[GH-ISSUE #2453] Add support for older AMD GPU gfx803, gfx802, gfx805 (e.g. Radeon RX 580, FirePro W7100) #1434

New Issue

@mnccouk commented on GitHub (Sep 15, 2024):

I have some success running Ollama using the 5.7.1 Rocm libraries with Radeon RX 580 GPU - Ubuntu 22.04.4 LTS, from within a docker container. Only tried so far with the llama3.1 model but seems to work ok.

The process I went through was just to focus and fix errors I encountered while trying to get something working, so this is by no means a proper fix for this issue.

The changes are based on the main Ollama branch from 6th Sept 2024, it would be interesting if somebody else could try to see if it works them too.

Repo with the changes - https://github.com/mnccouk/ollama/tree/rx580_gpu

See the top of the readme for some basic instructions to build. I only focused on the docker build, so make sure to install Rocm libraries on the docker host machine first (5.7.1).

@mnccouk commented on GitHub (Sep 15, 2024): I have some success running Ollama using the 5.7.1 Rocm libraries with Radeon RX 580 GPU - Ubuntu 22.04.4 LTS, from within a docker container. Only tried so far with the llama3.1 model but seems to work ok. The process I went through was just to focus and fix errors I encountered while trying to get something working, so this is by no means a proper fix for this issue. The changes are based on the main Ollama branch from 6th Sept 2024, it would be interesting if somebody else could try to see if it works them too. Repo with the changes - https://github.com/mnccouk/ollama/tree/rx580_gpu See the top of the readme for some basic instructions to build. I only focused on the docker build, so make sure to install Rocm libraries on the docker host machine first (5.7.1).

GiteaMirror commented

@jeromechungmf commented on GitHub (Sep 15, 2024):

Hello @Tamila-2017
I have been trying to run AI large language models (LLMs) on my PC using an RX580 GPU. I've tried two main approaches so far:
My Experience:

Linux with Vulkan:
I install Ubuntu 24.04 or 22.04, configuring it to use the Vulkan drivers, and then installing LMStudio. The goal is to make this setup work with my RX580 GPU for optimal LLM performance. However, attempting to install Vulkan drivers directly from AMD's website (for Ubuntu 22.04) proved unsuccessful. Instead, I found success by downloading the corresponding Vulkan driver package from https://vulkan.lunarg.com/sdk/home#linux

This allowed LMStudio to detect my RX580 GPU and use it for LLM calculations.

My Installation:
Currently, I am running Ubuntu Release 1.3.290 Ubuntu 22.04.
Recommendations
Thorough Driver Research: Before installing anything, make sure you download the correct Vulkan driver based on your Ubuntu version from https://vulkan.lunarg.com/sdk/home#linux

Windows with Vulkan:
In Windows OS Enviroment，just download AMD Driver Installer ， follow the GUI step by step，Finish the gpu driver install.The Vulkan driver for windows os have been ready .Finish the AMD Driver install，you can download LMStudio and Run it.you can see the LMStudio show GPU Infomation VRAM 8.0GB.

This is my setup for running AI language models using an RX580 graphics card. Please take a look and see if you can provide any guidance or suggestions.

Good luck.

Jerome

@jeromechungmf commented on GitHub (Sep 15, 2024): Hello @Tamila-2017 I have been trying to run AI large language models (LLMs) on my PC using an RX580 GPU. I've tried two main approaches so far: My Experience: Linux with Vulkan: I install Ubuntu 24.04 or 22.04, configuring it to use the Vulkan drivers, and then installing LMStudio. The goal is to make this setup work with my RX580 GPU for optimal LLM performance. However, attempting to install Vulkan drivers directly from AMD's website (for Ubuntu 22.04) proved unsuccessful. Instead, I found success by downloading the corresponding Vulkan driver package from https://vulkan.lunarg.com/sdk/home#linux This allowed LMStudio to detect my RX580 GPU and use it for LLM calculations. My Installation: Currently, I am running Ubuntu Release 1.3.290 Ubuntu 22.04. Recommendations Thorough Driver Research: Before installing anything, make sure you download the correct Vulkan driver based on your Ubuntu version from https://vulkan.lunarg.com/sdk/home#linux Windows with Vulkan: In Windows OS Enviroment，just download AMD Driver Installer ， follow the GUI step by step，Finish the gpu driver install.The Vulkan driver for windows os have been ready .Finish the AMD Driver install，you can download LMStudio and Run it.you can see the LMStudio show GPU Infomation VRAM 8.0GB. This is my setup for running AI language models using an RX580 graphics card. Please take a look and see if you can provide any guidance or suggestions. Good luck. Jerome

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 15, 2024):

Hi guys,

I am very grateful to all of you for your detailed account of how to solve my problem with the RX580.
Now I appeared a great chance to solve this problem and use RX580 to communicate with AI.

Can you please tell me, what is the speed increase of working with RX580 with AI compared to СPU N100?
Is it 10x, 100x, or maybe 1000x? :-)

@Tamila-2017 commented on GitHub (Sep 15, 2024): Hi guys, I am very grateful to all of you for your detailed account of how to solve my problem with the RX580. Now I appeared a great chance to solve this problem and use RX580 to communicate with AI. Can you please tell me, what is the speed increase of working with RX580 with AI compared to СPU N100? Is it 10x, 100x, or maybe 1000x? :-)

GiteaMirror commented

2026-04-12 11:18:55 -05:00

@KhazAkar commented on GitHub (Sep 15, 2024):

You can all try ollama-for-amd fork. I'm using it myself with my RX5500M card and ROCm 6.2. works wonders.

@KhazAkar commented on GitHub (Sep 15, 2024): You can all try ollama-for-amd fork. I'm using it myself with my RX5500M card and ROCm 6.2. works wonders.

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 15, 2024):

KhazAkar,

Thank you, this is also very interesting information. Could you tell us more about this?
For example, what exactly is the name of this fork, where can I get it and how to use it.
And most importantly: you are using the RX 5500M. Will this fork work with the RX580?

@Tamila-2017 commented on GitHub (Sep 15, 2024): KhazAkar, Thank you, this is also very interesting information. Could you tell us more about this? For example, what exactly is the name of this fork, where can I get it and how to use it. And most importantly: you are using the RX 5500M. Will this fork work with the RX580?

GiteaMirror commented

2026-04-12 11:18:55 -05:00

@Tamila-2017 commented on GitHub (Sep 16, 2024):

jeromechungmf

I chose your advice first:

Linux with Vulkan:
I install Ubuntu 24.04 or 22.04, configuring it to use the Vulkan drivers, and then installing LMStudio. The goal is to make this setup work with my RX580 GPU for optimal LLM performance. However, attempting to install Vulkan drivers directly from AMD's website (for Ubuntu 22.04) proved unsuccessful. Instead, I found success by downloading the corresponding Vulkan driver package from https://vulkan.lunarg.com/sdk/home#linux

This allowed LMStudio to detect my RX580 GPU and use it for LLM calculations.

And it ended in failure. What I did, in detail:

Install Ubuntu-24.04.1 LTS Server Minimal (without X's)
wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo tee /etc/apt/trusted.gpg.d/lunarg.asc sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-noble.list https://packages.lunarg.com/vulkan/lunarg-vulkan-noble.list sudo apt update sudo apt install vulkan-sdk
Download LM_Studio-0.3.2.AppImage to /home/ai/.appimage
Run LM Studio: needs FUSE
sudo add-apt-repository universe sudo apt install libfuse2t64
Run LM Studio: needs libnss3.so
sudo apt install libnss3
Run LM Studio: needs libasound
sudo apt install libasound2t64
Run LM Studio:

FATAL:setuid_sandbox_host.cc(158)] The SUID sandbox helper binary was found, but is not configured correctly.
Rather than run without sandboxing I'm aborting now.
You need to make sure that /tmp/.mount_LM_StusEylzO/chrome-sandbox is owned by root and has mode 4755.

But this file does not exist!

Next, I began to understand that a graphical environment is required to run LM Studio and I did not like it at all, because I want to using AI in a console environment, without X's.

That's why I stopped this experiment, and now I want to try KhazAkar advice.
But unfortunately he's not giving any details yet :-(

@Tamila-2017 commented on GitHub (Sep 16, 2024): **jeromechungmf** I chose your advice first: > Linux with Vulkan: > I install Ubuntu 24.04 or 22.04, configuring it to use the Vulkan drivers, and then installing LMStudio. The goal is to make this setup work with my RX580 GPU for optimal LLM performance. However, attempting to install Vulkan drivers directly from AMD's website (for Ubuntu 22.04) proved unsuccessful. Instead, I found success by downloading the corresponding Vulkan driver package from https://vulkan.lunarg.com/sdk/home#linux > > This allowed LMStudio to detect my RX580 GPU and use it for LLM calculations. And it ended in failure. What I did, in detail: 1. Install **Ubuntu-24.04.1 LTS Server Minimal** (without X's) 2. ` wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo tee /etc/apt/trusted.gpg.d/lunarg.asc sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-noble.list https://packages.lunarg.com/vulkan/lunarg-vulkan-noble.list sudo apt update sudo apt install vulkan-sdk` 3. Download **LM_Studio-0.3.2.AppImage** to /home/ai/.appimage 4. Run LM Studio: needs **FUSE** 6. `sudo add-apt-repository universe sudo apt install libfuse2t64 ` 7. Run LM Studio: needs **libnss3.so** 8. `sudo apt install libnss3` 9. Run LM Studio: needs **libasound** 10. `sudo apt install libasound2t64` 11. Run LM Studio: > FATAL:setuid_sandbox_host.cc(158)] The SUID sandbox helper binary was found, but is not configured correctly. > Rather than run without sandboxing I'm aborting now. > You need to make sure that /tmp/.mount_LM_StusEylzO/chrome-sandbox is owned by root and has mode 4755. But this file does not exist! Next, I began to understand that a graphical environment is required to run LM Studio and I did not like it at all, because I want to using AI in a console environment, without X's. That's why I stopped this experiment, and now I want to try **KhazAkar** advice. But unfortunately he's not giving any details yet :-(

GiteaMirror commented

2026-04-12 11:18:56 -05:00

@mnccouk commented on GitHub (Sep 16, 2024):

Hi guys,

I am very grateful to all of you for your detailed account of how to solve my problem with the RX580. Now I appeared a great chance to solve this problem and use RX580 to communicate with AI.

Can you please tell me, what is the speed increase of working with RX580 with AI compared to СPU N100? Is it 10x, 100x, or maybe 1000x? :-)

Here's my performance figures

CPU - AMD Ryzen 5 3400G with Radeon Vega Graphics (16GB RAM)

"eval_count":342,"eval_duration":66228944000 = 5.1639 tokens a second

GPU - Radeon Rx580 8GB vRAM

"eval_count":308,"eval_duration":11708779000} = 26.305 tokens a second

Same question asked in both cases:

curl http://192.168.0.156:11434/api/generate -d '{
  "model": "llama3.1",
  "prompt": "Why is the sky blue?"
}'

@mnccouk commented on GitHub (Sep 16, 2024): > Hi guys, > > I am very grateful to all of you for your detailed account of how to solve my problem with the RX580. Now I appeared a great chance to solve this problem and use RX580 to communicate with AI. > > Can you please tell me, what is the speed increase of working with RX580 with AI compared to СPU N100? Is it 10x, 100x, or maybe 1000x? :-) Here's my performance figures CPU - AMD Ryzen 5 3400G with Radeon Vega Graphics (16GB RAM) "eval_count":342,"eval_duration":66228944000 = 5.1639 tokens a second GPU - Radeon Rx580 8GB vRAM "eval_count":308,"eval_duration":11708779000} = 26.305 tokens a second Same question asked in both cases: ``` curl http://192.168.0.156:11434/api/generate -d '{ "model": "llama3.1", "prompt": "Why is the sky blue?" }' ```

GiteaMirror commented

2026-04-12 11:18:56 -05:00

@Tamila-2017 commented on GitHub (Sep 17, 2024):

mnccouk,
thank you so much for the convincing test! :-)
It turns out that the gain about speed is small, about 5 times.
At the same time, the TDP of:
AMD Ryzen 5 3400G = 65 Watts
Radeon Vega 16 = 75 Watts.
AMD RX580 = 185 Watts.

@Tamila-2017 commented on GitHub (Sep 17, 2024): **mnccouk**, thank you so much for the convincing test! :-) It turns out that the gain about speed is small, about 5 times. At the same time, the TDP of: AMD Ryzen 5 3400G = 65 Watts Radeon Vega 16 = 75 Watts. AMD RX580 = 185 Watts.

GiteaMirror commented

2026-04-12 11:18:56 -05:00

@jeromechungmf commented on GitHub (Sep 17, 2024):

jeromechungmf

I chose your advice first:

Linux with Vulkan:
I install Ubuntu 24.04 or 22.04, configuring it to use the Vulkan drivers, and then installing LMStudio. The goal is to make this setup work with my RX580 GPU for optimal LLM performance. However, attempting to install Vulkan drivers directly from AMD's website (for Ubuntu 22.04) proved unsuccessful. Instead, I found success by downloading the corresponding Vulkan driver package from https://vulkan.lunarg.com/sdk/home#linux
This allowed LMStudio to detect my RX580 GPU and use it for LLM calculations.

And it ended in failure. What I did, in detail:

Install Ubuntu-24.04.1 LTS Server Minimal (without X's)

wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo tee /etc/apt/trusted.gpg.d/lunarg.asc sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-noble.list https://packages.lunarg.com/vulkan/lunarg-vulkan-noble.list sudo apt update sudo apt install vulkan-sdk

Download LM_Studio-0.3.2.AppImage to /home/ai/.appimage

Run LM Studio: needs FUSE

sudo add-apt-repository universe sudo apt install libfuse2t64

Run LM Studio: needs libnss3.so

sudo apt install libnss3

Run LM Studio: needs libasound

sudo apt install libasound2t64

Run LM Studio:

FATAL:setuid_sandbox_host.cc(158)] The SUID sandbox helper binary was found, but is not configured correctly.
Rather than run without sandboxing I'm aborting now.
You need to make sure that /tmp/.mount_LM_StusEylzO/chrome-sandbox is owned by root and has mode 4755.

But this file does not exist!

Next, I began to understand that a graphical environment is required to run LM Studio and I did not like it at all, because I want to using AI in a console environment, without X's.

That's why I stopped this experiment, and now I want to try KhazAkar advice. But unfortunately he's not giving any details yet :-(

Hi Tamila-2017

Sorry，I should show my install process step by step:

1.downlaod the Ubuntu 22.04 Desktop iso file,my file name:ubuntu-22.04.4-desktop-amd64.iso .
2.make the iso file to usb boot Disk.In the linux you can use the dd command make it.ex:sudo dd bs=4M if=/path/to/ubuntu-22.04.4-desktop-amd64.iso of=/dev/sdX status=progress oflag=sync
3.I booted from a USB drive and installed Ubuntu 22.04 on my computer.You can choice minimal option，follw the gui guide step by step.
4.open the firefox paste the https://vulkan.lunarg.com/sdk/home#linux, you can browse the vulkan page，in the Linux content,click the Ubuntu Packages tab,then roll down until see the Release 1.3.290 link,click it.you can see the Ubuntu 22.04 (Jammy Jellyfish) and Ubuntu 24.04 (Noble Numbat).in my case,i click the Ubuntu 22.04 (Jammy Jellyfish) item.it's show the command like below:
wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo tee /etc/apt/trusted.gpg.d/lunarg.asc
sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-1.3.290-jammy.list https://packages.lunarg.com/vulkan/1.3.290/lunarg-vulkan-1.3.290-jammy.list
sudo apt update
sudo apt install vulkan-sdk
5.open the terminal follow the command and execute.finish to install vulkan driver in my ubuntu 22.04 desktop enviroment.
6.reboot
7.download the lmstudio and follow the https://www.tecmint.com/lm-studio-run-llms-linux/ to run lmstudio.

In my experince,it's should be work.

PS:use xdrp to longin Ubuntu should cause by privilege issue of lmstudio,if you want to remote login ubuntu to run lmstudio.Using like RealVNC Viewer for remote access is a better option, as it can avoid permission problems and allow the lmstudio can detection of the RX580 GPU.

Good luck.

Jerome

@jeromechungmf commented on GitHub (Sep 17, 2024): > **jeromechungmf** > > I chose your advice first: > > > Linux with Vulkan: > > I install Ubuntu 24.04 or 22.04, configuring it to use the Vulkan drivers, and then installing LMStudio. The goal is to make this setup work with my RX580 GPU for optimal LLM performance. However, attempting to install Vulkan drivers directly from AMD's website (for Ubuntu 22.04) proved unsuccessful. Instead, I found success by downloading the corresponding Vulkan driver package from https://vulkan.lunarg.com/sdk/home#linux > > This allowed LMStudio to detect my RX580 GPU and use it for LLM calculations. > > And it ended in failure. What I did, in detail: > > 1. Install **Ubuntu-24.04.1 LTS Server Minimal** (without X's) > 2. ` wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo tee /etc/apt/trusted.gpg.d/lunarg.asc sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-noble.list https://packages.lunarg.com/vulkan/lunarg-vulkan-noble.list sudo apt update sudo apt install vulkan-sdk` > 3. Download **LM_Studio-0.3.2.AppImage** to /home/ai/.appimage > 4. Run LM Studio: needs **FUSE** > 5. `sudo add-apt-repository universe sudo apt install libfuse2t64 ` > 6. Run LM Studio: needs **libnss3.so** > 7. `sudo apt install libnss3` > 8. Run LM Studio: needs **libasound** > 9. `sudo apt install libasound2t64` > 10. Run LM Studio: > > > FATAL:setuid_sandbox_host.cc(158)] The SUID sandbox helper binary was found, but is not configured correctly. > > Rather than run without sandboxing I'm aborting now. > > You need to make sure that /tmp/.mount_LM_StusEylzO/chrome-sandbox is owned by root and has mode 4755. > > But this file does not exist! > > Next, I began to understand that a graphical environment is required to run LM Studio and I did not like it at all, because I want to using AI in a console environment, without X's. > > That's why I stopped this experiment, and now I want to try **KhazAkar** advice. But unfortunately he's not giving any details yet :-( Hi Tamila-2017 Sorry，I should show my install process step by step: 1.downlaod the Ubuntu 22.04 Desktop iso file,my file name:ubuntu-22.04.4-desktop-amd64.iso . 2.make the iso file to usb boot Disk.In the linux you can use the dd command make it.ex:sudo dd bs=4M if=/path/to/ubuntu-22.04.4-desktop-amd64.iso of=/dev/sdX status=progress oflag=sync 3.I booted from a USB drive and installed Ubuntu 22.04 on my computer.You can choice minimal option，follw the gui guide step by step. 4.open the firefox paste the https://vulkan.lunarg.com/sdk/home#linux, you can browse the vulkan page，in the Linux content,click the Ubuntu Packages tab,then roll down until see the Release 1.3.290 link,click it.you can see the Ubuntu 22.04 (Jammy Jellyfish) and Ubuntu 24.04 (Noble Numbat).in my case,i click the Ubuntu 22.04 (Jammy Jellyfish) item.it's show the command like below: wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo tee /etc/apt/trusted.gpg.d/lunarg.asc sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-1.3.290-jammy.list https://packages.lunarg.com/vulkan/1.3.290/lunarg-vulkan-1.3.290-jammy.list sudo apt update sudo apt install vulkan-sdk 5.open the terminal follow the command and execute.finish to install vulkan driver in my ubuntu 22.04 desktop enviroment. 6.reboot 7.download the lmstudio and follow the https://www.tecmint.com/lm-studio-run-llms-linux/ to run lmstudio. In my experince,it's should be work. PS:use xdrp to longin Ubuntu should cause by privilege issue of lmstudio,if you want to remote login ubuntu to run lmstudio.Using like RealVNC Viewer for remote access is a better option, as it can avoid permission problems and allow the lmstudio can detection of the RX580 GPU. Good luck. Jerome

GiteaMirror commented

2026-04-12 11:18:57 -05:00

@Tamila-2017 commented on GitHub (Sep 17, 2024):

jeromechungmf,
Thank you for your attention and detailed description of the installation.
And I'm sorry that I still have a little experience in Linux.
So, I'm starting to install the recommended one for me ubuntu-22.04.4-desktop-amd64

@Tamila-2017 commented on GitHub (Sep 17, 2024): **jeromechungmf**, Thank you for your attention and detailed description of the installation. And I'm sorry that I still have a little experience in Linux. So, I'm starting to install the recommended one for me ubuntu-22.04.4-desktop-amd64

GiteaMirror commented

2026-04-12 11:18:57 -05:00

@KhazAkar commented on GitHub (Sep 17, 2024):

For vulkan, you don't need to downgrade to 22.04. 24.04 is also fine, even better. Ollama does not yet have vulkan support, but will gain it. For vulkan in CLI, you can either build Ollama with vulkan support from PR in this repo or build llama.cpp from ggreganov repository with vulkan support enabled. If you need GUI, then GPT4All is best one for GUI apps available, eventually LM Studio, but since I prefer open source, I'd pick GPT4ALL. In general I think we should move such discussion somewhere else and here focus on trying to get Ollama running with older AMD GPUs. So my proposition to try vulkan PR in Ollama marches the best.

@KhazAkar commented on GitHub (Sep 17, 2024): For vulkan, you don't need to downgrade to 22.04. 24.04 is also fine, even better. Ollama does not yet have vulkan support, but will gain it. For vulkan in CLI, you can either build Ollama with vulkan support from PR in this repo or build llama.cpp from ggreganov repository with vulkan support enabled. If you need GUI, then GPT4All is best one for GUI apps available, eventually LM Studio, but since I prefer open source, I'd pick GPT4ALL. In general I think we should move such discussion somewhere else and here focus on trying to get Ollama running with older AMD GPUs. So my proposition to try vulkan PR in Ollama marches the best.

GiteaMirror commented

2026-04-12 11:18:57 -05:00

@Tamila-2017 commented on GitHub (Sep 17, 2024):

So, I successfully installed Ubuntu-22.04.4-desktop-amd64 and followed steps "1-4".
However, unfortunately, action "5" is not executed, because there is no such command: execute.finish

.............
KhazAkar, oh, I'm glad you're here :-)
However, your message is vague and therefore not entirely clear
Could you elaborate on your recommendations, step by step?

@Tamila-2017 commented on GitHub (Sep 17, 2024): So, I successfully installed Ubuntu-22.04.4-desktop-amd64 and followed steps "1-4". However, unfortunately, action "5" is not executed, because there is no such command: **execute.finish** ............. KhazAkar, oh, I'm glad you're here :-) However, your message is vague and therefore not entirely clear Could you elaborate on your recommendations, step by step?

GiteaMirror commented

2026-04-12 11:18:58 -05:00

@mnccouk commented on GitHub (Sep 17, 2024):

@Tamila-2017

I happen to have a real time power meter on the server, The results for power consumption:-

CPU while processing - "Why is the sky blue?"
= Duration = 66.7 seconds with power consumption = 104 Watts
= 0.001926889 kWh
= 1.92 Wh

GPU while processing - "Why is the sky blue?"
Duration = 11.7 seconds with power consumption = 230 Watts
= 0.001661111 kWh
= 1.66 Wh

Server power while GPU\CPU idle is 40 Watts

@mnccouk commented on GitHub (Sep 17, 2024): @Tamila-2017 I happen to have a real time power meter on the server, The results for power consumption:- CPU while processing - "Why is the sky blue?" = Duration = 66.7 seconds with power consumption = 104 Watts = 0.001926889 kWh = 1.92 Wh GPU while processing - "Why is the sky blue?" Duration = 11.7 seconds with power consumption = 230 Watts = 0.001661111 kWh = 1.66 Wh Server power while GPU\CPU idle is 40 Watts

GiteaMirror commented

2026-04-12 11:18:58 -05:00

@Tamila-2017 commented on GitHub (Sep 17, 2024):

mnccouk,
Measurements are very useful, thank you

@Tamila-2017 commented on GitHub (Sep 17, 2024): **mnccouk**, Measurements are very useful, thank you

GiteaMirror commented

2026-04-12 11:18:59 -05:00

@Tamila-2017 commented on GitHub (Sep 17, 2024):

jeromechungmf,

I don't understand what the execute command execute.finish, which does not exist.

Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/
and I managed to launch the LM Studio.

Unfortunately, LM Studio also doesn't see the RX580 :-(
Although Ubuntu and nvtop sees it perfectly.

@Tamila-2017 commented on GitHub (Sep 17, 2024): **jeromechungmf**, I don't understand what the execute command **execute.finish**, which does not exist. Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio. Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly.

GiteaMirror commented

2026-04-12 11:18:59 -05:00

@Tamila-2017 commented on GitHub (Sep 17, 2024):

KhazAkar

If you need GUI, then ...

I don't need the GUI, I'm strongly against it because I need to work with Ollama in the console.

In general I think we should move such discussion somewhere else and here focus on trying to get Ollama running with older AMD GPUs.

I completely agree with you! But for some reason you're stubbornly not sharing the details of this success of yours -

You can all try ollama-for-amd fork. I'm using it myself with my RX5500M card and ROCm 6.2. works wonders.

even though I've asked you about it before.
Why don't you tell me about it? Is it your big secret or it your was just a bad joke?

Unlike you jeromechungmf detailed his experience and I tried to use it, although I'm not thrilled with the graphical environment because I need a console server option that works with Ollama.

@Tamila-2017 commented on GitHub (Sep 17, 2024): **KhazAkar** > If you need GUI, then ... I don't need the GUI, I'm strongly against it because I need to work with Ollama in the console. > In general I think we should move such discussion somewhere else and here focus on trying to get Ollama running with older AMD GPUs. I completely agree with you! But for some reason you're stubbornly not sharing the details of this success of yours - > You can all try ollama-for-amd fork. I'm using it myself with my RX5500M card and ROCm 6.2. works wonders. even though I've asked you about it before. Why don't you tell me about it? Is it your big secret or it your was just a bad joke? Unlike you **jeromechungmf** detailed his experience and I tried to use it, although I'm not thrilled with the graphical environment because I need a console server option that works with Ollama.

GiteaMirror commented

2026-04-12 11:19:00 -05:00

@KhazAkar commented on GitHub (Sep 17, 2024):

@Tamila-2017 it's not a joke. Fork is here: https://github.com/likelovewant/ollama-for-amd
But I can't assure you of anything with older card than I currently have. This one does not use vulkan but ROCm and I don't know if your card is supported by ROCm 6.x.
For Vulkan usage, if you're feeling brave, you can try building Ollama with vulkan support by cloning repository and applying this PR:
https://github.com/ollama/ollama/pull/5059

In both cases, building Ollama by hand is inevitable.

@KhazAkar commented on GitHub (Sep 17, 2024): @Tamila-2017 it's not a joke. Fork is here: https://github.com/likelovewant/ollama-for-amd But I can't assure you of anything with older card than I currently have. This one does not use vulkan but ROCm and I don't know if your card is supported by ROCm 6.x. For Vulkan usage, if you're feeling brave, you can try building Ollama with vulkan support by cloning repository and applying this PR: https://github.com/ollama/ollama/pull/5059 In both cases, building Ollama by hand is inevitable.

GiteaMirror commented

2026-04-12 11:19:01 -05:00

@KhazAkar commented on GitHub (Sep 17, 2024):

Plus, @Tamila-2017 - only Info in this fork I've found is in foreign language to me.
https://github.com/likelovewant/ollama-for-amd/issues/1
You can translate it if you want.
And it would be great idea that you join Ollama discord server :)

@KhazAkar commented on GitHub (Sep 17, 2024): Plus, @Tamila-2017 - only Info in this fork I've found is in foreign language to me. https://github.com/likelovewant/ollama-for-amd/issues/1 You can translate it if you want. And it would be great idea that you join Ollama discord server :)

GiteaMirror commented

2026-04-12 11:19:01 -05:00

@jeromechungmf commented on GitHub (Sep 18, 2024):

jeromechungmf,

I don't understand what the execute command execute.finish, which does not exist.

Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio.

Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly.

Hi Tamila-2017

Attach my screenshot,it's contain execute the vulkan driver install commands,my be you can refer.

Good lucks.

Jerome

@jeromechungmf commented on GitHub (Sep 18, 2024): > **jeromechungmf**, > > I don't understand what the execute command **execute.finish**, which does not exist. > > Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio. > > Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly. Hi Tamila-2017 ![1726621208963](https://github.com/user-attachments/assets/e2bf8df2-bdfe-4cd0-91ae-962bf0fbbe78) ![1726621296912](https://github.com/user-attachments/assets/9128e80c-8f5d-4b38-9c4c-a62e97bf20d5) ![1726621331628](https://github.com/user-attachments/assets/2ec0db74-7ca0-4fa2-9682-5b94a10d15ba) ![1726621374395](https://github.com/user-attachments/assets/5598d839-556b-46bb-bcb8-cf9b7ea72906) Attach my screenshot,it's contain execute the vulkan driver install commands,my be you can refer. Good lucks. Jerome

GiteaMirror commented

2026-04-12 11:19:02 -05:00

@jeromechungmf commented on GitHub (Sep 18, 2024):

jeromechungmf,

I don't understand what the execute command execute.finish, which does not exist.

Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio.

Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly.

Hi Tamila-2017

Attach the LMStudio Configuration sccreenshot,in the developer tab,click the LM Runtimes,you can check llama.cpp Vulkan should be v0.0.4.Then Click Settings tab you can see the VRAM 8 GB.

Jerome

@jeromechungmf commented on GitHub (Sep 18, 2024): > **jeromechungmf**, > > I don't understand what the execute command **execute.finish**, which does not exist. > > Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio. > > Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly. ![DSC_0532](https://github.com/user-attachments/assets/3399e75c-7c37-4726-91ee-738d460fdac8) ![DSC_0533](https://github.com/user-attachments/assets/535820c8-878d-4dbb-a282-f2d0987ac419) Hi Tamila-2017 Attach the LMStudio Configuration sccreenshot,in the developer tab,click the LM Runtimes,you can check llama.cpp Vulkan should be v0.0.4.Then Click Settings tab you can see the VRAM 8 GB. Jerome

GiteaMirror commented

2026-04-12 11:19:03 -05:00

@bellmancity commented on GitHub (Sep 19, 2024):

jeromechungmf,
I don't understand what the execute command execute.finish, which does not exist.
Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio.
Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly.

Hi Tamila-2017

Attach the LMStudio Configuration sccreenshot,in the developer tab,click the LM Runtimes,you can check llama.cpp Vulkan should be v0.0.4.Then Click Settings tab you can see the VRAM 8 GB.

Jerome

I tried LM Studio using your method (vulkan + lmstudio appimage) on my Linux Mint 22 with RX580. It just works right away. My RX580 is "refurbished" 16Gb from Aliexpress.

10.81 tok/sec 531 tokens 0.91s to first token

Hopefully Ollama will works with RX580 soon.

@bellmancity commented on GitHub (Sep 19, 2024): > > **jeromechungmf**, > > I don't understand what the execute command **execute.finish**, which does not exist. > > Therefore, I used this instruction - https://www.tecmint.com/lm-studio-run-llms-linux/ and I managed to launch the LM Studio. > > Unfortunately, LM Studio also doesn't see the RX580 :-( Although Ubuntu and nvtop sees it perfectly. > > ![DSC_0532](https://private-user-images.githubusercontent.com/4884739/368399186-3399e75c-7c37-4726-91ee-738d460fdac8.JPG?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.WsAGZompR7AsVtwcvQFxrIFUzVuWPCG887iyc8ZijmY) ![DSC_0533](https://private-user-images.githubusercontent.com/4884739/368399207-535820c8-878d-4dbb-a282-f2d0987ac419.JPG?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.fn0GUlyi5wueNrSzY4naRd3txE2yQOs3k_sZmkGG_Dk) > > Hi Tamila-2017 > > Attach the LMStudio Configuration sccreenshot,in the developer tab,click the LM Runtimes,you can check llama.cpp Vulkan should be v0.0.4.Then Click Settings tab you can see the VRAM 8 GB. > > Jerome I tried LM Studio using your method (vulkan + lmstudio appimage) on my Linux Mint 22 with RX580. It just works right away. My RX580 is "refurbished" 16Gb from Aliexpress. ![Screenshot from 2024-09-19 12-25-05](https://github.com/user-attachments/assets/c9a785d9-6456-444b-93f4-6fdb53ed6954) 10.81 tok/sec 531 tokens 0.91s to first token Hopefully Ollama will works with RX580 soon.

GiteaMirror commented

2026-04-12 11:19:03 -05:00

@Tamila-2017 commented on GitHub (Sep 19, 2024):

Hello all,

Thanks all for the tips, but unfortunately, I can't handle these software tricks on my own.
If you want to help me remotely, I will give you access to my an AI computer, and you can configure the RX580's personally.

@Tamila-2017 commented on GitHub (Sep 19, 2024): Hello all, Thanks all for the tips, but unfortunately, I can't handle these software tricks on my own. If you want to help me remotely, I will give you access to my an AI computer, and you can configure the RX580's personally.

GiteaMirror commented

2026-04-12 11:19:04 -05:00

@Tamila-2017 commented on GitHub (Sep 19, 2024):

KhazAkar

And it would be great idea that you join Ollama discord server :)

I tried to register on Discord - https://discord.com
But it first demanded my phone number and then told me that the this number was incorrect.
Discord stupid, he doesn't know about phone numbers.

@Tamila-2017 commented on GitHub (Sep 19, 2024): **KhazAkar** > And it would be great idea that you join Ollama discord server :) I tried to register on Discord - https://discord.com But it first demanded my phone number and then told me that the this number was incorrect. Discord stupid, he doesn't know about phone numbers.

GiteaMirror commented

2026-04-12 11:19:04 -05:00

@mnccouk commented on GitHub (Sep 19, 2024):

I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone.
https://hub.docker.com/r/mnccouk/ollama-gpu-rx580

@mnccouk commented on GitHub (Sep 19, 2024): I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone. https://hub.docker.com/r/mnccouk/ollama-gpu-rx580

GiteaMirror commented

2026-04-12 11:19:05 -05:00

@bellmancity commented on GitHub (Sep 21, 2024):

I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone. https://hub.docker.com/r/mnccouk/ollama-gpu-rx580

I have tried your docker image and it runs successfully.
Eval rate = 15.25 tokes/s. Very impressive compared to LM Studio above.

RX580 2048 with 16GB VRAM
Ryzen 5 4650G with 16GB Memory.

Thank you for your effort.

@bellmancity commented on GitHub (Sep 21, 2024): > I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone. https://hub.docker.com/r/mnccouk/ollama-gpu-rx580 ![amd_gpu](https://github.com/user-attachments/assets/5a43f495-13f7-4d59-a818-288aa243f8c4) ![result_ollama](https://github.com/user-attachments/assets/6342b2e4-59e5-4689-a62f-b61422c8a047) I have tried your docker image and it runs successfully. Eval rate = 15.25 tokes/s. Very impressive compared to LM Studio above. RX580 2048 with 16GB VRAM Ryzen 5 4650G with 16GB Memory. Thank you for your effort.

GiteaMirror commented

2026-04-12 11:19:06 -05:00

@mnccouk commented on GitHub (Sep 21, 2024):

@bellmancity, Awesome! Glad it worked for you.

@mnccouk commented on GitHub (Sep 21, 2024): @bellmancity, Awesome! Glad it worked for you.

GiteaMirror commented

2026-04-12 11:19:06 -05:00

@Tamila-2017 commented on GitHub (Sep 21, 2024):

I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone.
https://hub.docker.com/r/mnccouk/ollama-gpu-rx580

I can't install ROCm 5.7.0 because there are obscure errors when compiling.

mnccouk,

please give the exact download link of the distro you used here

@Tamila-2017 commented on GitHub (Sep 21, 2024): > I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone. > https://hub.docker.com/r/mnccouk/ollama-gpu-rx580 I can't install ROCm 5.7.0 because there are obscure errors when compiling. **mnccouk**, please give the exact download link of the distro you used [here](https://github.com/ollama/ollama/issues/2453#issuecomment-2362217923)

GiteaMirror commented

2026-04-12 11:19:06 -05:00

@eliot-akira commented on GitHub (Sep 21, 2024):

Maybe this link will help:

Follow this documentation for rocm installation, just substitute the 5.7.0 references to 5.7.1 in the documentation. --https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html

From https://github.com/mnccouk/ollama/tree/rx580_gpu?tab=readme-ov-file#linux-with-rx580-radeon-gpu

@eliot-akira commented on GitHub (Sep 21, 2024): Maybe this link will help: > Follow this documentation for rocm installation, just substitute the 5.7.0 references to 5.7.1 in the documentation. --https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html From https://github.com/mnccouk/ollama/tree/rx580_gpu?tab=readme-ov-file#linux-with-rx580-radeon-gpu

GiteaMirror commented

2026-04-12 11:19:07 -05:00

@Tamila-2017 commented on GitHub (Sep 21, 2024):

eliot-akira,

Thanks, these links have been known to me for a long time.
But I am not aware of the exact distribution that uses mnccouk.
I'm waiting for the exact link from mnccouk to download the distro, because a lot depends on the distro.

@Tamila-2017 commented on GitHub (Sep 21, 2024): **eliot-akira**, Thanks, these links have been known to me for a long time. But I am not aware of the exact distribution that uses mnccouk. I'm waiting for the exact link from **mnccouk** to download the distro, because a lot depends on the distro.

GiteaMirror commented

2026-04-12 11:19:07 -05:00

@KhazAkar commented on GitHub (Sep 21, 2024):

Eventually just make use of Vulkan driver, either in ollama from PR in this repo, or llama.cpp from https://github.com/ggerganov/llama.cpp built from source....

@KhazAkar commented on GitHub (Sep 21, 2024): Eventually just make use of Vulkan driver, either in ollama from PR in this repo, or llama.cpp from https://github.com/ggerganov/llama.cpp built from source....

GiteaMirror commented

2026-04-12 11:19:08 -05:00

@Tamila-2017 commented on GitHub (Sep 21, 2024):

I've tried using Vulkan before, but nothing good came out of it.

So I want to use the docker that mnccouk created and it worked for him and also at bellmancity

@Tamila-2017 commented on GitHub (Sep 21, 2024): I've tried using Vulkan before, but nothing good came out of it. So I want to use the docker that **mnccouk** created and it worked for him and also at **bellmancity**

GiteaMirror commented

2026-04-12 11:19:08 -05:00

@KhazAkar commented on GitHub (Sep 21, 2024):

I've tried using Vulkan before, but nothing good came out of it.

So I want to use the docker that mnccouk created and it worked for him and also at bellmancity

Something feels off for me here. I'm not talking about gui apps like somebody here tried to give LMStudio here
Plus, I think fastest way would be to use vulkan based inference, even building ollama from source, which is well written in docs here.
ROCm 5.7.x is quite... old. 5.7.1 is shipped with ubuntu 24.04 and debian 12 I think, under slightly different name, search for hip. Current versions of Ollama work with 6.0.x at least and I don't know if they work right with 5.7.x series. Screenshot on top says otherwise, but currently RX 580, besides pletinful of VRAM, ended it's life after cryptomining boom.
Best bet for its usage for AI is to push vulkan implementation for Ollama to be merged, available as PR here:
https://github.com/ollama/ollama/pull/5059

Either way, signing out. This thread got quite derailed IMHO and 3/4 of it should be on official discord server.

@KhazAkar commented on GitHub (Sep 21, 2024): > I've tried using Vulkan before, but nothing good came out of it. > > So I want to use the docker that **mnccouk** created and it worked for him and also at **bellmancity** Something feels off for me here. I'm not talking about gui apps like somebody here tried to give LMStudio here Plus, I think fastest way would be to use vulkan based inference, even building ollama from source, which is well written in docs here. ROCm 5.7.x is quite... old. 5.7.1 is shipped with ubuntu 24.04 and debian 12 I think, under slightly different name, search for `hip`. Current versions of Ollama work with 6.0.x at least and I don't know if they work right with 5.7.x series. Screenshot on top says otherwise, but currently RX 580, besides pletinful of VRAM, ended it's life after cryptomining boom. Best bet for its usage for AI is to push vulkan implementation for Ollama to be merged, available as PR here: https://github.com/ollama/ollama/pull/5059 Either way, signing out. This thread got quite derailed IMHO and 3/4 of it should be on official discord server.

GiteaMirror commented

2026-04-12 11:19:08 -05:00

@Tamila-2017 commented on GitHub (Sep 21, 2024):

KhazAkar ,
You narrate very much and interestingly.
Unfortunately, your thoughts do not lead to the final result, because they are based only on probabilistic guesses, but without their practical verification.

The only one who has a real success here without using the DE graphical environment and Vulkan - is modest mnccouk.

I'm waiting for him to give me the exact link to download the distro he uses.
Once again: not the name of the distro, but the EXACT LINK to download.

@Tamila-2017 commented on GitHub (Sep 21, 2024): **KhazAkar** , You narrate very much and interestingly. Unfortunately, your thoughts do not lead to the final result, because they are based only on probabilistic guesses, but without their practical verification. The only one who has a real success here without using the DE graphical environment and Vulkan - is modest **mnccouk**. I'm waiting for him to give me the exact link to download the distro he uses. Once again: not the name of the distro, but the EXACT LINK to download.

GiteaMirror commented

2026-04-12 11:19:09 -05:00

@mnccouk commented on GitHub (Sep 21, 2024):

@Tamila-2017, I installed Ubuntu 20.04 then did a distro upgrade to Ubuntu 22.04.4. Only reason I didn't install 22.04 directly was due to usb boot issue using 22.04 installation media, but that's another story.
Direct link (https://releases.ubuntu.com/22.04.4/) to 22.04.4 doesn't work for some reason. However, there is a link for 22.04.5, https://releases.ubuntu.com/jammy/ubuntu-22.04.5-desktop-amd64.iso.

Then install the Rocm libraries @eliot-akira provided the link to (above).

Sorry can't provide 22.04.4 working direct link but can't see why 22.04.5 shouldn't work for you.

Obviously, also make sure to install docker, this can be done using the packages that come with Ubuntu 22.04.x

@mnccouk commented on GitHub (Sep 21, 2024): @Tamila-2017, I installed Ubuntu 20.04 then did a distro upgrade to Ubuntu 22.04.4. Only reason I didn't install 22.04 directly was due to usb boot issue using 22.04 installation media, but that's another story. Direct link (https://releases.ubuntu.com/22.04.4/) to 22.04.4 doesn't work for some reason. However, there is a link for 22.04.5, https://releases.ubuntu.com/jammy/ubuntu-22.04.5-desktop-amd64.iso. Then install the Rocm libraries @eliot-akira provided the link to (above). Sorry can't provide 22.04.4 working direct link but can't see why 22.04.5 shouldn't work for you. Obviously, also make sure to install docker, this can be done using the packages that come with Ubuntu 22.04.x

GiteaMirror commented

2026-04-12 11:19:10 -05:00

@SirDubbins commented on GitHub (Sep 22, 2024):

I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone. https://hub.docker.com/r/mnccouk/ollama-gpu-rx580

This worked for me using an RX480

@SirDubbins commented on GitHub (Sep 22, 2024): > I've added the docker image I'm using with the rx580 with Ollama to docker hub, Hopefully it might prove to be useful to someone. https://hub.docker.com/r/mnccouk/ollama-gpu-rx580 This worked for me using an RX480

GiteaMirror commented

2026-04-12 11:19:10 -05:00

@Tamila-2017 commented on GitHub (Sep 22, 2024):

Dear mnccouk,

Here is where you can download any version of Ubuntu 22.04.04 -

https://old-releases.ubuntu.com/releases/jammy/

But there are 2 options here:

ubuntu-22.04.4-desktop-amd64.iso (Xorg)
ubuntu-22.04.4-live-server-amd64.iso (console)

Can I use the version without Xorg and Gnom?

I.e. ubuntu-22.04.4-live-server-amd64.iso ?

@Tamila-2017 commented on GitHub (Sep 22, 2024): Dear **mnccouk**, Here is where you can download any version of Ubuntu 22.04.04 - https://old-releases.ubuntu.com/releases/jammy/ But there are 2 options here: - ubuntu-22.04.4-desktop-amd64.iso (Xorg) - ubuntu-22.04.4-live-server-amd64.iso (console) Can I use the version without Xorg and Gnom? I.e. **ubuntu-22.04.4-live-server-amd64.iso** ?

GiteaMirror commented

2026-04-12 11:19:11 -05:00

@mnccouk commented on GitHub (Sep 22, 2024):

Ah nice, glad you found the link.
I installed the GUI version on my server as I have a monitor connected to it and wanted to use GUI tools to do stuff. If you have a monitor plugged into your server maybe this is a good option to go with in the beginning.

I'm sure ubuntu-22.04.4-live-server-amd64.iso should also work fine but can't validate that personally.

Maybe get it working first with the desktop version, then once you know all the steps switch to the server version, of cause the choice is yours.

@mnccouk commented on GitHub (Sep 22, 2024): Ah nice, glad you found the link. I installed the GUI version on my server as I have a monitor connected to it and wanted to use GUI tools to do stuff. If you have a monitor plugged into your server maybe this is a good option to go with in the beginning. I'm sure ubuntu-22.04.4-live-server-amd64.iso should also work fine but can't validate that personally. Maybe get it working first with the desktop version, then once you know all the steps switch to the server version, of cause the choice is yours.

GiteaMirror commented

2026-04-12 11:19:11 -05:00

@SirDubbins commented on GitHub (Sep 22, 2024):

Has anyone been successful using multiple GPUs? I was trying to add a couple 8gb RX470 but I can't seem to get them to show up in rocminfo.

@SirDubbins commented on GitHub (Sep 22, 2024): Has anyone been successful using multiple GPUs? I was trying to add a couple 8gb RX470 but I can't seem to get them to show up in rocminfo.

GiteaMirror commented

2026-04-12 11:19:11 -05:00

@Tamila-2017 commented on GitHub (Sep 22, 2024):

This nightmarishly unfinished manual -
https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html
this is not an instruction, but a mockery of common sense.

This is the third time I have installed Ubuntu-22.04.04.04-Server "Jammy", then followed these instructions, remembering to replace 5.7.0 with 5.7.1, but fail every time:

Only the CPU is visible, but the graphics card is not.

I don't understand why this is the case :-(

Dear mnccouk,

could you do me the courtesy of creating a linear bash script instead of this nightmarish instruction, which I could execute without having to think about unnecessary complications to get a working ROCm 5.7.1 ?

@Tamila-2017 commented on GitHub (Sep 22, 2024): This nightmarishly unfinished manual - https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html this is not an instruction, but a mockery of common sense. This is the third time I have installed Ubuntu-22.04.04.04-Server "Jammy", then followed these instructions, remembering to replace 5.7.0 with 5.7.1, but fail every time: - Only the CPU is visible, but the graphics card is not. I don't understand why this is the case :-( Dear **mnccouk**, could you do me the courtesy of creating a linear bash script instead of this nightmarish instruction, which I could execute without having to think about unnecessary complications to get a working ROCm 5.7.1 ?

GiteaMirror commented

2026-04-12 11:19:12 -05:00

@mnccouk commented on GitHub (Sep 22, 2024):

Just been through the docs, and back through my notes. This should get you close if not up and running. Haven't tested but hopefully this should get you on the right track

sudo mkdir --parents --mode=0755 /etc/apt/keyrings

wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | \
    gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
    
echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/amdgpu/5.7.1/ubuntu jammy main" \
    | sudo tee /etc/apt/sources.list.d/amdgpu.list
    
sudo apt update

echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600

sudo apt install amdgpu-dkms

sudo reboot

for ver in 5.7.1; do
echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/$ver jammy main" \
    | sudo tee --append /etc/apt/sources.list.d/rocm.list
done

echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' \
    | sudo tee /etc/apt/preferences.d/rocm-pin-600
    
sudo apt update

sudo apt install rocm-hip-sdk #if that didn't work try -> sudo apt install rocm-hip-sdk5.7.1

#Think these may already be installed through deps, but no harm in trying 
sudo apt install rocm-utils #if that didn't work again try-> sudo apt install rocm-utils5.7.1

sudo tee --append /etc/ld.so.conf.d/rocm.conf <<EOF
/opt/rocm/lib
/opt/rocm/lib64
EOF

sudo ldconfig

#You'll need to add this in you user profile, this will work if you stay in the same terminal
export PATH=$PATH:/opt/rocm-5.7.1/bin:/opt/rocm-5.7.1/opencl/bin

sudo usermod -a -G render your_linux_username_here

#I still had permissions issue so brute forced access
sudo chmod 666 /dev/kfd

#to varify the libraries\driver are\is working
/opt/rocm/bin/rocminfo

@mnccouk commented on GitHub (Sep 22, 2024): Just been through the docs, and back through my notes. This should get you close if not up and running. Haven't tested but hopefully this should get you on the right track ``` sudo mkdir --parents --mode=0755 /etc/apt/keyrings wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | \ gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/amdgpu/5.7.1/ubuntu jammy main" \ | sudo tee /etc/apt/sources.list.d/amdgpu.list sudo apt update echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600 sudo apt install amdgpu-dkms sudo reboot for ver in 5.7.1; do echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/$ver jammy main" \ | sudo tee --append /etc/apt/sources.list.d/rocm.list done echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' \ | sudo tee /etc/apt/preferences.d/rocm-pin-600 sudo apt update sudo apt install rocm-hip-sdk #if that didn't work try -> sudo apt install rocm-hip-sdk5.7.1 #Think these may already be installed through deps, but no harm in trying sudo apt install rocm-utils #if that didn't work again try-> sudo apt install rocm-utils5.7.1 sudo tee --append /etc/ld.so.conf.d/rocm.conf <<EOF /opt/rocm/lib /opt/rocm/lib64 EOF sudo ldconfig #You'll need to add this in you user profile, this will work if you stay in the same terminal export PATH=$PATH:/opt/rocm-5.7.1/bin:/opt/rocm-5.7.1/opencl/bin sudo usermod -a -G render your_linux_username_here #I still had permissions issue so brute forced access sudo chmod 666 /dev/kfd #to varify the libraries\driver are\is working /opt/rocm/bin/rocminfo ```

GiteaMirror commented

2026-04-12 11:19:12 -05:00

@Tamila-2017 commented on GitHub (Sep 22, 2024):

Wow! Thank you so much you dear mnccouk, for your quick and creative help!

I'm immediately proceeding with the 4th install ubuntu-22.04.4-live-server-amd64 :-)

@Tamila-2017 commented on GitHub (Sep 22, 2024): Wow! Thank you so much you dear **mnccouk**, for your quick and creative help! I'm immediately proceeding with the 4th install ubuntu-22.04.4-live-server-amd64 :-)

GiteaMirror commented

2026-04-12 11:19:12 -05:00

@kth8 commented on GitHub (Sep 22, 2024):

I did a fresh install of ubuntu 22.04.5 then updated and rebooted but getting stuck at the sudo apt install amdgpu-dkms part with this error

Setting up amdgpu-dkms-firmware (1:6.2.4.50701-1664922.22.04) ...
Setting up dctrl-tools (2.24-3build2) ...
Setting up dkms (2.8.7-2ubuntu2.2) ...
Setting up amdgpu-dkms (1:6.2.4.50701-1664922.22.04) ...
Loading new amdgpu-6.2.4-1664922.22.04 DKMS files...
Building for 6.8.0-45-generic
Building for architecture x86_64
Building initial module for 6.8.0-45-generic
ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/amdgpu-dkms.0.crash'
Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64)
Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information.
dpkg: error processing package amdgpu-dkms (--configure):
 installed amdgpu-dkms package post-installation script subprocess returned error exit status 10
Processing triggers for man-db (2.10.2-1) ...
Errors were encountered while processing:
 amdgpu-dkms
E: Sub-process /usr/bin/dpkg returned an error code (1)

from that make.log: https://termbin.com/hjob

@kth8 commented on GitHub (Sep 22, 2024): I did a fresh install of ubuntu 22.04.5 then updated and rebooted but getting stuck at the `sudo apt install amdgpu-dkms` part with this error ``` Setting up amdgpu-dkms-firmware (1:6.2.4.50701-1664922.22.04) ... Setting up dctrl-tools (2.24-3build2) ... Setting up dkms (2.8.7-2ubuntu2.2) ... Setting up amdgpu-dkms (1:6.2.4.50701-1664922.22.04) ... Loading new amdgpu-6.2.4-1664922.22.04 DKMS files... Building for 6.8.0-45-generic Building for architecture x86_64 Building initial module for 6.8.0-45-generic ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/amdgpu-dkms.0.crash' Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64) Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information. dpkg: error processing package amdgpu-dkms (--configure): installed amdgpu-dkms package post-installation script subprocess returned error exit status 10 Processing triggers for man-db (2.10.2-1) ... Errors were encountered while processing: amdgpu-dkms E: Sub-process /usr/bin/dpkg returned an error code (1) ``` from that make.log: https://termbin.com/hjob

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 23, 2024):

Dear mnccouk,

Your magical script executed without errors, only I had to split it into two parts, where the

sudo reboot

However, I got the same result as before, in which I don't see a graphics card, only a Celeron CPU:

~$ /opt/rocm/bin/rocminfo 
ROCk module is loaded
=====================    
HSA System Attributes    
=====================    
Runtime Version:         1.1
System Timestamp Freq.:  1000.000000MHz
Sig. Max Wait Duration:  18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model:           LARGE                              
System Endianness:       LITTLE                             
Mwaitx:                  DISABLED
DMAbuf Support:          YES

==========               
HSA Agents               
==========               
*******                  
Agent 1                  
*******                  
  Name:                    Intel(R) Celeron(R) CPU G1620 @ 2.70GHz
  Uuid:                    CPU-XX                             
  Marketing Name:          Intel(R) Celeron(R) CPU G1620 @ 2.70GHz
  Vendor Name:             CPU                                
  Feature:                 None specified                     
  Profile:                 FULL_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        0(0x0)                             
  Queue Min Size:          0(0x0)                             
  Queue Max Size:          0(0x0)                             
  Queue Type:              MULTI                              
  Node:                    0                                  
  Device Type:             CPU                                
  Cache Info:              
    L1:                      32768(0x8000) KB                   
  Chip ID:                 0(0x0)                             
  ASIC Revision:           0(0x0)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   2700                               
  BDFID:                   0                                  
  Internal Node ID:        0                                  
  Compute Unit:            2                                  
  SIMDs per CU:            0                                  
  Shader Engines:          0                                  
  Shader Arrs. per Eng.:   0                                  
  WatchPts on Addr. Ranges:1                                  
  Features:                None
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    3970100(0x3c9434) KB               
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 2                   
      Segment:                 GLOBAL; FLAGS: KERNARG, FINE GRAINED
      Size:                    3970100(0x3c9434) KB               
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    3970100(0x3c9434) KB               
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
  ISA Info:                
*** Done ***

What could be the reason for this failure?

@Tamila-2017 commented on GitHub (Sep 23, 2024): Dear **mnccouk**, Your magical script executed without errors, only I had to split it into two parts, where the `sudo reboot` However, I got the same result as before, in which I don't see a graphics card, only a Celeron CPU: ``` ~$ /opt/rocm/bin/rocminfo ROCk module is loaded ===================== HSA System Attributes ===================== Runtime Version: 1.1 System Timestamp Freq.: 1000.000000MHz Sig. Max Wait Duration: 18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count) Machine Model: LARGE System Endianness: LITTLE Mwaitx: DISABLED DMAbuf Support: YES ========== HSA Agents ========== ******* Agent 1 ******* Name: Intel(R) Celeron(R) CPU G1620 @ 2.70GHz Uuid: CPU-XX Marketing Name: Intel(R) Celeron(R) CPU G1620 @ 2.70GHz Vendor Name: CPU Feature: None specified Profile: FULL_PROFILE Float Round Mode: NEAR Max Queue Number: 0(0x0) Queue Min Size: 0(0x0) Queue Max Size: 0(0x0) Queue Type: MULTI Node: 0 Device Type: CPU Cache Info: L1: 32768(0x8000) KB Chip ID: 0(0x0) ASIC Revision: 0(0x0) Cacheline Size: 64(0x40) Max Clock Freq. (MHz): 2700 BDFID: 0 Internal Node ID: 0 Compute Unit: 2 SIMDs per CU: 0 Shader Engines: 0 Shader Arrs. per Eng.: 0 WatchPts on Addr. Ranges:1 Features: None Pool Info: Pool 1 Segment: GLOBAL; FLAGS: FINE GRAINED Size: 3970100(0x3c9434) KB Allocatable: TRUE Alloc Granule: 4KB Alloc Alignment: 4KB Accessible by all: TRUE Pool 2 Segment: GLOBAL; FLAGS: KERNARG, FINE GRAINED Size: 3970100(0x3c9434) KB Allocatable: TRUE Alloc Granule: 4KB Alloc Alignment: 4KB Accessible by all: TRUE Pool 3 Segment: GLOBAL; FLAGS: COARSE GRAINED Size: 3970100(0x3c9434) KB Allocatable: TRUE Alloc Granule: 4KB Alloc Alignment: 4KB Accessible by all: TRUE ISA Info: *** Done *** ``` What could be the reason for this failure?

GiteaMirror commented

@SirDubbins commented on GitHub (Sep 23, 2024):

I did a fresh install of ubuntu 22.04.5 then updated and rebooted but getting stuck at the sudo apt install amdgpu-dkms part with this error

Setting up amdgpu-dkms-firmware (1:6.2.4.50701-1664922.22.04) ...
Setting up dctrl-tools (2.24-3build2) ...
Setting up dkms (2.8.7-2ubuntu2.2) ...
Setting up amdgpu-dkms (1:6.2.4.50701-1664922.22.04) ...
Loading new amdgpu-6.2.4-1664922.22.04 DKMS files...
Building for 6.8.0-45-generic
Building for architecture x86_64
Building initial module for 6.8.0-45-generic
ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/amdgpu-dkms.0.crash'
Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64)
Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information.
dpkg: error processing package amdgpu-dkms (--configure):
 installed amdgpu-dkms package post-installation script subprocess returned error exit status 10
Processing triggers for man-db (2.10.2-1) ...
Errors were encountered while processing:
 amdgpu-dkms
E: Sub-process /usr/bin/dpkg returned an error code (1)

from that make.log: https://termbin.com/hjob

I think i encountered a similar problem. I wound up deleting the 6.8.0-45 generic kernel and am using the older 5.19.0-32 generic kernel instead.

@SirDubbins commented on GitHub (Sep 23, 2024): > I did a fresh install of ubuntu 22.04.5 then updated and rebooted but getting stuck at the `sudo apt install amdgpu-dkms` part with this error > > ``` > Setting up amdgpu-dkms-firmware (1:6.2.4.50701-1664922.22.04) ... > Setting up dctrl-tools (2.24-3build2) ... > Setting up dkms (2.8.7-2ubuntu2.2) ... > Setting up amdgpu-dkms (1:6.2.4.50701-1664922.22.04) ... > Loading new amdgpu-6.2.4-1664922.22.04 DKMS files... > Building for 6.8.0-45-generic > Building for architecture x86_64 > Building initial module for 6.8.0-45-generic > ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/amdgpu-dkms.0.crash' > Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64) > Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information. > dpkg: error processing package amdgpu-dkms (--configure): > installed amdgpu-dkms package post-installation script subprocess returned error exit status 10 > Processing triggers for man-db (2.10.2-1) ... > Errors were encountered while processing: > amdgpu-dkms > E: Sub-process /usr/bin/dpkg returned an error code (1) > ``` > > from that make.log: https://termbin.com/hjob I think i encountered a similar problem. I wound up deleting the 6.8.0-45 generic kernel and am using the older 5.19.0-32 generic kernel instead.

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 23, 2024):

SirDubbins,

I used to get this error all the time too.
But when I used magical the script by mnccouk, I didn't notice it, because the lines on the screen were running very fast.
Maybe it didn't exist?

@Tamila-2017 commented on GitHub (Sep 23, 2024): SirDubbins, I used to get this error all the time too. But when I used magical the script by [mnccouk](https://github.com/ollama/ollama/issues/2453#issuecomment-2366982275), I didn't notice it, because the lines on the screen were running very fast. Maybe it didn't exist?

GiteaMirror commented

2026-04-12 11:19:14 -05:00

@mnccouk commented on GitHub (Sep 23, 2024):

@Tamila-2017
Probably best to run that script step by step, one command at a time, After each command validate you have no errors, apologies for the confusion. You may have had similar issue to @kth8.but not noticed due to lack of error trapping when running all the commands at once.

@SirDubbins proposed a fix, by trying again after downgrading your kernel, or reinstall ubuntu again(sorry!).

Looking at the doc - https://rocm.docs.amd.com/en/docs-5.7.1/release/gpu_os_support.html - Ubuntu 22.04.3 has been validated with kernel 6.2. Try that version(22.04.3) of Ubuntu instead, also don't install any updates as part of the Ubuntu installation.

I went from an earlier of Ubuntu 20.04 then upgraded through to Ubuntu 22.04.4, the whole process wasn't smooth for me either. It's quite possible the kernel drivers could have compiled against an earlier kernel version.

Just for info - I'm actually on kernel version 6.8.0-40-generic now.

Sorry, but you may have to target an install of (Ubuntu 22.04.3) or try downgrading your kernel but that introduce other issues. I'd opt for the fresh install again if you have the luxury of being able to.

@mnccouk commented on GitHub (Sep 23, 2024): @Tamila-2017 Probably best to run that script step by step, one command at a time, After each command validate you have no errors, apologies for the confusion. You may have had similar issue to @kth8.but not noticed due to lack of error trapping when running all the commands at once. @SirDubbins proposed a fix, by trying again after downgrading your kernel, or reinstall ubuntu again(sorry!). Looking at the doc - https://rocm.docs.amd.com/en/docs-5.7.1/release/gpu_os_support.html - Ubuntu 22.04.3 has been validated with kernel 6.2. Try that version(22.04.3) of Ubuntu instead, also don't install any updates as part of the Ubuntu installation. I went from an earlier of Ubuntu 20.04 then upgraded through to Ubuntu 22.04.4, the whole process wasn't smooth for me either. It's quite possible the kernel drivers could have compiled against an earlier kernel version. Just for info - I'm actually on kernel version 6.8.0-40-generic now. Sorry, but you may have to target an install of (Ubuntu 22.04.3) or try downgrading your kernel but that introduce other issues. I'd opt for the fresh install again if you have the luxury of being able to.

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 23, 2024):

Dear mnccouk,

I've reinstalled again ubuntu-22.04.4-live-server-amd64.iso

uname -a
Linux ai-server 5.15.0-122-generic #132-Ubuntu SMP Thu Aug 29 13:45:52 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

Then, carefully step by step, I executed each command of your magic script separately, watching for mistakes.
There were no mistakes.

However, I got the same result without the GPU -

/opt/rocm/bin/rocminfo
ROCk module is loaded
=====================    
HSA System Attributes    
=====================    
Runtime Version:         1.1
System Timestamp Freq.:  1000.000000MHz
Sig. Max Wait Duration:  18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model:           LARGE                              
System Endianness:       LITTLE                             
Mwaitx:                  DISABLED
DMAbuf Support:          YES

==========               
HSA Agents               
==========               
*******                  
Agent 1                  
*******                  
  Name:                    Intel(R) Celeron(R) CPU G1620 @ 2.70GHz
  Uuid:                    CPU-XX                             
  Marketing Name:          Intel(R) Celeron(R) CPU G1620 @ 2.70GHz
  Vendor Name:             CPU                                
  Feature:                 None specified                     
  Profile:                 FULL_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        0(0x0)                             
  Queue Min Size:          0(0x0)                             
  Queue Max Size:          0(0x0)                             
  Queue Type:              MULTI                              
  Node:                    0                                  
  Device Type:             CPU                                
  Cache Info:              
    L1:                      32768(0x8000) KB                   
  Chip ID:                 0(0x0)                             
  ASIC Revision:           0(0x0)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   2700                               
  BDFID:                   0                                  
  Internal Node ID:        0                                  
  Compute Unit:            2                                  
  SIMDs per CU:            0                                  
  Shader Engines:          0                                  
  Shader Arrs. per Eng.:   0                                  
  WatchPts on Addr. Ranges:1                                  
  Features:                None
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    3970100(0x3c9434) KB               
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 2                   
      Segment:                 GLOBAL; FLAGS: KERNARG, FINE GRAINED
      Size:                    3970100(0x3c9434) KB               
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    3970100(0x3c9434) KB               
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
  ISA Info:                
*** Done ***

What do I do next?

@Tamila-2017 commented on GitHub (Sep 23, 2024): Dear **mnccouk**, I've reinstalled again **ubuntu-22.04.4-live-server-amd64.iso** ``` uname -a Linux ai-server 5.15.0-122-generic #132-Ubuntu SMP Thu Aug 29 13:45:52 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux ``` Then, carefully step by step, I executed each command of your [magic script](https://github.com/ollama/ollama/issues/2453#issuecomment-2366982275) separately, watching for mistakes. There were no mistakes. However, I got the same result without the GPU - ``` /opt/rocm/bin/rocminfo ROCk module is loaded ===================== HSA System Attributes ===================== Runtime Version: 1.1 System Timestamp Freq.: 1000.000000MHz Sig. Max Wait Duration: 18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count) Machine Model: LARGE System Endianness: LITTLE Mwaitx: DISABLED DMAbuf Support: YES ========== HSA Agents ========== ******* Agent 1 ******* Name: Intel(R) Celeron(R) CPU G1620 @ 2.70GHz Uuid: CPU-XX Marketing Name: Intel(R) Celeron(R) CPU G1620 @ 2.70GHz Vendor Name: CPU Feature: None specified Profile: FULL_PROFILE Float Round Mode: NEAR Max Queue Number: 0(0x0) Queue Min Size: 0(0x0) Queue Max Size: 0(0x0) Queue Type: MULTI Node: 0 Device Type: CPU Cache Info: L1: 32768(0x8000) KB Chip ID: 0(0x0) ASIC Revision: 0(0x0) Cacheline Size: 64(0x40) Max Clock Freq. (MHz): 2700 BDFID: 0 Internal Node ID: 0 Compute Unit: 2 SIMDs per CU: 0 Shader Engines: 0 Shader Arrs. per Eng.: 0 WatchPts on Addr. Ranges:1 Features: None Pool Info: Pool 1 Segment: GLOBAL; FLAGS: FINE GRAINED Size: 3970100(0x3c9434) KB Allocatable: TRUE Alloc Granule: 4KB Alloc Alignment: 4KB Accessible by all: TRUE Pool 2 Segment: GLOBAL; FLAGS: KERNARG, FINE GRAINED Size: 3970100(0x3c9434) KB Allocatable: TRUE Alloc Granule: 4KB Alloc Alignment: 4KB Accessible by all: TRUE Pool 3 Segment: GLOBAL; FLAGS: COARSE GRAINED Size: 3970100(0x3c9434) KB Allocatable: TRUE Alloc Granule: 4KB Alloc Alignment: 4KB Accessible by all: TRUE ISA Info: *** Done *** ``` What do I do next?

GiteaMirror commented

2026-04-12 11:19:14 -05:00

@mnccouk commented on GitHub (Sep 23, 2024):

What's the output of

cat /var/log/dmesg | grep -e amdgpu -e drm

@mnccouk commented on GitHub (Sep 23, 2024): What's the output of ``` cat /var/log/dmesg | grep -e amdgpu -e drm ```

GiteaMirror commented

2026-04-12 11:19:15 -05:00

@Tamila-2017 commented on GitHub (Sep 23, 2024):

# cat /var/log/dmesg | grep -e amdgpu -e drm
cat: /var/log/dmesg: No such file or directory

# ls
alternatives.log  btmp                   dist-upgrade  installer  private              wtmp
apt               cloud-init-output.log  dpkg.log      journal    rocm_smi_lib
bootstrap.log     cloud-init.log         faillog       lastlog    unattended-upgrades

@Tamila-2017 commented on GitHub (Sep 23, 2024): ``` # cat /var/log/dmesg | grep -e amdgpu -e drm cat: /var/log/dmesg: No such file or directory ``` ``` # ls alternatives.log btmp dist-upgrade installer private wtmp apt cloud-init-output.log dpkg.log journal rocm_smi_lib bootstrap.log cloud-init.log faillog lastlog unattended-upgrades ```

GiteaMirror commented

2026-04-12 11:19:15 -05:00

@mnccouk commented on GitHub (Sep 23, 2024):

How about?

sudo dmesg | grep -e amdgpu -e drm

@mnccouk commented on GitHub (Sep 23, 2024): How about? ``` sudo dmesg | grep -e amdgpu -e drm ```

GiteaMirror commented

2026-04-12 11:19:15 -05:00

@mnccouk commented on GitHub (Sep 23, 2024):

And...

lsmod | grep amdgpu

@mnccouk commented on GitHub (Sep 23, 2024): And... ``` lsmod | grep amdgpu ```

GiteaMirror commented

2026-04-12 11:19:16 -05:00

@Tamila-2017 commented on GitHub (Sep 23, 2024):

sudo dmesg | grep -e amdgpu -e drm
[    5.127309] [drm] amdgpu kernel modesetting enabled.
[    5.156647] [drm] amdgpu version: 6.2.4
[    5.185611] [drm] OS DRM version: 5.15.0
[    5.244324] amdgpu: CRAT table not found
[    5.305139] amdgpu: Virtual CRAT table created for CPU
[    5.368709] amdgpu: Topology: Add CPU node
[    5.448129] amdgpu: PeerDirect support was initialized successfully
[    5.480922] [drm] initializing kernel modesetting (POLARIS10 0x1002:0x67DF 0x1DA2:0xE366 0xE7).
[    5.513856] [drm] register mmio base: 0xF7E00000
[    5.546704] [drm] register mmio size: 262144
[    5.579527] [drm] add ip block number 0 <vi_common>
[    5.612338] [drm] add ip block number 1 <gmc_v8_0>
[    5.645021] [drm] add ip block number 2 <tonga_ih>
[    5.677367] [drm] add ip block number 3 <gfx_v8_0>
[    5.709378] [drm] add ip block number 4 <sdma_v3_0>
[    5.741213] [drm] add ip block number 5 <powerplay>
[    5.772408] [drm] add ip block number 6 <dm>
[    5.802951] [drm] add ip block number 7 <uvd_v6_0>
[    5.833281] [drm] add ip block number 8 <vce_v3_0>
[    5.925618] amdgpu 0000:01:00.0: No more image in the PCI ROM
[    5.956693] amdgpu 0000:01:00.0: amdgpu: Fetched VBIOS from ROM BAR
[    5.987774] amdgpu: ATOM BIOS: 113-1E3660U-O51
[    6.018348] [drm] UVD is enabled in VM mode
[    6.048418] [drm] UVD ENC is enabled in VM mode
[    6.078313] [drm] VCE enabled in VM mode
[    6.108367] amdgpu 0000:01:00.0: vgaarb: deactivate vga console
[    6.108372] amdgpu 0000:01:00.0: amdgpu: Trusted Memory Zone (TMZ) feature not supported
[    6.108376] amdgpu 0000:01:00.0: amdgpu: PCIE atomic ops is not supported
[    6.108593] [drm] vm size is 64 GB, 2 levels, block size is 10-bit, fragment size is 9-bit
[    6.108638] amdgpu 0000:01:00.0: amdgpu: VRAM: 8192M 0x000000F400000000 - 0x000000F5FFFFFFFF (8192M used)
[    6.108643] amdgpu 0000:01:00.0: amdgpu: GART: 256M 0x000000FF00000000 - 0x000000FF0FFFFFFF
[    6.108734] [drm] Detected VRAM RAM=8192M, BAR=256M
[    6.108738] [drm] RAM width 256bits GDDR5
[    6.108788] [drm] amdgpu: 8192M of VRAM memory ready
[    6.108792] [drm] amdgpu: 1938M of GTT memory ready.
[    6.108862] [drm] GART: num cpu pages 65536, num gpu pages 65536
[    6.109304] [drm] PCIE GART of 256M enabled (table at 0x000000F400800000).
[    6.109428] [drm] Chained IB support enabled!
[    6.110196] amdgpu: [powerplay] hwmgr_sw_init smu backed is polaris10_smu
[    6.110341] [drm] Found UVD firmware Version: 1.130 Family ID: 16
[    6.110955] [drm] Found VCE firmware Version: 53.26 Binary ID: 3
[    6.179076] [drm] Display Core v3.2.241 initialized on DCE 11.2
[    6.274702] [drm] UVD and UVD ENC initialized successfully.
[    6.384641] [drm] VCE initialized successfully.
[    6.385053] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0
[    6.385072] amdgpu 0000:01:00.0: amdgpu: SE 4, SH per SE 1, CU per SH 9, active_cu_number 36
[    6.388350] amdgpu: legacy kernel without apple_gmux_detect()
[    6.388356] amdgpu 0000:01:00.0: amdgpu: Using BACO for runtime pm
[    6.389279] [drm] Initialized amdgpu 3.54.0 20150101 for 0000:01:00.0 on minor 0
[    6.403933] fbcon: amdgpudrmfb (fb0) is primary device
[    6.479039] amdgpu 0000:01:00.0: [drm] fb0: amdgpudrmfb frame buffer device
[    7.845748] systemd[1]: Starting Load Kernel Module drm...
[    9.367404] snd_hda_intel 0000:01:00.1: bound 0000:01:00.0 (ops amdgpu_dm_audio_component_bind_ops [amdgpu])

lsmod | grep amdgpu
amdgpu              13590528  0
amddrm_ttm_helper      16384  1 amdgpu
amdttm                 94208  2 amdgpu,amddrm_ttm_helper
amdxcp                 16384  1 amdgpu
iommu_v2               24576  1 amdgpu
amddrm_buddy           20480  1 amdgpu
amd_sched              49152  1 amdgpu
amdkcl                 32768  3 amd_sched,amdttm,amdgpu
i2c_algo_bit           16384  1 amdgpu
drm_kms_helper        311296  3 amdgpu
drm                   622592  9 drm_kms_helper,amd_sched,amdttm,amdgpu,amddrm_buddy,amdkcl,amddrm_ttm_helper,amdxcp

@Tamila-2017 commented on GitHub (Sep 23, 2024): ``` sudo dmesg | grep -e amdgpu -e drm [ 5.127309] [drm] amdgpu kernel modesetting enabled. [ 5.156647] [drm] amdgpu version: 6.2.4 [ 5.185611] [drm] OS DRM version: 5.15.0 [ 5.244324] amdgpu: CRAT table not found [ 5.305139] amdgpu: Virtual CRAT table created for CPU [ 5.368709] amdgpu: Topology: Add CPU node [ 5.448129] amdgpu: PeerDirect support was initialized successfully [ 5.480922] [drm] initializing kernel modesetting (POLARIS10 0x1002:0x67DF 0x1DA2:0xE366 0xE7). [ 5.513856] [drm] register mmio base: 0xF7E00000 [ 5.546704] [drm] register mmio size: 262144 [ 5.579527] [drm] add ip block number 0 <vi_common> [ 5.612338] [drm] add ip block number 1 <gmc_v8_0> [ 5.645021] [drm] add ip block number 2 <tonga_ih> [ 5.677367] [drm] add ip block number 3 <gfx_v8_0> [ 5.709378] [drm] add ip block number 4 <sdma_v3_0> [ 5.741213] [drm] add ip block number 5 <powerplay> [ 5.772408] [drm] add ip block number 6 <dm> [ 5.802951] [drm] add ip block number 7 <uvd_v6_0> [ 5.833281] [drm] add ip block number 8 <vce_v3_0> [ 5.925618] amdgpu 0000:01:00.0: No more image in the PCI ROM [ 5.956693] amdgpu 0000:01:00.0: amdgpu: Fetched VBIOS from ROM BAR [ 5.987774] amdgpu: ATOM BIOS: 113-1E3660U-O51 [ 6.018348] [drm] UVD is enabled in VM mode [ 6.048418] [drm] UVD ENC is enabled in VM mode [ 6.078313] [drm] VCE enabled in VM mode [ 6.108367] amdgpu 0000:01:00.0: vgaarb: deactivate vga console [ 6.108372] amdgpu 0000:01:00.0: amdgpu: Trusted Memory Zone (TMZ) feature not supported [ 6.108376] amdgpu 0000:01:00.0: amdgpu: PCIE atomic ops is not supported [ 6.108593] [drm] vm size is 64 GB, 2 levels, block size is 10-bit, fragment size is 9-bit [ 6.108638] amdgpu 0000:01:00.0: amdgpu: VRAM: 8192M 0x000000F400000000 - 0x000000F5FFFFFFFF (8192M used) [ 6.108643] amdgpu 0000:01:00.0: amdgpu: GART: 256M 0x000000FF00000000 - 0x000000FF0FFFFFFF [ 6.108734] [drm] Detected VRAM RAM=8192M, BAR=256M [ 6.108738] [drm] RAM width 256bits GDDR5 [ 6.108788] [drm] amdgpu: 8192M of VRAM memory ready [ 6.108792] [drm] amdgpu: 1938M of GTT memory ready. [ 6.108862] [drm] GART: num cpu pages 65536, num gpu pages 65536 [ 6.109304] [drm] PCIE GART of 256M enabled (table at 0x000000F400800000). [ 6.109428] [drm] Chained IB support enabled! [ 6.110196] amdgpu: [powerplay] hwmgr_sw_init smu backed is polaris10_smu [ 6.110341] [drm] Found UVD firmware Version: 1.130 Family ID: 16 [ 6.110955] [drm] Found VCE firmware Version: 53.26 Binary ID: 3 [ 6.179076] [drm] Display Core v3.2.241 initialized on DCE 11.2 [ 6.274702] [drm] UVD and UVD ENC initialized successfully. [ 6.384641] [drm] VCE initialized successfully. [ 6.385053] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0 [ 6.385072] amdgpu 0000:01:00.0: amdgpu: SE 4, SH per SE 1, CU per SH 9, active_cu_number 36 [ 6.388350] amdgpu: legacy kernel without apple_gmux_detect() [ 6.388356] amdgpu 0000:01:00.0: amdgpu: Using BACO for runtime pm [ 6.389279] [drm] Initialized amdgpu 3.54.0 20150101 for 0000:01:00.0 on minor 0 [ 6.403933] fbcon: amdgpudrmfb (fb0) is primary device [ 6.479039] amdgpu 0000:01:00.0: [drm] fb0: amdgpudrmfb frame buffer device [ 7.845748] systemd[1]: Starting Load Kernel Module drm... [ 9.367404] snd_hda_intel 0000:01:00.1: bound 0000:01:00.0 (ops amdgpu_dm_audio_component_bind_ops [amdgpu]) ``` ``` lsmod | grep amdgpu amdgpu 13590528 0 amddrm_ttm_helper 16384 1 amdgpu amdttm 94208 2 amdgpu,amddrm_ttm_helper amdxcp 16384 1 amdgpu iommu_v2 24576 1 amdgpu amddrm_buddy 20480 1 amdgpu amd_sched 49152 1 amdgpu amdkcl 32768 3 amd_sched,amdttm,amdgpu i2c_algo_bit 16384 1 amdgpu drm_kms_helper 311296 3 amdgpu drm 622592 9 drm_kms_helper,amd_sched,amdttm,amdgpu,amddrm_buddy,amdkcl,amddrm_ttm_helper,amdxcp ```

GiteaMirror commented

2026-04-12 11:19:16 -05:00

@mnccouk commented on GitHub (Sep 23, 2024):

I'm running out of ideas here, but I don't like the look of:

[    6.385053] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0

Think it maybe to do with PCI spec of your hardware - see this link for related topic - https://www.reddit.com/r/ROCm/comments/ba6tvq/atomics_on_my_hardware/

@mnccouk commented on GitHub (Sep 23, 2024): I'm running out of ideas here, but I don't like the look of: ``` [ 6.385053] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0 ``` Think it maybe to do with PCI spec of your hardware - see this link for related topic - https://www.reddit.com/r/ROCm/comments/ba6tvq/atomics_on_my_hardware/

GiteaMirror commented

2026-04-12 11:19:17 -05:00

@Tamila-2017 commented on GitHub (Sep 23, 2024):

It's incredible! You are doing well, but I am not, even though I follow your instructions exactly

So I offer you remote access via ssh, and you personally can install this cranky ROCm 5.7.1

PS. I used to be able to mine Bitcoin, Litecoin and Ethereum Classic on the same motherboard MSI H61MU-E35 (B3) and AMD RX580 without any problems.
I don't understand why they refuse to work together with ROCm :-o

@Tamila-2017 commented on GitHub (Sep 23, 2024): It's incredible! You are doing well, but I am not, even though I follow your instructions exactly So I offer you remote access via ssh, and you personally can install this cranky ROCm 5.7.1 PS. I used to be able to mine Bitcoin, Litecoin and Ethereum Classic on the same motherboard MSI H61MU-E35 (B3) and AMD RX580 without any problems. I don't understand why they refuse to work together with ROCm :-o

GiteaMirror commented

2026-04-12 11:19:17 -05:00

@mnccouk commented on GitHub (Sep 23, 2024):

I can't be certain on this, but what I'm reading is the ROCm requires PCIe 3.0 AtomicOp feature, looking at the spec of your CPU - https://www.intel.com/content/www/us/en/products/sku/71073/intel-celeron-processor-g1620-2m-cache-2-70-ghz/specifications.html
It looks like it supports PCIe revision 2.0 which unfortunately is not compatible with the ROCm system. Also motherboard seems to be PCIe 2.x too.

I'm sorry to say I don't think ROCm will work for you with that CPU\Motherboard GPU combination, I maybe wrong but that's I think. Open to any other thoughts.

@mnccouk commented on GitHub (Sep 23, 2024): I can't be certain on this, but what I'm reading is the ROCm requires PCIe 3.0 AtomicOp feature, looking at the spec of your CPU - https://www.intel.com/content/www/us/en/products/sku/71073/intel-celeron-processor-g1620-2m-cache-2-70-ghz/specifications.html It looks like it supports PCIe revision 2.0 which unfortunately is not compatible with the ROCm system. Also motherboard seems to be PCIe 2.x too. I'm sorry to say I don't think ROCm will work for you with that CPU\Motherboard GPU combination, I maybe wrong but that's I think. Open to any other thoughts.

GiteaMirror commented

2026-04-12 11:19:17 -05:00

@Tamila-2017 commented on GitHub (Sep 23, 2024):

Dear mnccouk 💓

You have no need to apologize :-) On the contrary, I am very grateful to you for your courtesy and your great work you have done to analyze my situation and create a magic script to build ROCm.
I think you are right about my outdated hardware.
I will consider buying a more modern CPU and Motherboard, and I would be grateful if you could tell me the system requirements for them.

@Tamila-2017 commented on GitHub (Sep 23, 2024): Dear **mnccouk** 💓 You have no need to apologize :-) On the contrary, I am very grateful to you for your courtesy and your great work you have done to analyze my situation and create a magic script to build ROCm. I think you are right about my outdated hardware. I will consider buying a more modern CPU and Motherboard, and I would be grateful if you could tell me the system requirements for them.

GiteaMirror commented

@mnccouk commented on GitHub (Sep 24, 2024):

@Tamila-2017

I can only forward the motherboard and CPU I'm using as I know that would work:
Motherboard - B450M PRO-VDH MAX - https://www.msi.com/Motherboard/B450M-PRO-VDH-MAX/support
CPU - AMD Ryzen 5 3400G with Radeon Vega Graphics
And of course the RX580

There is a discussion here - https://github.com/ROCm/ROCm/issues/237 about a similar issue to what you have, except in this instance it's just one of the PCIe ports that is failing when attempting to use two graphics cards, even though it has PCIe 3.0 specification.

Maybe others would share what hardware(motherboard\cpu combination) they have successfully used with ROCm to give you some alternatives?

@mnccouk commented on GitHub (Sep 24, 2024): @Tamila-2017 I can only forward the motherboard and CPU I'm using as I know that would work: Motherboard - B450M PRO-VDH MAX - https://www.msi.com/Motherboard/B450M-PRO-VDH-MAX/support CPU - AMD Ryzen 5 3400G with Radeon Vega Graphics And of course the RX580 There is a discussion here - https://github.com/ROCm/ROCm/issues/237 about a similar issue to what you have, except in this instance it's just one of the PCIe ports that is failing when attempting to use two graphics cards, even though it has PCIe 3.0 specification. Maybe others would share what hardware(motherboard\cpu combination) they have successfully used with ROCm to give you some alternatives?

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 24, 2024):

Ok. Guys, how do you measure speed in tokens/second? I want to measure the speed of CPU N100.

@Tamila-2017 commented on GitHub (Sep 24, 2024): Ok. Guys, how do you measure speed in tokens/second? I want to measure the speed of CPU N100.

GiteaMirror commented

@bellmancity commented on GitHub (Sep 24, 2024):

Ok. Guys, how do you measure speed in tokens/second? I want to measure the speed of CPU N100.
I think this is the command, in ollama prompt.
/set verbose

@bellmancity commented on GitHub (Sep 24, 2024): > Ok. Guys, how do you measure speed in tokens/second? I want to measure the speed of CPU N100. I think this is the command, in ollama prompt. /set verbose

GiteaMirror commented

2026-04-12 11:19:19 -05:00

@Tamila-2017 commented on GitHub (Sep 24, 2024):

/set verbose

So I need to run some model to measure the speed?
I ran the Llama 3.1 model and executed your command, but it did not report the speed:

>>> /set verbose
Set 'verbose' mode.

@Tamila-2017 commented on GitHub (Sep 24, 2024): > /set verbose So I need to run some model to measure the speed? I ran the Llama 3.1 model and executed your command, but it did not report the speed: ``` >>> /set verbose Set 'verbose' mode. ```

GiteaMirror commented

@mnccouk commented on GitHub (Sep 24, 2024):

Then you make a prompt:-

@mnccouk commented on GitHub (Sep 24, 2024): Then you make a prompt:- ![Screenshot from 2024-09-24 20-36-45](https://github.com/user-attachments/assets/ea587bb0-3028-4d17-86fa-61e69ac34afd)

GiteaMirror commented

2026-04-12 11:19:19 -05:00

@Tamila-2017 commented on GitHub (Sep 24, 2024):

Surprisingly, I got a different answer:

~$ ollama run llama3.1
/set verbose
Set 'verbose' mode.
why is the sky blue
The sky appears blue to us because of a phenomenon called scattering which occurs when sunlight interacts with the tiny molecules of gases in the atmosphere. Here's a simplified explanation:

Sunlight enters Earth's atmosphere: When the sun shines, it emits a wide range of electromagnetic radiation, including visible light, ultraviolet (UV) light, and infrared (IR) radiation.

Light scatters off gas molecules: As sunlight travels through the atmosphere, it encounters tiny molecules of gases like nitrogen (N2) and oxygen (O2). These molecules are much smaller than the wavelength of light, so they scatter the shorter (blue) wavelengths more efficiently than the longer (red) wavelengths.

Blue light is scattered in all directions: The scattering of blue light by gas molecules causes it to be distributed throughout the atmosphere in all directions. This means that we see this scattered blue light from almost everywhere around us.

Red light continues straight: On the other hand, the longer wavelengths (like red and infrared) continue traveling in a more direct path, with less scattering occurring. As a result, these colors are not as prominently visible to our eyes.

Why doesn't the sky always appear blue?

Time of day: During sunrise and sunset, the sun's light travels through more of the Earth's atmosphere, which scatters shorter wavelengths > like blue even further. This results in the appearance of reds and oranges.

Clouds and dust: When clouds or atmospheric particles are present, they can absorb or scatter light in different ways, affecting the apparent color of the sky.

Atmospheric conditions: Weather events, pollution, or other factors can alter the way light interacts with the atmosphere, making the sky appear more hazy, gray, or even greenish.

Additional fun facts:

The sky appears blue to us because our eyes are more sensitive to the shorter (blue) wavelengths of light.

If we were to live on a planet with a different atmospheric composition or no atmosphere at all, the sky might appear differently colored or even not blue at all!

Some people claim that the sky can appear violet under certain conditions. However, this is relatively rare and usually requires specific optical phenomena.

I hope you enjoyed learning about why the sky appears blue!

And the performance of Llama 3.1 on CPU N100 turned out like this:

@Tamila-2017 commented on GitHub (Sep 24, 2024): Surprisingly, I got a different answer: > ~$ ollama run llama3.1 > /set verbose > Set 'verbose' mode. > why is the sky blue > The sky appears blue to us because of a phenomenon called scattering which occurs when sunlight interacts with the tiny molecules of gases in the atmosphere. Here's a simplified explanation: > > 1. **Sunlight enters Earth's atmosphere**: When the sun shines, it emits a wide range of electromagnetic radiation, including visible light, ultraviolet (UV) light, and infrared (IR) radiation. > 2. **Light scatters off gas molecules**: As sunlight travels through the atmosphere, it encounters tiny molecules of gases like nitrogen (N2) and oxygen (O2). These molecules are much smaller than the wavelength of light, so they scatter the shorter (blue) wavelengths more efficiently than the longer (red) wavelengths. > 3. **Blue light is scattered in all directions**: The scattering of blue light by gas molecules causes it to be distributed throughout the atmosphere in all directions. This means that we see this scattered blue light from almost everywhere around us. > 4. **Red light continues straight**: On the other hand, the longer wavelengths (like red and infrared) continue traveling in a more direct path, with less scattering occurring. As a result, these colors are not as prominently visible to our eyes. > > **Why doesn't the sky always appear blue?** > > 1. **Time of day**: During sunrise and sunset, the sun's light travels through more of the Earth's atmosphere, which scatters shorter wavelengths > like blue even further. This results in the appearance of reds and oranges. > 2. **Clouds and dust**: When clouds or atmospheric particles are present, they can absorb or scatter light in different ways, affecting the apparent color of the sky. > 3. **Atmospheric conditions**: Weather events, pollution, or other factors can alter the way light interacts with the atmosphere, making the sky appear more hazy, gray, or even greenish. > > **Additional fun facts:** > > * The sky appears blue to us because our eyes are more sensitive to the shorter (blue) wavelengths of light. > * If we were to live on a planet with a different atmospheric composition or no atmosphere at all, the sky might appear differently colored or even not blue at all! > * Some people claim that the sky can appear violet under certain conditions. However, this is relatively rare and usually requires specific optical phenomena. > > I hope you enjoyed learning about why the sky appears blue! And the performance of Llama 3.1 on CPU N100 turned out like this: ![CPU_N100](https://github.com/user-attachments/assets/83b294db-3931-4eff-8047-1661f251b8b8)

GiteaMirror commented

2026-04-12 11:19:19 -05:00

@Tamila-2017 commented on GitHub (Sep 24, 2024):

The performance of Llama 3.1 on CPU Intel-i3-4330-3 50GHz:

@Tamila-2017 commented on GitHub (Sep 24, 2024): The performance of Llama 3.1 on CPU Intel-i3-4330-3 50GHz: ![Intel-i3-4330-3 50GHz](https://github.com/user-attachments/assets/8dc5f36a-969c-4a03-b0e9-748cf958e0d6)

GiteaMirror commented

2026-04-12 11:19:20 -05:00

@Tamila-2017 commented on GitHub (Sep 24, 2024):

Dear mnccouk,

You have achieved a performance of ~26 tokens/sec.
Let me clarify: this is the total performance AMD Ryzen 5 3400G + Radeon Vega Graphics + RX580 ?
Or is the performance of only one RX580 ?

@Tamila-2017 commented on GitHub (Sep 24, 2024): Dear **mnccouk**, You have achieved a performance of ~26 tokens/sec. Let me clarify: this is the total performance AMD Ryzen 5 3400G **+** Radeon Vega Graphics **+** RX580 ? Or is the performance of only one RX580 ?

GiteaMirror commented

2026-04-12 11:19:20 -05:00

@AustinPowers1935 commented on GitHub (Sep 25, 2024):

docker exec -it ollama_gpu ollama run llama3.1

/set verbose
Set 'verbose' mode.
Fix grammar and polish message:
...
... @mnccouk Thank you! Works great for models which fit into GPU memory.
...
... For those struggling with installation on Ubuntu:
... just skip "2. Add the AMDGPU Repository and Install the Kernel-mode Driver" from this instruction: https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html.
... You do not need it: recent Ubuntu releases already contain amdgpu driver.
...
... I got it worked without any problems within less than half an hour.
... 22.04.5 LTS
... 6.8.0-40-generic
... RX590
Here is the revised message with grammar and polish:

"@mnccouk Thank you! Works great for models that fit into GPU memory.

For those struggling to install on Ubuntu, here's a tip:
You can skip step 2 ("Add the AMDGPU Repository and Install the Kernel-mode Driver") from this instruction set: https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html.
You don't need it, as recent Ubuntu releases already contain the amdgpu driver.

I was able to get it working without any issues within less than half an hour on:
22.04.5 LTS
6.8.0-40-generic
RX590"

Note: I made some minor changes for clarity and readability, such as changing "those struggling with installation" to "those struggling to install", and rephrasing the sentence about skipping step 2
to make it more concise. Let me know if you'd like any further adjustments!

total duration: 31.579130243s
load duration: 28.163993ms
prompt eval count: 137 token(s)
prompt eval duration: 24.679448s
prompt eval rate: 5.55 tokens/s
eval count: 202 token(s)
eval duration: 6.829664s
eval rate: 29.58 tokens/s

@AustinPowers1935 commented on GitHub (Sep 25, 2024): > docker exec -it ollama_gpu ollama run llama3.1 > >>> /set verbose > Set 'verbose' mode. > >>> Fix grammar and polish message: > ... > ... @mnccouk Thank you! Works great for models which fit into GPU memory. > ... > ... For those struggling with installation on Ubuntu: > ... just skip "2. Add the AMDGPU Repository and Install the Kernel-mode Driver" from this instruction: https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html. > ... You do not need it: recent Ubuntu releases already contain amdgpu driver. > ... > ... I got it worked without any problems within less than half an hour. > ... 22.04.5 LTS > ... 6.8.0-40-generic > ... RX590 > Here is the revised message with grammar and polish: > > "@mnccouk Thank you! Works great for models that fit into GPU memory. > > For those struggling to install on Ubuntu, here's a tip: > You can skip step 2 ("Add the AMDGPU Repository and Install the Kernel-mode Driver") from this instruction set: https://rocm.docs.amd.com/en/docs-5.7.0/deploy/linux/os-native/install.html. > You don't need it, as recent Ubuntu releases already contain the amdgpu driver. > > I was able to get it working without any issues within less than half an hour on: > 22.04.5 LTS > 6.8.0-40-generic > RX590" > > Note: I made some minor changes for clarity and readability, such as changing "those struggling with installation" to "those struggling to install", and rephrasing the sentence about skipping step 2 > to make it more concise. Let me know if you'd like any further adjustments! > > total duration: 31.579130243s > load duration: 28.163993ms > prompt eval count: 137 token(s) > prompt eval duration: 24.679448s > prompt eval rate: 5.55 tokens/s > eval count: 202 token(s) > eval duration: 6.829664s > eval rate: 29.58 tokens/s >

GiteaMirror commented

2026-04-12 11:19:20 -05:00

@AustinPowers1935 commented on GitHub (Sep 25, 2024):

@kth8 @Tamila-2017 ^^^

@AustinPowers1935 commented on GitHub (Sep 25, 2024): @kth8 @Tamila-2017 ^^^

GiteaMirror commented

2026-04-12 11:19:21 -05:00

@kth8 commented on GitHub (Sep 25, 2024):

After doing a new install of Ubuntu 22.04 I followed the guide except step 2 of installing amdgpu-dkms but after rebooting and running rocminfo it seems my GPU is not discovered.

Here are my system logs: https://termbin.com/svjq

I am using RX 470.

@kth8 commented on GitHub (Sep 25, 2024): After doing a new install of Ubuntu 22.04 I followed the guide except step 2 of installing `amdgpu-dkms` but after rebooting and running `rocminfo` it seems my GPU is not discovered. Here are my system logs: https://termbin.com/svjq I am using RX 470.

GiteaMirror commented

2026-04-12 11:19:21 -05:00

@mnccouk commented on GitHub (Sep 25, 2024):

Dear mnccouk,

You have achieved a performance of ~26 tokens/sec. Let me clarify: this is the total performance AMD Ryzen 5 3400G + Radeon Vega Graphics + RX580 ? Or is the performance of only one RX580 ?

This is just with 1 RX580 + CPU, the CPU is still utilised to some degree but the GPU is performing all the heavy lifting.

As far as I'm aware the embedded Radeon Vega Graphics(embedded in the CPU) in not utilised.

@mnccouk commented on GitHub (Sep 25, 2024): > Dear **mnccouk**, > > You have achieved a performance of ~26 tokens/sec. Let me clarify: this is the total performance AMD Ryzen 5 3400G **+** Radeon Vega Graphics **+** RX580 ? Or is the performance of only one RX580 ? This is just with 1 RX580 + CPU, the CPU is still utilised to some degree but the GPU is performing all the heavy lifting. As far as I'm aware the embedded Radeon Vega Graphics(embedded in the CPU) in not utilised.

GiteaMirror commented

2026-04-12 11:19:22 -05:00

@mnccouk commented on GitHub (Sep 25, 2024):

@kth8
looks like you have similar problem to @Tamila-2017
from your log

[   24.636266] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0

@mnccouk commented on GitHub (Sep 25, 2024): @kth8 looks like you have similar problem to @Tamila-2017 from your log ``` [ 24.636266] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0 ```

GiteaMirror commented

2026-04-12 11:19:22 -05:00

@kth8 commented on GitHub (Sep 25, 2024):

@kth8 looks like you have similar problem to @Tamila-2017 from your log
[   24.636266] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0

Do you know what that means? Is there something wrong with my motherboard or graphics card? Do I need to change a BIOS setting or set a kernel parameter or use a different kernel or something else?

@kth8 commented on GitHub (Sep 25, 2024): > @kth8 looks like you have similar problem to @Tamila-2017 from your log > > ``` > [ 24.636266] kfd kfd: amdgpu: skipped device 1002:67df, PCI rejects atomics 730<0 > ``` Do you know what that means? Is there something wrong with my motherboard or graphics card? Do I need to change a BIOS setting or set a kernel parameter or use a different kernel or something else?

GiteaMirror commented

@mnccouk commented on GitHub (Sep 25, 2024):

I believe it's to do with PCI specification - see chat from earlier - https://github.com/ollama/ollama/issues/2453#issuecomment-2369372398

@mnccouk commented on GitHub (Sep 25, 2024): I believe it's to do with PCI specification - see chat from earlier - https://github.com/ollama/ollama/issues/2453#issuecomment-2369372398

GiteaMirror commented

@kth8 commented on GitHub (Sep 25, 2024):

I believe it's to do with PCI specification - see chat from earlier - #2453 (comment)

If this is a hardware issue then I guess there is no way to solve this?

@kth8 commented on GitHub (Sep 25, 2024): > I believe it's to do with PCI specification - see chat from earlier - [#2453 (comment)](https://github.com/ollama/ollama/issues/2453#issuecomment-2369372398) If this is a hardware issue then I guess there is no way to solve this?

GiteaMirror commented

@mnccouk commented on GitHub (Sep 25, 2024):

Only by upgrading motherboard and CPU, I think. (I'm no expert, only deduced this from what I've read, so, i'm open to other thoughts.)

Your logs list your system with detected chipset - https://www.intel.com/content/www/us/en/products/sku/66416/intel-c216-chipset/specifications.html
https://www.intel.com/content/www/us/en/products/sku/65693/intel-core-i33220-processor-3m-cache-3-30-ghz/specifications.html

Which both are PCIe v2.0,

@mnccouk commented on GitHub (Sep 25, 2024): Only by upgrading motherboard and CPU, I think. (I'm no expert, only deduced this from what I've read, so, i'm open to other thoughts.) Your logs list your system with detected chipset - https://www.intel.com/content/www/us/en/products/sku/66416/intel-c216-chipset/specifications.html https://www.intel.com/content/www/us/en/products/sku/65693/intel-core-i33220-processor-3m-cache-3-30-ghz/specifications.html Which both are PCIe v2.0,

GiteaMirror commented

2026-04-12 11:19:24 -05:00

@Tamila-2017 commented on GitHub (Sep 26, 2024):

Guys, let me express my humble opinion, perhaps erroneous.
Yes, outdated hardware can cause and failure ROCm to at it work.
But it should not affect ROCm assembly, i.e. ordinary compilation, imho.

@Tamila-2017 commented on GitHub (Sep 26, 2024): Guys, let me express my humble opinion, perhaps erroneous. Yes, outdated hardware can cause and failure ROCm to at it work. But it should not affect ROCm assembly, i.e. ordinary compilation, imho.

GiteaMirror commented

@Tamila-2017 commented on GitHub (Sep 27, 2024):

About Atomics and Motherboards: https://github.com/ROCm/ROCm/issues/1146#issuecomment-758624560

Very sad information.... It turns out that ROCm is a capricious and unreliable project :-(

@Tamila-2017 commented on GitHub (Sep 27, 2024): About Atomics and Motherboards: https://github.com/ROCm/ROCm/issues/1146#issuecomment-758624560 Very sad information.... It turns out that ROCm is a capricious and unreliable project :-(

GiteaMirror commented

2026-04-12 11:19:24 -05:00

@Tamila-2017 commented on GitHub (Sep 29, 2024):

My Test NVIDIA Graphics Cards Performance:

Of course, modern NVIDIA graphics cards work fast.
But modern AMD cards also have good performance

AMD's problem lies in a completely different, non-caring attitude: the disgusting support for video cards.
As a result, the ROCm project is in a terribly moody state. It requires special motherboards with an undocumented Atomics option for the PCI bus, which their manufacturers usually do not report in their specifications. Therefore, it is impossible to guess in advance whether any motherboard will work or not.

The Ollama project also doesn't give a damn about old AMD graphics cards, paying attention only to new models.

As a result, I completely wasted a lot of time trying to get ROCm to work on the RX580, but I didn't get any of it.

I am completely disappointed in AMD and ROCm for it.
Therefore, I stop any attempts to work with the terrible AMD platform and completely switch to NVIDIA, which does not have these problems and everything is configured in a few minutes and works stably.

Good luck to you, guys! 💓

@Tamila-2017 commented on GitHub (Sep 29, 2024): **My Test NVIDIA Graphics Cards Performance:** ![RTX-2080-TI](https://github.com/user-attachments/assets/ec6650c8-9ce6-4f6c-9db6-ae5a27b99e3e) ![RTX-3080](https://github.com/user-attachments/assets/fba34ede-7774-4fd6-8fa4-f37a10af5718) Of course, modern NVIDIA graphics cards work fast. But modern AMD cards also have good performance AMD's problem lies in a completely different, non-caring attitude: the disgusting support for video cards. As a result, the ROCm project is in a terribly moody state. It requires special motherboards with an undocumented Atomics option for the PCI bus, which their manufacturers usually do not report in their specifications. Therefore, it is impossible to guess in advance whether any motherboard will work or not. The Ollama project also doesn't give a damn about old AMD graphics cards, paying attention only to new models. As a result, I completely wasted a lot of time trying to get ROCm to work on the RX580, but I didn't get any of it. I am completely disappointed in AMD and ROCm for it. Therefore, I stop any attempts to work with the terrible AMD platform and completely switch to NVIDIA, which does not have these problems and everything is configured in a few minutes and works stably. Good luck to you, guys! 💓

GiteaMirror commented

2026-04-12 11:19:24 -05:00

@T-Shilov commented on GitHub (Oct 8, 2024):

Hello everyone,

I have read this topic carefully, and I want to buy a PRIME X370-PRO motherboard.
Do you think it will be able to work with two AMD RX580 graphics cards?

@T-Shilov commented on GitHub (Oct 8, 2024): Hello everyone, I have read this [topic](https://github.com/ROCm/ROCm/issues/1146) carefully, and I want to buy a [PRIME X370-PRO](https://www.asus.com/supportonly/prime%20x370-pro/helpdesk_cpu/) motherboard. Do you think it will be able to work with **two** AMD RX580 graphics cards?

GiteaMirror commented

2026-04-12 11:19:25 -05:00

@Darin755 commented on GitHub (Oct 8, 2024):

Hello everyone,

I have read this topic carefully, and I want to buy a PRIME X370-PRO motherboard. Do you think it will be able to work with two AMD RX580 graphics cards?

It should work the big question is what kind of performance you will get. That will depend heavily on the CPU and motherboard. For optimal performance you want two x16 slots but the board you have is either a single x16 or dual x8. That might be fine for your needs but it is something you should keep in mind.

If you want something that can handle two cards you want a server board with a server CPU which can be pretty expensive. The other option would be to just build two systems. Either way this issue is not really the place for this so I would recommend that you open up a new discussion.

@Darin755 commented on GitHub (Oct 8, 2024): > Hello everyone, > > I have read this [topic](https://github.com/ROCm/ROCm/issues/1146) carefully, and I want to buy a [PRIME X370-PRO](https://www.asus.com/supportonly/prime%20x370-pro/helpdesk_cpu/) motherboard. Do you think it will be able to work with **two** AMD RX580 graphics cards? It should work the big question is what kind of performance you will get. That will depend heavily on the CPU and motherboard. For optimal performance you want two x16 slots but the board you have is either a single x16 or dual x8. That might be fine for your needs but it is something you should keep in mind. If you want something that can handle two cards you want a server board with a server CPU which can be pretty expensive. The other option would be to just build two systems. Either way this issue is not really the place for this so I would recommend that you open up a new discussion.

GiteaMirror commented

2026-04-12 11:19:25 -05:00

@T-Shilov commented on GitHub (Oct 8, 2024):

@Darin755,

For optimal performance you want two x16 slots but the board you have is either a single x16 or dual x8.

Thank you for your important observation, thank you!
So, the PRIME X370-PRO for the two RX580s is not suitable :-(
Will need another motherboard, and I would appreciate it if you could tell me its model.

That will depend heavily on the CPU

I'm sorry, but is AI performance dependent on the CPU?
After all, graphics cards are used here, which provide the main performance of AI.
Take a look at this, please

Regarding the question of the other thread. The use of RX580 is closely related to the type of motherboard, and it was already discussed above. So I also have the right to discuss here the choice of motherboard for the RX580, Isn't that right?
However, I took your advice into account and asked my question here as well.

@T-Shilov commented on GitHub (Oct 8, 2024): @Darin755, > For optimal performance you want two x16 slots but the board you have is either a single x16 or dual x8. ![2 x PCI Express 3 0](https://github.com/user-attachments/assets/75769e3a-d371-4688-b742-da2d79262cf2) Thank you for your important observation, thank you! So, the PRIME X370-PRO for the two RX580s is not suitable :-( Will need another motherboard, and I would appreciate it if you could tell me its model. > That will depend heavily on the CPU I'm sorry, but is AI performance dependent on the CPU? After all, graphics cards are used here, which provide the main performance of AI. Take a look at [this]( https://github.com/ollama/ollama/issues/2453#issuecomment-2375210546), please Regarding the question of the other thread. The use of RX580 is closely related to the type of motherboard, and it was already discussed [above](https://github.com/ollama/ollama/issues/2453#issuecomment-2371070242). So I also have the right to discuss here the choice of motherboard for the RX580, Isn't that right? However, I took your advice into account and asked my question [here](https://github.com/ROCm/ROCm/issues/1146#issuecomment-2400184705) as well.

GiteaMirror commented

2026-04-12 11:19:25 -05:00

@T-Shilov commented on GitHub (Oct 10, 2024):

mnccouk,
I used Ubuntu 24.04 and your handy script: https://github.com/ollama/ollama/issues/2453#issuecomment-2366982275
However, when executing this script, several errors occurred and the script stopped:

<skip>
update-alternatives: using /usr/lib/shim/shimx64.efi.signed.latest to provide /usr/lib/shim/shimx64.efi.signed (shimx64.e
fi.signed) in auto mode
Generating a new Secure Boot signing key:
Can't load /var/lib/shim-signed/mok/.rnd into RNG
40474B4D84740000:error:12000079:random number generator:RAND_load_file:Cannot open file:../crypto/rand/randfile.c:106:Fil
ename=/var/lib/shim-signed/mok/.rnd
<skip>

EFI variables are not supported on this system
/sys/firmware/efi/efivars not found, aborting.
Setting up libgprofng0:amd64 (2.42-4ubuntu2) ...
<skip>

Setting up dkms (3.0.11-1ubuntu13) ...
Setting up amdgpu-dkms (1:6.2.4.50701-1664922.22.04) ...
Loading new amdgpu-6.2.4-1664922.22.04 DKMS files...
Building for 6.8.0-45-generic
Building for architecture x86_64
Building initial module for 6.8.0-45-generic
Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64)
Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information.
dpkg: error processing package amdgpu-dkms (--configure):
 installed amdgpu-dkms package post-installation script subprocess returned error exit status 10
Setting up g++-x86-64-linux-gnu (4:13.2.0-7ubuntu1) ...
Setting up g++-13 (13.2.0-23ubuntu4) ...
Setting up g++ (4:13.2.0-7ubuntu1) ...
update-alternatives: using /usr/bin/g++ to provide /usr/bin/c++ (c++) in auto mode
Setting up build-essential (12.10ubuntu1) ...
Processing triggers for libc-bin (2.39-0ubuntu8.3) ...
Processing triggers for man-db (2.12.0-4build2) ...
Processing triggers for install-info (7.1-3build2) ...
Errors were encountered while processing:
 amdgpu-dkms
E: Sub-process /usr/bin/dpkg returned an error code (1)

Then I used an older Ubuntu 22.04.3 LTS, but other errors occurred and the script stopped again:

<skip>
Loading new amdgpu-6.2.4-1664922.22.04 DKMS files...
Building for 6.8.0-45-generic
Building for architecture x86_64
Building initial module for 6.8.0-45-generic
configure: error: cannot detect CFLAGS...
Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64)
Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information.
dpkg: error processing package amdgpu-dkms (--configure):
 installed amdgpu-dkms package post-installation script subprocess returned error exit status 10
Setting up g++ (4:11.2.0-1ubuntu1) ...
update-alternatives: using /usr/bin/g++ to provide /usr/bin/c++ (c++) in auto mode
Setting up build-essential (12.9ubuntu3) ...
Processing triggers for libc-bin (2.35-0ubuntu3.1) ...
Processing triggers for man-db (2.10.2-1) ...
Processing triggers for install-info (6.8-4build1) ...
Errors were encountered while processing:
 amdgpu-dkms
E: Sub-process /usr/bin/dpkg returned an error code (1)

Can you please advise me on how this unfortunate situation can be remedied?

@T-Shilov commented on GitHub (Oct 10, 2024): mnccouk, I used Ubuntu 24.04 and your handy script: https://github.com/ollama/ollama/issues/2453#issuecomment-2366982275 However, when executing this script, several errors occurred and the script stopped: ``` <skip> update-alternatives: using /usr/lib/shim/shimx64.efi.signed.latest to provide /usr/lib/shim/shimx64.efi.signed (shimx64.e fi.signed) in auto mode Generating a new Secure Boot signing key: Can't load /var/lib/shim-signed/mok/.rnd into RNG 40474B4D84740000:error:12000079:random number generator:RAND_load_file:Cannot open file:../crypto/rand/randfile.c:106:Fil ename=/var/lib/shim-signed/mok/.rnd <skip> EFI variables are not supported on this system /sys/firmware/efi/efivars not found, aborting. Setting up libgprofng0:amd64 (2.42-4ubuntu2) ... <skip> Setting up dkms (3.0.11-1ubuntu13) ... Setting up amdgpu-dkms (1:6.2.4.50701-1664922.22.04) ... Loading new amdgpu-6.2.4-1664922.22.04 DKMS files... Building for 6.8.0-45-generic Building for architecture x86_64 Building initial module for 6.8.0-45-generic Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64) Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information. dpkg: error processing package amdgpu-dkms (--configure): installed amdgpu-dkms package post-installation script subprocess returned error exit status 10 Setting up g++-x86-64-linux-gnu (4:13.2.0-7ubuntu1) ... Setting up g++-13 (13.2.0-23ubuntu4) ... Setting up g++ (4:13.2.0-7ubuntu1) ... update-alternatives: using /usr/bin/g++ to provide /usr/bin/c++ (c++) in auto mode Setting up build-essential (12.10ubuntu1) ... Processing triggers for libc-bin (2.39-0ubuntu8.3) ... Processing triggers for man-db (2.12.0-4build2) ... Processing triggers for install-info (7.1-3build2) ... Errors were encountered while processing: amdgpu-dkms E: Sub-process /usr/bin/dpkg returned an error code (1) ``` Then I used an older Ubuntu 22.04.3 LTS, but other errors occurred and the script stopped again: ``` <skip> Loading new amdgpu-6.2.4-1664922.22.04 DKMS files... Building for 6.8.0-45-generic Building for architecture x86_64 Building initial module for 6.8.0-45-generic configure: error: cannot detect CFLAGS... Error! Bad return status for module build on kernel: 6.8.0-45-generic (x86_64) Consult /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log for more information. dpkg: error processing package amdgpu-dkms (--configure): installed amdgpu-dkms package post-installation script subprocess returned error exit status 10 Setting up g++ (4:11.2.0-1ubuntu1) ... update-alternatives: using /usr/bin/g++ to provide /usr/bin/c++ (c++) in auto mode Setting up build-essential (12.9ubuntu3) ... Processing triggers for libc-bin (2.35-0ubuntu3.1) ... Processing triggers for man-db (2.10.2-1) ... Processing triggers for install-info (6.8-4build1) ... Errors were encountered while processing: amdgpu-dkms E: Sub-process /usr/bin/dpkg returned an error code (1) ``` Can you please advise me on how this unfortunate situation can be remedied?

GiteaMirror commented

@T-Shilov commented on GitHub (Oct 11, 2024):

This topic was created on February 12. It's been 310 days already!
It contains a lot of questions and various tips on using AMD RX580.
For example, there was created a handy docker for using Ollama.

However, I studied this topic carefully, and realized that the main problem here is a failed installation of ROCm 5.7.1.
After that I tried using the most promising tips for installing ROCm 5.7.1, but unfortunately none of them work.

Well, can someone please courage the trouble to summarize and create a how to for a RELIABLE installation of Rock 5.7.1? :-)

@T-Shilov commented on GitHub (Oct 11, 2024): This topic was created on February 12. It's been 310 days already! It contains a lot of questions and various tips on using AMD RX580. For example, there was created a handy docker for using Ollama. However, I studied this topic carefully, and realized that the main problem here is a failed installation of ROCm 5.7.1. After that I tried using the most promising tips for installing ROCm 5.7.1, but unfortunately none of them work. Well, can someone please courage the trouble to summarize and create a how to for a RELIABLE installation of Rock 5.7.1? :-)

GiteaMirror commented

@mnccouk commented on GitHub (Oct 11, 2024):

Regarding your error:

Check inside your build log -
/var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log

See if that gives you some extra information on why the build failed.

I noticed that you are using 24.04 of ubuntu, ROCm 5.7.1 drivers are
relatively old so there's a chance that something might not be compatible
between the two.

On Fri, 11 Oct 2024 at 15:34, T-Shilov @.***> wrote:

This topic was created on February 12. It's been 310 days already!
It contains a lot of questions and various tips on using AMD RX580. For
example, there was created a handy docker for using Ollama.

However, I studied this topic carefully, and realized that the main
problem here is a failed installation of ROCm 5.7.1.
After that I tried using the most promising tips for installing ROCm
5.7.1, but unfortunately none of them work.

Well, can someone please courage the trouble to summarize and create a how
to for a RELIABLE installation of Rock 5.7.1? :-)

—
Reply to this email directly, view it on GitHub
https://github.com/ollama/ollama/issues/2453#issuecomment-2407547842,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ABOLJO4NJJQRC6BVXJDUMQDZ27OX7AVCNFSM6AAAAABDD3P55SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBXGU2DOOBUGI
.
You are receiving this because you were mentioned.Message ID:
@.***>

@mnccouk commented on GitHub (Oct 11, 2024): Regarding your error: Check inside your build log - /var/lib/dkms/amdgpu/6.2.4-1664922.22.04/build/make.log See if that gives you some extra information on why the build failed. I noticed that you are using 24.04 of ubuntu, ROCm 5.7.1 drivers are relatively old so there's a chance that something might not be compatible between the two. On Fri, 11 Oct 2024 at 15:34, T-Shilov ***@***.***> wrote: > This topic was created on February 12. It's been 310 days already! > It contains a lot of questions and various tips on using AMD RX580. For > example, there was created a handy docker for using Ollama. > > However, I studied this topic carefully, and realized that the main > problem here is a failed installation of ROCm 5.7.1. > After that I tried using the most promising tips for installing ROCm > 5.7.1, but unfortunately none of them work. > > Well, can someone please courage the trouble to summarize and create a how > to for a RELIABLE installation of Rock 5.7.1? :-) > > — > Reply to this email directly, view it on GitHub > <https://github.com/ollama/ollama/issues/2453#issuecomment-2407547842>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ABOLJO4NJJQRC6BVXJDUMQDZ27OX7AVCNFSM6AAAAABDD3P55SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBXGU2DOOBUGI> > . > You are receiving this because you were mentioned.Message ID: > ***@***.***> >

GiteaMirror commented

@T-Shilov commented on GitHub (Oct 11, 2024):

mnccouk, thank you the quick response.
But since my attempts were unsuccessful, during this time I have already to variant to the option using Ubuntu 20.04.
A little later I will tell you what happened with it.

@T-Shilov commented on GitHub (Oct 11, 2024): mnccouk, thank you the quick response. But since my attempts were unsuccessful, during this time I have already to variant to the option using Ubuntu 20.04. A little later I will tell you what happened with it.

GiteaMirror commented