github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #13535] Can't override OLLAMA_LIBRARY_PATH [Ubuntu 24.04.3 LTS | ollama 0.13.5 systemd | AMD ROCm] #55428

New Issue
Closed
opened 2026-04-29 09:10:12 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-29 09:10:12 -05:00
Owner
Copy Link

Originally created by @osirisOfGit on GitHub (Dec 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13535

What is the issue?

What it says on the tin - followed instructions to install AMD ROCm native on my system, all lights green on that front -

post-installation steps results 
❯ rocm-smi


WARNING: AMD GPU device(s) is/are in a low-power state. Check power control/runtime_status

========================================= ROCm System Management Interface =========================================
=================================================== Concise Info ===================================================
Device  Node  IDs              Temp    Power   Partitions          SCLK   MCLK     Fan  Perf  PwrCap  VRAM%  GPU%  
              (DID,     GUID)  (Edge)  (Avg)   (Mem, Compute, ID)                                                  
====================================================================================================================
0       1     0x7550,   45211  37.0°C  15.0W   N/A, N/A, 0         16Mhz  456Mhz   0%   auto  330.0W  4%     2%    
1       2     0x13c0,   21135  37.0°C  0.015W  N/A, N/A, 0         N/A    2800Mhz  0%   auto  N/A     0%     0%    
====================================================================================================================
=============================================== End of ROCm SMI Log ================================================

~ 
❯ rocminfo | grep -i "Marketing Name:"
  Marketing Name:          AMD Ryzen 7 9800X3D 8-Core Processor
  Marketing Name:          AMD Radeon RX 9070 XT              
  Marketing Name:          AMD Radeon Graphics

and ollama is added to the requisite groups

❯ id ollama
uid=997(ollama) gid=984(ollama) groups=984(ollama),44(video),992(render)

but all my driver stuff is installed to /opt/rocm/

ls -lart /opt/rocm/ 
❯ lt /opt/rocm/
total 40
lrwxrwxrwx  1 root root    10 Nov 20 14:52 llvm -> ./lib/llvm
lrwxrwxrwx  1 root root    32 Nov 20 14:53 amdgcn -> lib/llvm/lib/clang/20/lib/amdgcn
drwxr-xr-x  8 root root  4096 Dec 17 20:24 .
drwxr-xr-x 25 root root  4096 Dec 17 20:24 share
drwxr-xr-x 13 root root  4096 Dec 17 20:24 libexec
drwxr-xr-x 49 root root  4096 Dec 17 20:24 include
drwxr-xr-x  5 root root  4096 Dec 17 20:24 ..
drwxr-xr-x  3 root root  4096 Dec 17 20:24 bin
drwxr-xr-x 15 root root 12288 Dec 17 20:26 lib
drwxr-xr-x  2 root root  4096 Dec 20 00:09 .info

not /usr/local/lib/ollama/rocm as Ollama seems to default, causing level=INFO source=runner.go:464 msg="failure during GPU discovery" OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" extra_envs=map[] error="runner crashed".

I've tried literally every way of overwriting the OLLAMA_LIBRARY_PATH= (system) environment variable that i can think of, but the result is always the same - the discovery action always seems to use the defaults and doesn't allow me to point it anywhere else

❯ sudo systemctl daemon-reload

~ 
❯ sudo systemctl restart ollama

~ 
❯ sudo systemctl status ollama
● ollama.service - Ollama Service
     Loaded: loaded (/etc/systemd/system/ollama.service; enabled; preset: enabled)
    Drop-In: /etc/systemd/system/ollama.service.d
             └─override.conf
     Active: active (running) since Sat 2025-12-20 01:57:18 CST; 1s ago
   Main PID: 31823 (ollama)
      Tasks: 10 (limit: 73894)
     Memory: 10.4M (peak: 360.4M)
        CPU: 586ms
     CGroup: /system.slice/ollama.service
             └─31823 /usr/local/bin/ollama serve

Dec 20 01:57:18 Anubis ollama[31823]: "/usr/local/lib/ollama/rocm/rocblas/library/TensileLibrary_lazy_gfx1200.dat"
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=runner.go:464 msg="failure during GPU discovery" OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" extra_envs=map[] error="runner crashed"
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=TRACE source=runner.go:467 msg="runner enumerated devices" OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" devices=[]
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=DEBUG source=runner.go:437 msg="bootstrap discovery took" duration=616.404417ms OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" extra_envs=map[]
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=runner.go:106 msg="experimental Vulkan support disabled.  To enable, set OLLAMA_VULKAN=1"
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=DEBUG source=runner.go:124 msg="evaluating which, if any, devices to filter out" initial_count=0
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=TRACE source=runner.go:174 msg="supported GPU library combinations before filtering" supported=map[]
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=DEBUG source=runner.go:40 msg="GPU bootstrap discovery took" duration=644.461991ms
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=types.go:60 msg="inference compute" id=cpu library=cpu compute="" name=cpu description=cpu libdirs=ollama driver="" pci_id="" type="" total="60.4 GiB" available="55.5 GiB"
Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=routes.go:1648 msg="entering low vram mode" "total vram"="0 B" threshold="20.0 GiB"

Below image is just proof that i've tried the most sensible way and the rest of the environment variables work, just not this one:

Ultrawide image screenshot

(Ignore the S[Service] in this, that was a fat finger as i was taking the screenshot, i was actually using [Service]

Image

Thankfully, Vulkan does appear to work:

Dec 20 02:11:34 Anubis ollama[33593]: time=2025-12-20T02:11:34.424-06:00 level=INFO source=types.go:42 msg="inference compute" id=00000000-7c00-0000-0000-000000000000 filter_id="" library=Vulkan compute=0.0 name=Vulkan1 description="AMD Radeon Graphics (RADV RAPHAEL_MENDOCINO)" libdirs=ollama,vulkan driver=0.0 pci_id=0000:7c:00.0 type=iGPU total="32.2 GiB" available="32.2 GiB"
Dec 20 02:11:34 Anubis ollama[33593]: time=2025-12-20T02:11:34.424-06:00 level=INFO source=types.go:42 msg="inference compute" id=00000000-0300-0000-0000-000000000000 filter_id="" library=Vulkan compute=0.0 name=Vulkan0 description="AMD Radeon RX 9070 XT (RADV GFX1201)" libdirs=ollama,vulkan driver=0.0 pci_id=0000:03:00.0 type=discrete total="15.9 GiB" available="15.2 GiB"

but i figured i'd report it anyway.

Let me know if there's any more info i can provide or test out. Thanks in advance!

Full journal output, reversed (new -> old), includes vulkan
ollama.log

Relevant log output

OS

Linux

GPU

AMD

CPU

AMD

Ollama version

0.13.5

Originally created by @osirisOfGit on GitHub (Dec 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13535 ### What is the issue? What it says on the tin - followed instructions [to install AMD ROCm native on my system](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/install-methods/package-manager/package-manager-ubuntu.html), all lights green on that front - <details> <summary> post-installation steps results </summary> ```bash ❯ rocm-smi WARNING: AMD GPU device(s) is/are in a low-power state. Check power control/runtime_status ========================================= ROCm System Management Interface ========================================= =================================================== Concise Info =================================================== Device Node IDs Temp Power Partitions SCLK MCLK Fan Perf PwrCap VRAM% GPU% (DID, GUID) (Edge) (Avg) (Mem, Compute, ID) ==================================================================================================================== 0 1 0x7550, 45211 37.0°C 15.0W N/A, N/A, 0 16Mhz 456Mhz 0% auto 330.0W 4% 2% 1 2 0x13c0, 21135 37.0°C 0.015W N/A, N/A, 0 N/A 2800Mhz 0% auto N/A 0% 0% ==================================================================================================================== =============================================== End of ROCm SMI Log ================================================ ~ ❯ rocminfo | grep -i "Marketing Name:" Marketing Name: AMD Ryzen 7 9800X3D 8-Core Processor Marketing Name: AMD Radeon RX 9070 XT Marketing Name: AMD Radeon Graphics ``` </details> and ollama is added to the requisite groups ```bash ❯ id ollama uid=997(ollama) gid=984(ollama) groups=984(ollama),44(video),992(render) ``` but all my driver stuff is installed to `/opt/rocm/` <details> <summary> ls -lart /opt/rocm/ </summary> ```bash ❯ lt /opt/rocm/ total 40 lrwxrwxrwx 1 root root 10 Nov 20 14:52 llvm -> ./lib/llvm lrwxrwxrwx 1 root root 32 Nov 20 14:53 amdgcn -> lib/llvm/lib/clang/20/lib/amdgcn drwxr-xr-x 8 root root 4096 Dec 17 20:24 . drwxr-xr-x 25 root root 4096 Dec 17 20:24 share drwxr-xr-x 13 root root 4096 Dec 17 20:24 libexec drwxr-xr-x 49 root root 4096 Dec 17 20:24 include drwxr-xr-x 5 root root 4096 Dec 17 20:24 .. drwxr-xr-x 3 root root 4096 Dec 17 20:24 bin drwxr-xr-x 15 root root 12288 Dec 17 20:26 lib drwxr-xr-x 2 root root 4096 Dec 20 00:09 .info ``` </details> not `/usr/local/lib/ollama/rocm` as Ollama seems to default, causing `level=INFO source=runner.go:464 msg="failure during GPU discovery" OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" extra_envs=map[] error="runner crashed"`. I've tried _literally_ every way of overwriting the `OLLAMA_LIBRARY_PATH=` (system) environment variable that i can think of, but the result is always the same - the discovery action always seems to use the defaults and doesn't allow me to point it anywhere else ```bash ❯ sudo systemctl daemon-reload ~ ❯ sudo systemctl restart ollama ~ ❯ sudo systemctl status ollama ● ollama.service - Ollama Service Loaded: loaded (/etc/systemd/system/ollama.service; enabled; preset: enabled) Drop-In: /etc/systemd/system/ollama.service.d └─override.conf Active: active (running) since Sat 2025-12-20 01:57:18 CST; 1s ago Main PID: 31823 (ollama) Tasks: 10 (limit: 73894) Memory: 10.4M (peak: 360.4M) CPU: 586ms CGroup: /system.slice/ollama.service └─31823 /usr/local/bin/ollama serve Dec 20 01:57:18 Anubis ollama[31823]: "/usr/local/lib/ollama/rocm/rocblas/library/TensileLibrary_lazy_gfx1200.dat" Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=runner.go:464 msg="failure during GPU discovery" OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" extra_envs=map[] error="runner crashed" Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=TRACE source=runner.go:467 msg="runner enumerated devices" OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" devices=[] Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=DEBUG source=runner.go:437 msg="bootstrap discovery took" duration=616.404417ms OLLAMA_LIBRARY_PATH="[/usr/local/lib/ollama /usr/local/lib/ollama/rocm]" extra_envs=map[] Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=runner.go:106 msg="experimental Vulkan support disabled. To enable, set OLLAMA_VULKAN=1" Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=DEBUG source=runner.go:124 msg="evaluating which, if any, devices to filter out" initial_count=0 Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=TRACE source=runner.go:174 msg="supported GPU library combinations before filtering" supported=map[] Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=DEBUG source=runner.go:40 msg="GPU bootstrap discovery took" duration=644.461991ms Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=types.go:60 msg="inference compute" id=cpu library=cpu compute="" name=cpu description=cpu libdirs=ollama driver="" pci_id="" type="" total="60.4 GiB" available="55.5 GiB" Dec 20 01:57:18 Anubis ollama[31823]: time=2025-12-20T01:57:18.826-06:00 level=INFO source=routes.go:1648 msg="entering low vram mode" "total vram"="0 B" threshold="20.0 GiB" ``` Below image is just proof that i've tried the most sensible way and the rest of the environment variables work, just not this one: <details> <summary> Ultrawide image screenshot </summary> (Ignore the `S[Service]` in this, that was a fat finger as i was taking the screenshot, i was actually using `[Service]` <img width="2625" height="1208" alt="Image" src="https://github.com/user-attachments/assets/a029176e-c17a-446e-a6a9-d5017b790d5b" /> </details> Thankfully, Vulkan does appear to work: ```bash Dec 20 02:11:34 Anubis ollama[33593]: time=2025-12-20T02:11:34.424-06:00 level=INFO source=types.go:42 msg="inference compute" id=00000000-7c00-0000-0000-000000000000 filter_id="" library=Vulkan compute=0.0 name=Vulkan1 description="AMD Radeon Graphics (RADV RAPHAEL_MENDOCINO)" libdirs=ollama,vulkan driver=0.0 pci_id=0000:7c:00.0 type=iGPU total="32.2 GiB" available="32.2 GiB" Dec 20 02:11:34 Anubis ollama[33593]: time=2025-12-20T02:11:34.424-06:00 level=INFO source=types.go:42 msg="inference compute" id=00000000-0300-0000-0000-000000000000 filter_id="" library=Vulkan compute=0.0 name=Vulkan0 description="AMD Radeon RX 9070 XT (RADV GFX1201)" libdirs=ollama,vulkan driver=0.0 pci_id=0000:03:00.0 type=discrete total="15.9 GiB" available="15.2 GiB" ``` but i figured i'd report it anyway. Let me know if there's any more info i can provide or test out. Thanks in advance! Full journal output, reversed (new -> old), includes vulkan [ollama.log](https://github.com/user-attachments/files/24270557/ollama.log) ### Relevant log output ```shell ``` ### OS Linux ### GPU AMD ### CPU AMD ### Ollama version 0.13.5
GiteaMirror added the bug label 2026-04-29 09:10:12 -05:00
GiteaMirror commented 2026-04-29 09:10:15 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 20, 2025):

#12908

<!-- gh-comment-id:3677772817 --> @rick-github commented on GitHub (Dec 20, 2025): #12908
GiteaMirror commented 2026-04-29 09:10:15 -05:00
Author
Owner
Copy Link

@osirisOfGit commented on GitHub (Dec 20, 2025):

Ah, it turns out, it wasn't a problem with the library path - Ollama was just insisting that it use my iGPU instead of my discrete graphics card. I finally got it to only look at my Radeon discrete with:

### Editing /etc/systemd/system/ollama.service.d/override.conf
### Anything between here and the comment below will become the contents of the drop-in file

[Service]
# Tell GGML to use ROCm (not CUDA) and set the visible device.
Environment="GGML_USE_ROCM=1"
Environment="GGML_USE_CUDA=0"


# Ollama‑specific helpers
Environment="OLLAMA_ROCM=1"
Environment="OLLAMA_DEBUG=0"
Environment="OLLAMA_VULKAN=0"
Environment="ROCR_VISIBLE_DEVICES=0"
Environment="ROCM_VISIBLE_DEVICES=0"

specifically

Environment="ROCR_VISIBLE_DEVICES=0"
Environment="ROCM_VISIBLE_DEVICES=0"

(the appropriate device id) - it took both of these for it to finally work - disregard this bug!

<!-- gh-comment-id:3678052640 --> @osirisOfGit commented on GitHub (Dec 20, 2025): Ah, it turns out, it wasn't a problem with the library path - Ollama was just _insisting_ that it use my iGPU instead of my discrete graphics card. I finally got it to only look at my Radeon discrete with: ```ini ### Editing /etc/systemd/system/ollama.service.d/override.conf ### Anything between here and the comment below will become the contents of the drop-in file [Service] # Tell GGML to use ROCm (not CUDA) and set the visible device. Environment="GGML_USE_ROCM=1" Environment="GGML_USE_CUDA=0" # Ollama‑specific helpers Environment="OLLAMA_ROCM=1" Environment="OLLAMA_DEBUG=0" Environment="OLLAMA_VULKAN=0" Environment="ROCR_VISIBLE_DEVICES=0" Environment="ROCM_VISIBLE_DEVICES=0" ``` specifically ``` Environment="ROCR_VISIBLE_DEVICES=0" Environment="ROCM_VISIBLE_DEVICES=0" ``` (the appropriate device id) - it took both of these for it to finally work - disregard this bug!
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#55428
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?