github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-15 12:48:05 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #11271] Potential Incompatibility between Rocky Linux 9.6 and DaVinci Resolve 20 #7431

New Issue
Closed
opened 2026-04-12 19:30:47 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-12 19:30:47 -05:00
Owner
Copy Link

Originally created by @KevinTurnbull on GitHub (Jul 2, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11271

What is the issue?

I'm running an NVidia Quadro RTX5000. This is the 2018 edition with 16gb VRAM.
I'm running on a similar era Xeon Processor with 24gb RAM and 64gb of SSD SWAP.

When the system was running Linux Mint 22.04 I would get high TPS using Qwen3:14b. Installing DaVinci Resolve 20 was incompatible with Linux Mint so I migrated to Rocky Linux 9.6. I had to disable Wayland for compatibility with some screencasting software that I use.

I have performed: https://docs.rockylinux.org/desktop/display/installing_nvidia_gpu_drivers/

linai@linai:~$ nvidia-smi
Wed Jul  2 08:27:27 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 575.57.08              Driver Version: 575.57.08      CUDA Version: 12.9     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  Quadro RTX 5000                Off |   00000000:01:00.0 Off |                    0 |
| 33%   32C    P8             13W /  230W |     208MiB /  15360MiB |     12%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1928      G   /usr/libexec/Xorg                        63MiB |
|    0   N/A  N/A            2021      G   /usr/bin/gnome-shell                    138MiB |
+-----------------------------------------------------------------------------------------+

Running QWEN3:14b was very slow. Concurrent PMON will follow:

linai@linai:~$ ollama run qwen3:14b
>>> Hi
Thinking...
Okay, the user said "Hi". I need to respond appropriately^C

>>> /bye
linai@linai:~$ ollama ps
NAME         ID              SIZE     PROCESSOR    UNTIL              
qwen3:14b    bdbd181c33f2    12 GB    100% GPU     4 minutes from now

This is a readout of a nvidia-smi pmon -o T running for the minute surrounding the above commands.

linai@linai:~$ nvidia-smi pmon -o T
#Time         gpu         pid   type     sm    mem    enc    dec    jpg    ofa    command 
#HH:MM:SS     Idx           #    C/G      %      %      %      %      %      %    name 
 08:29:06       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:06       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:07       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:07       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:08       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:08       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:09       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:09       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:10       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:10       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:11       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:11       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:12       0       1928     G     74     27      -      -      -      -    Xorg           
 08:29:12       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:13       0       1928     G     18      5      -      -      -      -    Xorg           
 08:29:13       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:14       0       1928     G      2      0      -      -      -      -    Xorg           
 08:29:14       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:15       0       1928     G      2      1      -      -      -      -    Xorg           
 08:29:15       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:16       0       1928     G      2      1      -      -      -      -    Xorg           
 08:29:16       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:17       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:17       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:19       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:19       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:20       0       1928     G      6      0      -      -      -      -    Xorg           
 08:29:20       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:21       0       1928     G      3      0      -      -      -      -    Xorg           
 08:29:21       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:22       0       1928     G      9      2      -      -      -      -    Xorg           
 08:29:22       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:23       0       1928     G     13      4      -      -      -      -    Xorg           
 08:29:23       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:24       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:24       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:25       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:25       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:26       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:26       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:27       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:27       0       2021     G     10      4      -      -      -      -    gnome-shell    
 08:29:28       0       1928     G     13      5      -      -      -      -    Xorg           
 08:29:28       0       2021     G      3      1      -      -      -      -    gnome-shell    
#Time         gpu         pid   type     sm    mem    enc    dec    jpg    ofa    command 
#HH:MM:SS     Idx           #    C/G      %      %      %      %      %      %    name 
 08:29:29       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:29       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:30       0       1928     G      3      1      -      -      -      -    Xorg           
 08:29:30       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:31       0       1928     G      3      1      -      -      -      -    Xorg           
 08:29:31       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:32       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:32       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:33       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:33       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:34       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:34       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:35       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:35       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:36       0       1928     G      4      1      -      -      -      -    Xorg           
 08:29:36       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:37       0       1928     G      6      1      -      -      -      -    Xorg           
 08:29:37       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:38       0       1928     G      8      2      -      -      -      -    Xorg           
 08:29:38       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:39       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:39       0       2021     G      1      0      -      -      -      -    gnome-shell    
 08:29:40       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:40       0       2021     G     37     15      -      -      -      -    gnome-shell    
 08:29:41       0       1928     G      9      5      -      -      -      -    Xorg           
 08:29:41       0       2021     G     36     16      -      -      -      -    gnome-shell    
 08:29:43       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:43       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:44       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:44       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:45       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:45       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:46       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:46       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:47       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:47       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:48       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:48       0       2021     G      -      -      -      -      -      -    gnome-shell    
 08:29:49       0       1928     G      -      -      -      -      -      -    Xorg           
 08:29:49       0       2021     G      -      -      -      -      -      -    gnome-shell

In between the 7:50 logs and logs provided in "Relevant log output" I have confirmed drivers are installed, rebooted the system and have not booted up DaVinci Resolve 20.

Relevant log output

Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.061-04:00 level=INFO source=sched.go:788 msg="new model will fit in available VRAM in single GPU, loading" model=/usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e gpu=GPU-3a4bb711-3dff-4796-72b9-7ad73c0f74d7 parallel=2 available=15214182400 required="11.2 GiB"
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.334-04:00 level=INFO source=server.go:135 msg="system memory" total="23.1 GiB" free="20.7 GiB" free_swap="64.0 GiB"
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.335-04:00 level=INFO source=server.go:175 msg=offload library=cuda layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[14.2 GiB]" memory.gpu_overhead="0 B" memory.required.full="11.2 GiB" memory.required.partial="11.2 GiB" memory.required.kv="1.2 GiB" memory.required.allocations="[11.2 GiB]" memory.weights.total="8.2 GiB" memory.weights.repeating="7.6 GiB" memory.weights.nonrepeating="608.6 MiB" memory.graph.full="1.0 GiB" memory.graph.partial="1.0 GiB"
## [[...removed...]]]
Jul 02 07:50:01 linai ollama[1117]: print_info: general.name     = Qwen3 14B
## [[...removed...]]]
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.566-04:00 level=INFO source=server.go:438 msg="starting llama server" cmd="/usr/local/bin/ollama runner --model /usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e --ctx-size 8192 --batch-size 512 --n-gpu-layers 41 --threads 6 --parallel 2 --port 45893"
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.567-04:00 level=INFO source=sched.go:483 msg="loaded runners" count=1
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.567-04:00 level=INFO source=server.go:598 msg="waiting for llama runner to start responding"
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.567-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server not responding"
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.576-04:00 level=INFO source=runner.go:815 msg="starting go runner"
Jul 02 07:50:01 linai ollama[1117]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-sandybridge.so
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.581-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc)
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.581-04:00 level=INFO source=runner.go:874 msg="Server listening on 127.0.0.1:45893"
## [[...removed...]]]
Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.818-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server loading model"
## [[...removed...]]]
Jul 02 07:50:02 linai ollama[1117]: time=2025-07-02T07:50:02.177-04:00 level=WARN source=sched.go:687 msg="gpu VRAM usage didn't recover within timeout" seconds=5.19217117 runner.size="11.2 GiB" runner.vram="11.2 GiB" runner.parallel=2 runner.pid=22392 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e
Jul 02 07:50:02 linai ollama[1117]: llama_kv_cache_unified:        CPU KV buffer size =  1280.00 MiB
Jul 02 07:50:02 linai ollama[1117]: llama_kv_cache_unified: KV self size  = 1280.00 MiB, K (f16):  640.00 MiB, V (f16):  640.00 MiB
Jul 02 07:50:02 linai ollama[1117]: llama_context:        CPU compute buffer size =   696.01 MiB
Jul 02 07:50:02 linai ollama[1117]: llama_context: graph nodes  = 1526
Jul 02 07:50:02 linai ollama[1117]: llama_context: graph splits = 1
Jul 02 07:50:02 linai ollama[1117]: time=2025-07-02T07:50:02.319-04:00 level=INFO source=server.go:637 msg="llama runner started in 0.75 seconds"
Jul 02 07:50:02 linai ollama[1117]: [GIN] 2025/07/02 - 07:50:02 | 200 |  1.577629607s |       127.0.0.1 | POST     "/api/generate"
## [[ Rebooted System as part of Reinstalling NVIDIA drivers ]]
Jul 02 08:25:51 linai systemd[1]: Started Ollama Service.
Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.595-04:00 level=INFO source=routes.go:1235 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTT>
Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.597-04:00 level=INFO source=images.go:476 msg="total blobs: 5"
Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.597-04:00 level=INFO source=images.go:483 msg="total unused blobs removed: 0"
Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.598-04:00 level=INFO source=routes.go:1288 msg="Listening on 127.0.0.1:11434 (version 0.9.3)"
Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.598-04:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.833-04:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-3a4bb711-3dff-4796-72b9-7ad73c0f74d7 library=cuda variant=v12 compute=7.5 driver=12.9 name="Quadro RTX 50>
Jul 02 08:29:19 linai ollama[945]: [GIN] 2025/07/02 - 08:29:19 | 200 |     421.568µs |       127.0.0.1 | HEAD     "/"
Jul 02 08:29:19 linai ollama[945]: [GIN] 2025/07/02 - 08:29:19 | 200 |   50.302177ms |       127.0.0.1 | POST     "/api/show"
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.358-04:00 level=INFO source=sched.go:788 msg="new model will fit in available VRAM in single GPU, loading" model=/usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f>
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.537-04:00 level=INFO source=server.go:135 msg="system memory" total="23.1 GiB" free="21.1 GiB" free_swap="64.0 GiB"
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.541-04:00 level=INFO source=server.go:175 msg=offload library=cuda layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[14.2 GiB]" memory.gpu>
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.763-04:00 level=INFO source=server.go:438 msg="starting llama server" cmd="/usr/local/bin/ollama runner --model /usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6>
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.763-04:00 level=INFO source=sched.go:483 msg="loaded runners" count=1
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.763-04:00 level=INFO source=server.go:598 msg="waiting for llama runner to start responding"
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.764-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server not responding"
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.774-04:00 level=INFO source=runner.go:815 msg="starting go runner"
Jul 02 08:29:19 linai ollama[945]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-sandybridge.so
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.781-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc)
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.781-04:00 level=INFO source=runner.go:874 msg="Server listening on 127.0.0.1:45463"
Jul 02 08:29:20 linai ollama[945]: load_tensors: loading model tensors, this can take a while... (mmap = true)
Jul 02 08:29:20 linai ollama[945]: time=2025-07-02T08:29:20.014-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server loading model"
Jul 02 08:29:24 linai ollama[945]: load_tensors:   CPU_Mapped model buffer size =  8840.78 MiB
Jul 02 08:29:25 linai ollama[945]: time=2025-07-02T08:29:25.031-04:00 level=INFO source=server.go:637 msg="llama runner started in 5.27 seconds"
Jul 02 08:29:25 linai ollama[945]: [GIN] 2025/07/02 - 08:29:25 | 200 |  5.940366649s |       127.0.0.1 | POST     "/api/generate"
Jul 02 08:29:34 linai ollama[945]: [GIN] 2025/07/02 - 08:29:34 | 200 |  7.463693701s |       127.0.0.1 | POST     "/api/chat"
Jul 02 08:29:38 linai ollama[945]: [GIN] 2025/07/02 - 08:29:38 | 200 |      17.555µs |       127.0.0.1 | HEAD     "/"
Jul 02 08:29:38 linai ollama[945]: [GIN] 2025/07/02 - 08:29:38 | 200 |      69.888µs |       127.0.0.1 | GET      "/api/ps"
Jul 02 08:34:39 linai ollama[945]: time=2025-07-02T08:34:39.335-04:00 level=WARN source=sched.go:687 msg="gpu VRAM usage didn't recover within timeout" seconds=5.196673098 runner.size="11.2 GiB" runner.vram="11.2 GiB" runner.parallel=2 r>
Jul 02 08:34:39 linai ollama[945]: time=2025-07-02T08:34:39.586-04:00 level=WARN source=sched.go:687 msg="gpu VRAM usage didn't recover within timeout" seconds=5.447431288 runner.size="11.2 GiB" runner.vram="11.2 GiB" runner.parallel=2 r>

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.9.3

Originally created by @KevinTurnbull on GitHub (Jul 2, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11271 ### What is the issue? I'm running an NVidia Quadro RTX5000. This is the 2018 edition with 16gb VRAM. I'm running on a similar era Xeon Processor with 24gb RAM and 64gb of SSD SWAP. When the system was running Linux Mint 22.04 I would get high TPS using Qwen3:14b. Installing DaVinci Resolve 20 was incompatible with Linux Mint so I migrated to Rocky Linux 9.6. I had to disable Wayland for compatibility with some screencasting software that I use. I have performed: https://docs.rockylinux.org/desktop/display/installing_nvidia_gpu_drivers/ ```nvidia-smi linai@linai:~$ nvidia-smi Wed Jul 2 08:27:27 2025 +-----------------------------------------------------------------------------------------+ | NVIDIA-SMI 575.57.08 Driver Version: 575.57.08 CUDA Version: 12.9 | |-----------------------------------------+------------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+========================+======================| | 0 Quadro RTX 5000 Off | 00000000:01:00.0 Off | 0 | | 33% 32C P8 13W / 230W | 208MiB / 15360MiB | 12% Default | | | | N/A | +-----------------------------------------+------------------------+----------------------+ +-----------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=========================================================================================| | 0 N/A N/A 1928 G /usr/libexec/Xorg 63MiB | | 0 N/A N/A 2021 G /usr/bin/gnome-shell 138MiB | +-----------------------------------------------------------------------------------------+ ``` Running QWEN3:14b was very slow. Concurrent PMON will follow: ```bash linai@linai:~$ ollama run qwen3:14b >>> Hi Thinking... Okay, the user said "Hi". I need to respond appropriately^C >>> /bye linai@linai:~$ ollama ps NAME ID SIZE PROCESSOR UNTIL qwen3:14b bdbd181c33f2 12 GB 100% GPU 4 minutes from now ``` This is a readout of a `nvidia-smi pmon -o T` running for the minute surrounding the above commands. ```pmon linai@linai:~$ nvidia-smi pmon -o T #Time gpu pid type sm mem enc dec jpg ofa command #HH:MM:SS Idx # C/G % % % % % % name 08:29:06 0 1928 G - - - - - - Xorg 08:29:06 0 2021 G - - - - - - gnome-shell 08:29:07 0 1928 G - - - - - - Xorg 08:29:07 0 2021 G - - - - - - gnome-shell 08:29:08 0 1928 G - - - - - - Xorg 08:29:08 0 2021 G - - - - - - gnome-shell 08:29:09 0 1928 G - - - - - - Xorg 08:29:09 0 2021 G - - - - - - gnome-shell 08:29:10 0 1928 G - - - - - - Xorg 08:29:10 0 2021 G - - - - - - gnome-shell 08:29:11 0 1928 G - - - - - - Xorg 08:29:11 0 2021 G - - - - - - gnome-shell 08:29:12 0 1928 G 74 27 - - - - Xorg 08:29:12 0 2021 G - - - - - - gnome-shell 08:29:13 0 1928 G 18 5 - - - - Xorg 08:29:13 0 2021 G - - - - - - gnome-shell 08:29:14 0 1928 G 2 0 - - - - Xorg 08:29:14 0 2021 G - - - - - - gnome-shell 08:29:15 0 1928 G 2 1 - - - - Xorg 08:29:15 0 2021 G - - - - - - gnome-shell 08:29:16 0 1928 G 2 1 - - - - Xorg 08:29:16 0 2021 G - - - - - - gnome-shell 08:29:17 0 1928 G - - - - - - Xorg 08:29:17 0 2021 G - - - - - - gnome-shell 08:29:19 0 1928 G - - - - - - Xorg 08:29:19 0 2021 G - - - - - - gnome-shell 08:29:20 0 1928 G 6 0 - - - - Xorg 08:29:20 0 2021 G - - - - - - gnome-shell 08:29:21 0 1928 G 3 0 - - - - Xorg 08:29:21 0 2021 G - - - - - - gnome-shell 08:29:22 0 1928 G 9 2 - - - - Xorg 08:29:22 0 2021 G - - - - - - gnome-shell 08:29:23 0 1928 G 13 4 - - - - Xorg 08:29:23 0 2021 G - - - - - - gnome-shell 08:29:24 0 1928 G - - - - - - Xorg 08:29:24 0 2021 G - - - - - - gnome-shell 08:29:25 0 1928 G - - - - - - Xorg 08:29:25 0 2021 G - - - - - - gnome-shell 08:29:26 0 1928 G - - - - - - Xorg 08:29:26 0 2021 G - - - - - - gnome-shell 08:29:27 0 1928 G - - - - - - Xorg 08:29:27 0 2021 G 10 4 - - - - gnome-shell 08:29:28 0 1928 G 13 5 - - - - Xorg 08:29:28 0 2021 G 3 1 - - - - gnome-shell #Time gpu pid type sm mem enc dec jpg ofa command #HH:MM:SS Idx # C/G % % % % % % name 08:29:29 0 1928 G - - - - - - Xorg 08:29:29 0 2021 G - - - - - - gnome-shell 08:29:30 0 1928 G 3 1 - - - - Xorg 08:29:30 0 2021 G - - - - - - gnome-shell 08:29:31 0 1928 G 3 1 - - - - Xorg 08:29:31 0 2021 G - - - - - - gnome-shell 08:29:32 0 1928 G - - - - - - Xorg 08:29:32 0 2021 G - - - - - - gnome-shell 08:29:33 0 1928 G - - - - - - Xorg 08:29:33 0 2021 G - - - - - - gnome-shell 08:29:34 0 1928 G - - - - - - Xorg 08:29:34 0 2021 G - - - - - - gnome-shell 08:29:35 0 1928 G - - - - - - Xorg 08:29:35 0 2021 G - - - - - - gnome-shell 08:29:36 0 1928 G 4 1 - - - - Xorg 08:29:36 0 2021 G - - - - - - gnome-shell 08:29:37 0 1928 G 6 1 - - - - Xorg 08:29:37 0 2021 G - - - - - - gnome-shell 08:29:38 0 1928 G 8 2 - - - - Xorg 08:29:38 0 2021 G - - - - - - gnome-shell 08:29:39 0 1928 G - - - - - - Xorg 08:29:39 0 2021 G 1 0 - - - - gnome-shell 08:29:40 0 1928 G - - - - - - Xorg 08:29:40 0 2021 G 37 15 - - - - gnome-shell 08:29:41 0 1928 G 9 5 - - - - Xorg 08:29:41 0 2021 G 36 16 - - - - gnome-shell 08:29:43 0 1928 G - - - - - - Xorg 08:29:43 0 2021 G - - - - - - gnome-shell 08:29:44 0 1928 G - - - - - - Xorg 08:29:44 0 2021 G - - - - - - gnome-shell 08:29:45 0 1928 G - - - - - - Xorg 08:29:45 0 2021 G - - - - - - gnome-shell 08:29:46 0 1928 G - - - - - - Xorg 08:29:46 0 2021 G - - - - - - gnome-shell 08:29:47 0 1928 G - - - - - - Xorg 08:29:47 0 2021 G - - - - - - gnome-shell 08:29:48 0 1928 G - - - - - - Xorg 08:29:48 0 2021 G - - - - - - gnome-shell 08:29:49 0 1928 G - - - - - - Xorg 08:29:49 0 2021 G - - - - - - gnome-shell ``` In between the 7:50 logs and logs provided in "Relevant log output" I have confirmed drivers are installed, rebooted the system and have not booted up DaVinci Resolve 20. ### Relevant log output ```shell Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.061-04:00 level=INFO source=sched.go:788 msg="new model will fit in available VRAM in single GPU, loading" model=/usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e gpu=GPU-3a4bb711-3dff-4796-72b9-7ad73c0f74d7 parallel=2 available=15214182400 required="11.2 GiB" Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.334-04:00 level=INFO source=server.go:135 msg="system memory" total="23.1 GiB" free="20.7 GiB" free_swap="64.0 GiB" Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.335-04:00 level=INFO source=server.go:175 msg=offload library=cuda layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[14.2 GiB]" memory.gpu_overhead="0 B" memory.required.full="11.2 GiB" memory.required.partial="11.2 GiB" memory.required.kv="1.2 GiB" memory.required.allocations="[11.2 GiB]" memory.weights.total="8.2 GiB" memory.weights.repeating="7.6 GiB" memory.weights.nonrepeating="608.6 MiB" memory.graph.full="1.0 GiB" memory.graph.partial="1.0 GiB" ## [[...removed...]]] Jul 02 07:50:01 linai ollama[1117]: print_info: general.name = Qwen3 14B ## [[...removed...]]] Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.566-04:00 level=INFO source=server.go:438 msg="starting llama server" cmd="/usr/local/bin/ollama runner --model /usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e --ctx-size 8192 --batch-size 512 --n-gpu-layers 41 --threads 6 --parallel 2 --port 45893" Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.567-04:00 level=INFO source=sched.go:483 msg="loaded runners" count=1 Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.567-04:00 level=INFO source=server.go:598 msg="waiting for llama runner to start responding" Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.567-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server not responding" Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.576-04:00 level=INFO source=runner.go:815 msg="starting go runner" Jul 02 07:50:01 linai ollama[1117]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-sandybridge.so Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.581-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc) Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.581-04:00 level=INFO source=runner.go:874 msg="Server listening on 127.0.0.1:45893" ## [[...removed...]]] Jul 02 07:50:01 linai ollama[1117]: time=2025-07-02T07:50:01.818-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server loading model" ## [[...removed...]]] Jul 02 07:50:02 linai ollama[1117]: time=2025-07-02T07:50:02.177-04:00 level=WARN source=sched.go:687 msg="gpu VRAM usage didn't recover within timeout" seconds=5.19217117 runner.size="11.2 GiB" runner.vram="11.2 GiB" runner.parallel=2 runner.pid=22392 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e Jul 02 07:50:02 linai ollama[1117]: llama_kv_cache_unified: CPU KV buffer size = 1280.00 MiB Jul 02 07:50:02 linai ollama[1117]: llama_kv_cache_unified: KV self size = 1280.00 MiB, K (f16): 640.00 MiB, V (f16): 640.00 MiB Jul 02 07:50:02 linai ollama[1117]: llama_context: CPU compute buffer size = 696.01 MiB Jul 02 07:50:02 linai ollama[1117]: llama_context: graph nodes = 1526 Jul 02 07:50:02 linai ollama[1117]: llama_context: graph splits = 1 Jul 02 07:50:02 linai ollama[1117]: time=2025-07-02T07:50:02.319-04:00 level=INFO source=server.go:637 msg="llama runner started in 0.75 seconds" Jul 02 07:50:02 linai ollama[1117]: [GIN] 2025/07/02 - 07:50:02 | 200 | 1.577629607s | 127.0.0.1 | POST "/api/generate" ## [[ Rebooted System as part of Reinstalling NVIDIA drivers ]] Jul 02 08:25:51 linai systemd[1]: Started Ollama Service. Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.595-04:00 level=INFO source=routes.go:1235 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTT> Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.597-04:00 level=INFO source=images.go:476 msg="total blobs: 5" Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.597-04:00 level=INFO source=images.go:483 msg="total unused blobs removed: 0" Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.598-04:00 level=INFO source=routes.go:1288 msg="Listening on 127.0.0.1:11434 (version 0.9.3)" Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.598-04:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs" Jul 02 08:25:51 linai ollama[945]: time=2025-07-02T08:25:51.833-04:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-3a4bb711-3dff-4796-72b9-7ad73c0f74d7 library=cuda variant=v12 compute=7.5 driver=12.9 name="Quadro RTX 50> Jul 02 08:29:19 linai ollama[945]: [GIN] 2025/07/02 - 08:29:19 | 200 | 421.568µs | 127.0.0.1 | HEAD "/" Jul 02 08:29:19 linai ollama[945]: [GIN] 2025/07/02 - 08:29:19 | 200 | 50.302177ms | 127.0.0.1 | POST "/api/show" Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.358-04:00 level=INFO source=sched.go:788 msg="new model will fit in available VRAM in single GPU, loading" model=/usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f> Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.537-04:00 level=INFO source=server.go:135 msg="system memory" total="23.1 GiB" free="21.1 GiB" free_swap="64.0 GiB" Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.541-04:00 level=INFO source=server.go:175 msg=offload library=cuda layers.requested=-1 layers.model=41 layers.offload=41 layers.split="" memory.available="[14.2 GiB]" memory.gpu> Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.763-04:00 level=INFO source=server.go:438 msg="starting llama server" cmd="/usr/local/bin/ollama runner --model /usr/share/ollama/.ollama/models/blobs/sha256-a8cc1361f3145dc01f6> Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.763-04:00 level=INFO source=sched.go:483 msg="loaded runners" count=1 Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.763-04:00 level=INFO source=server.go:598 msg="waiting for llama runner to start responding" Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.764-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server not responding" Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.774-04:00 level=INFO source=runner.go:815 msg="starting go runner" Jul 02 08:29:19 linai ollama[945]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-sandybridge.so Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.781-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc) Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.781-04:00 level=INFO source=runner.go:874 msg="Server listening on 127.0.0.1:45463" Jul 02 08:29:20 linai ollama[945]: load_tensors: loading model tensors, this can take a while... (mmap = true) Jul 02 08:29:20 linai ollama[945]: time=2025-07-02T08:29:20.014-04:00 level=INFO source=server.go:632 msg="waiting for server to become available" status="llm server loading model" Jul 02 08:29:24 linai ollama[945]: load_tensors: CPU_Mapped model buffer size = 8840.78 MiB Jul 02 08:29:25 linai ollama[945]: time=2025-07-02T08:29:25.031-04:00 level=INFO source=server.go:637 msg="llama runner started in 5.27 seconds" Jul 02 08:29:25 linai ollama[945]: [GIN] 2025/07/02 - 08:29:25 | 200 | 5.940366649s | 127.0.0.1 | POST "/api/generate" Jul 02 08:29:34 linai ollama[945]: [GIN] 2025/07/02 - 08:29:34 | 200 | 7.463693701s | 127.0.0.1 | POST "/api/chat" Jul 02 08:29:38 linai ollama[945]: [GIN] 2025/07/02 - 08:29:38 | 200 | 17.555µs | 127.0.0.1 | HEAD "/" Jul 02 08:29:38 linai ollama[945]: [GIN] 2025/07/02 - 08:29:38 | 200 | 69.888µs | 127.0.0.1 | GET "/api/ps" Jul 02 08:34:39 linai ollama[945]: time=2025-07-02T08:34:39.335-04:00 level=WARN source=sched.go:687 msg="gpu VRAM usage didn't recover within timeout" seconds=5.196673098 runner.size="11.2 GiB" runner.vram="11.2 GiB" runner.parallel=2 r> Jul 02 08:34:39 linai ollama[945]: time=2025-07-02T08:34:39.586-04:00 level=WARN source=sched.go:687 msg="gpu VRAM usage didn't recover within timeout" seconds=5.447431288 runner.size="11.2 GiB" runner.vram="11.2 GiB" runner.parallel=2 r> ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.9.3
GiteaMirror added the bug label 2026-04-12 19:30:47 -05:00
GiteaMirror commented 2026-04-12 19:30:48 -05:00
Author
Owner
Copy Link

@KevinTurnbull commented on GitHub (Jul 2, 2025):

Here's my nvidia-smi after the reinstall of nvidia drivers.

linai@linai:~$ nvidia-smi
Wed Jul  2 08:42:28 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 575.57.08              Driver Version: 575.57.08      CUDA Version: 12.9     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  Quadro RTX 5000                Off |   00000000:01:00.0 Off |                    0 |
| 33%   31C    P5             30W /  230W |     225MiB /  15360MiB |      1%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1928      G   /usr/libexec/Xorg                        77MiB |
|    0   N/A  N/A            2021      G   /usr/bin/gnome-shell                    141MiB |
+-----------------------------------------------------------------------------------------+
<!-- gh-comment-id:3027735475 --> @KevinTurnbull commented on GitHub (Jul 2, 2025): Here's my nvidia-smi after the reinstall of nvidia drivers. ``` linai@linai:~$ nvidia-smi Wed Jul 2 08:42:28 2025 +-----------------------------------------------------------------------------------------+ | NVIDIA-SMI 575.57.08 Driver Version: 575.57.08 CUDA Version: 12.9 | |-----------------------------------------+------------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+========================+======================| | 0 Quadro RTX 5000 Off | 00000000:01:00.0 Off | 0 | | 33% 31C P5 30W / 230W | 225MiB / 15360MiB | 1% Default | | | | N/A | +-----------------------------------------+------------------------+----------------------+ +-----------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=========================================================================================| | 0 N/A N/A 1928 G /usr/libexec/Xorg 77MiB | | 0 N/A N/A 2021 G /usr/bin/gnome-shell 141MiB | +-----------------------------------------------------------------------------------------+ ```
GiteaMirror commented 2026-04-12 19:30:49 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 2, 2025):

Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.774-04:00 level=INFO source=runner.go:815 msg="starting go runner"
Jul 02 08:29:19 linai ollama[945]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-sandybridge.so
Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.781-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc)

No GPU backends found. How did you install ollama?

<!-- gh-comment-id:3027760228 --> @rick-github commented on GitHub (Jul 2, 2025): ``` Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.774-04:00 level=INFO source=runner.go:815 msg="starting go runner" Jul 02 08:29:19 linai ollama[945]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-sandybridge.so Jul 02 08:29:19 linai ollama[945]: time=2025-07-02T08:29:19.781-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc) ``` No GPU backends found. How did you install ollama?
GiteaMirror commented 2026-04-12 19:30:49 -05:00
Author
Owner
Copy Link

@KevinTurnbull commented on GitHub (Jul 2, 2025):

curl -fsSL https://ollama.com/install.sh | sh

Maybe drivers weren't in place at install time. I'll try an uninstall-install cycle of Ollama itself.

<!-- gh-comment-id:3027789823 --> @KevinTurnbull commented on GitHub (Jul 2, 2025): `curl -fsSL https://ollama.com/install.sh | sh` Maybe drivers weren't in place at install time. I'll try an uninstall-install cycle of Ollama itself.
GiteaMirror commented 2026-04-12 19:30:49 -05:00
Author
Owner
Copy Link

@KevinTurnbull commented on GitHub (Jul 2, 2025):

As per: https://www.reddit.com/r/ollama/comments/1eelxbc/comment/mpqdmit

# Remove Ollama binary as suggested In the Ollama docs 'if you are upgrading from a prior version...' here: https://github.com/ollama/ollama/blob/25248f4bd5a1d720f6775ee1e45dfed718d6f98f/docs/linux.md#manual-install
sudo rm -rf /usr/lib/ollama

# Reinstall Ollama
curl -fsSL https://ollama.com/install.sh | sh
<!-- gh-comment-id:3027796777 --> @KevinTurnbull commented on GitHub (Jul 2, 2025): As per: https://www.reddit.com/r/ollama/comments/1eelxbc/comment/mpqdmit ```sh # Remove Ollama binary as suggested In the Ollama docs 'if you are upgrading from a prior version...' here: https://github.com/ollama/ollama/blob/25248f4bd5a1d720f6775ee1e45dfed718d6f98f/docs/linux.md#manual-install sudo rm -rf /usr/lib/ollama # Reinstall Ollama curl -fsSL https://ollama.com/install.sh | sh ```
GiteaMirror commented 2026-04-12 19:30:49 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 2, 2025):

sudo rm -rf /usr/lib/ollama

The install script already does this.

<!-- gh-comment-id:3027808891 --> @rick-github commented on GitHub (Jul 2, 2025): > `sudo rm -rf /usr/lib/ollama` The install script already does this.
GiteaMirror commented 2026-04-12 19:30:50 -05:00
Author
Owner
Copy Link

@KevinTurnbull commented on GitHub (Jul 2, 2025):

Ahh! That's more like it

linai@linai:~$ sudo rm -rf /usr/lib/ollama
linai@linai:~$ curl -fsSL https://ollama.com/install.sh | sh
>>> Cleaning up old version at /usr/local/lib/ollama
>>> Installing ollama to /usr/local
>>> Downloading Linux amd64 bundle
######################################################################## 100.0%
>>> Adding ollama user to render group...
>>> Adding ollama user to video group...
>>> Adding current user to ollama group...
>>> Creating ollama systemd service...
>>> Enabling and starting ollama service...
>>> NVIDIA GPU installed.

Enjoy the saga that Qwen3:14b hallucinated to the pleasant screeching sound of the RTX5000's transistors roared to life.

linai@linai:~$ ollama run qwen3:14b --verbose
>>> Write me a story about a github issue #11271 for the Ollama project.
Thinking...
Okay, the user wants a story about GitHub issue #11271 for the Ollama 
project. Let me start by recalling what Ollama is. Ollama is an 
open-source project that allows users to run large language models 
locally. So, the story should revolve around a specific issue in their 
GitHub repository.

First, I need to think about the structure of a GitHub issue. Typically, 
there's a title, description, steps to reproduce, expected vs actual 
behavior, and maybe some comments from users and developers. The story 
should probably follow the journey of this issue from being reported to 
being resolved.

I should create a realistic scenario. Maybe a user encountered a problem 
when trying to run a model, like an error message or performance issue. 
Let's say the issue is related to model loading times being too slow. That's a common problem and relatable.

The title could be something like "Model Loading Takes Excessively Long Time on Windows 10". Then, the user describes the problem, 
maybe including their environment details (Windows 10, specific model version). They might have tried troubleshooting steps like 
checking disk space, reinstalling, but no luck.

Then, comments from other users might chime in, saying they have the same issue on different hardware, or maybe different OS 
versions. The Ollama team would respond, asking for more info, maybe logs or system specs. Then, a developer might investigate, 
find a bottleneck, like a specific library causing delays on Windows. They could propose a fix, maybe an optimization in the code, 
and then release a patch. The issue gets closed with gratitude from the community.

I need to make sure the story flows, has some technical details but remains understandable. Maybe include some back-and-forth in 
the comments to show collaboration. Also, highlight the community aspect, as open-source projects rely on user feedback. Perhaps 
end with the resolution and positive outcome, showing how the issue was handled effectively.
...done thinking.

**GitHub Issue #11271: "Model Loading Takes Excessively Long Time on Windows 10"**  
*Opened by @user4238 on April 5, 2024*  

---

### **Title**  
**Model Loading Takes Excessively Long Time on Windows 10**  

---

### **Description**  
When attempting to load the `llama3-8b` model on Windows 10 (version 22H2), the process consistently hangs for **10+ minutes** 
before proceeding. This occurs even with an SSD and 32GB of RAM. The same model loads in under 30 seconds on macOS and Linux.  

**Steps to Reproduce:**  
1. Install Ollama via the official Windows installer (v0.3.7).  
2. Run `ollama run llama3-8b`.  
3. Observe the terminal output: after the initial "Downloading model..." message, no further progress is shown for ~10 minutes.  
4. Eventually, the model loads, but the delay is unacceptable for production use.  

**Expected Behavior:**  
Model should load within 1–2 minutes, consistent with cross-platform performance.  

**Actual Behavior:**  
Model loading freezes for 10+ minutes on Windows.  

**Additional Context:**  
- CPU: Intel i7-12700K  
- GPU: NVIDIA RTX 4080  
- Ollama logs show no errors, but the process consumes ~50% CPU during the hang.  
- Tried reinstalling Ollama and clearing the cache (`~/.ollama`), but the issue persists.  

---

### **Comments**  

**@user4238 (April 5, 2024):**  
> This is blocking our team from adopting Ollama for Windows-based workflows. Any insights?  

**@user9876 (April 6, 2024):**  
> Same issue here! Using Windows 11, Ryzen 7 6800H, and it’s even worse—model loading takes ~15 minutes.  

**@ollama-bot (April 7, 2024):**  
> Thank you for the detailed report! We’re investigating. Could you share the full Ollama logs (`~/.ollama/logs/ollama.log`) and 
the output of `ollama show llama3-8b`?  

**@user4238 (April 8, 2024):**  
> Attached logs here: [link to gist]. The `show` command outputs:  
> ```  
> Model: llama3-8b  
> Size: 8B  
> Platform: Windows 10  
> Status: Downloaded (but took 10 minutes to reach this point)  
> ```  

---

### **Developer Response**  

**@matt (Ollama Maintainer) (April 10, 2024):**  
> Hi team! Thanks for the report. We’ve reproduced this on Windows 10/11 and identified the root cause: a **file locking issue** in 
the `ggml` library when using Windows’ `CreateFileMapping` API. The library was incorrectly holding a lock on the model file during 
initialization, causing a blocking delay.  

> **Fix:** We’ve merged a patch to `ggml` (commit [abc123](https://github.com/ollama/ggml/commit/abc123)) that releases the lock 
early. This will be included in the next Ollama release (v0.3.8).  

> **Workaround:** For now, you can manually delete the `~/.ollama/models/llama3-8b` directory and rerun `ollama run llama3-8b`. The 
model will download fresh without the lock.  

---

### **Resolution**  

**@user4238 (April 12, 2024):**  
> Just tested v0.3.8—model loads in 2 minutes! Thanks for the quick fix.  

**@matt (April 13, 2024):**  
> Closed as fixed in v0.3.8. Huge thanks to @user4238 and @user9876 for the detailed reports! This was a tricky 
one—Windows-specific issues are always a pain.  

---

### **Legacy**  
Issue #11271 became a cornerstone of Ollama’s Windows optimization efforts. The fix not only improved performance but also spurred 
the team to add Windows-specific testing to their CI pipeline. Users celebrated the resolution in the Ollama Discord server, with 
one member quipping:  
> “Finally, Windows isn’t the redheaded stepchild of Ollama anymore!”  

The issue remains a testament to the power of community-driven debugging in open source. 🚀  

---  
*Closed by @matt on April 13, 2024.*

total duration:       51.892452056s
load duration:        34.719249ms
prompt eval count:    31 token(s)
prompt eval duration: 170.343926ms
prompt eval rate:     181.98 tokens/s
eval count:           1472 token(s)
eval duration:        51.686570457s
eval rate:            28.48 tokens/s
<!-- gh-comment-id:3027813131 --> @KevinTurnbull commented on GitHub (Jul 2, 2025): Ahh! That's more like it ```bash linai@linai:~$ sudo rm -rf /usr/lib/ollama linai@linai:~$ curl -fsSL https://ollama.com/install.sh | sh >>> Cleaning up old version at /usr/local/lib/ollama >>> Installing ollama to /usr/local >>> Downloading Linux amd64 bundle ######################################################################## 100.0% >>> Adding ollama user to render group... >>> Adding ollama user to video group... >>> Adding current user to ollama group... >>> Creating ollama systemd service... >>> Enabling and starting ollama service... >>> NVIDIA GPU installed. ``` Enjoy the saga that Qwen3:14b hallucinated to the pleasant screeching sound of the RTX5000's transistors roared to life. ``` linai@linai:~$ ollama run qwen3:14b --verbose >>> Write me a story about a github issue #11271 for the Ollama project. Thinking... Okay, the user wants a story about GitHub issue #11271 for the Ollama project. Let me start by recalling what Ollama is. Ollama is an open-source project that allows users to run large language models locally. So, the story should revolve around a specific issue in their GitHub repository. First, I need to think about the structure of a GitHub issue. Typically, there's a title, description, steps to reproduce, expected vs actual behavior, and maybe some comments from users and developers. The story should probably follow the journey of this issue from being reported to being resolved. I should create a realistic scenario. Maybe a user encountered a problem when trying to run a model, like an error message or performance issue. Let's say the issue is related to model loading times being too slow. That's a common problem and relatable. The title could be something like "Model Loading Takes Excessively Long Time on Windows 10". Then, the user describes the problem, maybe including their environment details (Windows 10, specific model version). They might have tried troubleshooting steps like checking disk space, reinstalling, but no luck. Then, comments from other users might chime in, saying they have the same issue on different hardware, or maybe different OS versions. The Ollama team would respond, asking for more info, maybe logs or system specs. Then, a developer might investigate, find a bottleneck, like a specific library causing delays on Windows. They could propose a fix, maybe an optimization in the code, and then release a patch. The issue gets closed with gratitude from the community. I need to make sure the story flows, has some technical details but remains understandable. Maybe include some back-and-forth in the comments to show collaboration. Also, highlight the community aspect, as open-source projects rely on user feedback. Perhaps end with the resolution and positive outcome, showing how the issue was handled effectively. ...done thinking. **GitHub Issue #11271: "Model Loading Takes Excessively Long Time on Windows 10"** *Opened by @user4238 on April 5, 2024* --- ### **Title** **Model Loading Takes Excessively Long Time on Windows 10** --- ### **Description** When attempting to load the `llama3-8b` model on Windows 10 (version 22H2), the process consistently hangs for **10+ minutes** before proceeding. This occurs even with an SSD and 32GB of RAM. The same model loads in under 30 seconds on macOS and Linux. **Steps to Reproduce:** 1. Install Ollama via the official Windows installer (v0.3.7). 2. Run `ollama run llama3-8b`. 3. Observe the terminal output: after the initial "Downloading model..." message, no further progress is shown for ~10 minutes. 4. Eventually, the model loads, but the delay is unacceptable for production use. **Expected Behavior:** Model should load within 1–2 minutes, consistent with cross-platform performance. **Actual Behavior:** Model loading freezes for 10+ minutes on Windows. **Additional Context:** - CPU: Intel i7-12700K - GPU: NVIDIA RTX 4080 - Ollama logs show no errors, but the process consumes ~50% CPU during the hang. - Tried reinstalling Ollama and clearing the cache (`~/.ollama`), but the issue persists. --- ### **Comments** **@user4238 (April 5, 2024):** > This is blocking our team from adopting Ollama for Windows-based workflows. Any insights? **@user9876 (April 6, 2024):** > Same issue here! Using Windows 11, Ryzen 7 6800H, and it’s even worse—model loading takes ~15 minutes. **@ollama-bot (April 7, 2024):** > Thank you for the detailed report! We’re investigating. Could you share the full Ollama logs (`~/.ollama/logs/ollama.log`) and the output of `ollama show llama3-8b`? **@user4238 (April 8, 2024):** > Attached logs here: [link to gist]. The `show` command outputs: > ``` > Model: llama3-8b > Size: 8B > Platform: Windows 10 > Status: Downloaded (but took 10 minutes to reach this point) > ``` --- ### **Developer Response** **@matt (Ollama Maintainer) (April 10, 2024):** > Hi team! Thanks for the report. We’ve reproduced this on Windows 10/11 and identified the root cause: a **file locking issue** in the `ggml` library when using Windows’ `CreateFileMapping` API. The library was incorrectly holding a lock on the model file during initialization, causing a blocking delay. > **Fix:** We’ve merged a patch to `ggml` (commit [abc123](https://github.com/ollama/ggml/commit/abc123)) that releases the lock early. This will be included in the next Ollama release (v0.3.8). > **Workaround:** For now, you can manually delete the `~/.ollama/models/llama3-8b` directory and rerun `ollama run llama3-8b`. The model will download fresh without the lock. --- ### **Resolution** **@user4238 (April 12, 2024):** > Just tested v0.3.8—model loads in 2 minutes! Thanks for the quick fix. **@matt (April 13, 2024):** > Closed as fixed in v0.3.8. Huge thanks to @user4238 and @user9876 for the detailed reports! This was a tricky one—Windows-specific issues are always a pain. --- ### **Legacy** Issue #11271 became a cornerstone of Ollama’s Windows optimization efforts. The fix not only improved performance but also spurred the team to add Windows-specific testing to their CI pipeline. Users celebrated the resolution in the Ollama Discord server, with one member quipping: > “Finally, Windows isn’t the redheaded stepchild of Ollama anymore!” The issue remains a testament to the power of community-driven debugging in open source. 🚀 --- *Closed by @matt on April 13, 2024.* total duration: 51.892452056s load duration: 34.719249ms prompt eval count: 31 token(s) prompt eval duration: 170.343926ms prompt eval rate: 181.98 tokens/s eval count: 1472 token(s) eval duration: 51.686570457s eval rate: 28.48 tokens/s ```
GiteaMirror commented 2026-04-12 19:30:50 -05:00
Author
Owner
Copy Link

@KevinTurnbull commented on GitHub (Jul 3, 2025):

For posterity; I have also now confirmed that Davinci Resolve 20 can be running while Ollama does inferencing. I also had a script running that does video transcoding while Davinci Resolve and ollama were both active.

No conflicts detected (though running out of vram is a definite possibility).

I suspect something went sideways during the initial video driver install and ollama didn't get a chance to hook things up nicely. Reinstall drivers, reinstall Ollama.

<!-- gh-comment-id:3031789769 --> @KevinTurnbull commented on GitHub (Jul 3, 2025): For posterity; I have also now confirmed that Davinci Resolve 20 can be running while Ollama does inferencing. I also had a script running that does video transcoding while Davinci Resolve and ollama were both active. No conflicts detected (though running out of vram is a definite possibility). I suspect something went sideways during the initial video driver install and ollama didn't get a chance to hook things up nicely. Reinstall drivers, reinstall Ollama.
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#7431
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?