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 #11323] Enable GPU support with non-sudo local installation #7472

New Issue
Closed
opened 2026-04-12 19:32:40 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-12 19:32:40 -05:00
Owner
Copy Link

Originally created by @vitaglianog on GitHub (Jul 7, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11323

Hello,
I have installed ollama locally on a slurm node using the method detailed in the docs
Since I do not have sudo access, instead of extracting the tar in the /usr folder, I did it in a local folder with:

curl -L https://ollama.com/download/ollama-linux-amd64.tgz -o ollama-linux-amd64.tgz
tar -C .local/ -xzf ollama-linux-amd64.tgz

Although the GPUs are seen by the ollama server, it seems that the ollama-runner is unable to load the cuda backend (see logs).
I tried adding the .local/bin and .local/lib folders to the path with the following commands, but that does not seem to help.

export PATH="$PATH:~/.local/bin"
export LD_LIBRARY_PATH=:~/.local/lib:~/.local/lib/ollama:$LD_LIBRARY_PATH

ollama.log

The log attached is obtained runningollama serve and ollama run llama2 on a machine with 2 40GB GPUs so VRAM should not be a limiting factor.

Any insights on how to enable GPU inference?

Relevant log output

time=2025-07-07T16:46:35.854-04:00 level=DEBUG source=server.go:291 msg="compatible gpu libraries" compatible=[]

time=2025-07-07T16:46:35.940-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.AVX512=1 CPU.0.AVX512_VBMI=1 CPU.0.AVX512_VNNI=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc)

load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: layer   0 assigned to device CPU, is_swa = 0
load_tensors: layer   1 assigned to device CPU, is_swa = 0
load_tensors: layer   2 assigned to device CPU, is_swa = 0
load_tensors: layer   3 assigned to device CPU, is_swa = 0
load_tensors: layer   4 assigned to device CPU, is_swa = 0
load_tensors: layer   5 assigned to device CPU, is_swa = 0
load_tensors: layer   6 assigned to device CPU, is_swa = 0
load_tensors: layer   7 assigned to device CPU, is_swa = 0
load_tensors: layer   8 assigned to device CPU, is_swa = 0
load_tensors: layer   9 assigned to device CPU, is_swa = 0
load_tensors: layer  10 assigned to device CPU, is_swa = 0
load_tensors: layer  11 assigned to device CPU, is_swa = 0
load_tensors: layer  12 assigned to device CPU, is_swa = 0
load_tensors: layer  13 assigned to device CPU, is_swa = 0
load_tensors: layer  14 assigned to device CPU, is_swa = 0
load_tensors: layer  15 assigned to device CPU, is_swa = 0
load_tensors: layer  16 assigned to device CPU, is_swa = 0
load_tensors: layer  17 assigned to device CPU, is_swa = 0
load_tensors: layer  18 assigned to device CPU, is_swa = 0
load_tensors: layer  19 assigned to device CPU, is_swa = 0
load_tensors: layer  20 assigned to device CPU, is_swa = 0
load_tensors: layer  21 assigned to device CPU, is_swa = 0
load_tensors: layer  22 assigned to device CPU, is_swa = 0
load_tensors: layer  23 assigned to device CPU, is_swa = 0
load_tensors: layer  24 assigned to device CPU, is_swa = 0
load_tensors: layer  25 assigned to device CPU, is_swa = 0
load_tensors: layer  26 assigned to device CPU, is_swa = 0
load_tensors: layer  27 assigned to device CPU, is_swa = 0
load_tensors: layer  28 assigned to device CPU, is_swa = 0
load_tensors: layer  29 assigned to device CPU, is_swa = 0
load_tensors: layer  30 assigned to device CPU, is_swa = 0
load_tensors: layer  31 assigned to device CPU, is_swa = 0
load_tensors: layer  32 assigned to device CPU, is_swa = 0

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

0.9.5

Originally created by @vitaglianog on GitHub (Jul 7, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11323 Hello, I have installed ollama locally on a slurm node using the method detailed in the [docs](https://github.com/ollama/ollama/blob/main/docs/linux.md#uninstall) Since I do not have sudo access, instead of extracting the tar in the `/usr` folder, I did it in a local folder with: ``` curl -L https://ollama.com/download/ollama-linux-amd64.tgz -o ollama-linux-amd64.tgz tar -C .local/ -xzf ollama-linux-amd64.tgz ``` Although the GPUs are seen by the ollama server, it seems that the ollama-runner is unable to load the cuda backend (see logs). I tried adding the `.local/bin` and `.local/lib` folders to the path with the following commands, but that does not seem to help. ``` export PATH="$PATH:~/.local/bin" export LD_LIBRARY_PATH=:~/.local/lib:~/.local/lib/ollama:$LD_LIBRARY_PATH ``` [ollama.log](https://github.com/user-attachments/files/21110000/ollama.log) The log attached is obtained running`ollama serve` and `ollama run llama2` on a machine with 2 40GB GPUs so VRAM should not be a limiting factor. Any insights on how to enable GPU inference? ### Relevant log output ```shell time=2025-07-07T16:46:35.854-04:00 level=DEBUG source=server.go:291 msg="compatible gpu libraries" compatible=[] time=2025-07-07T16:46:35.940-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.AVX512=1 CPU.0.AVX512_VBMI=1 CPU.0.AVX512_VNNI=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc) load_tensors: loading model tensors, this can take a while... (mmap = true) load_tensors: layer 0 assigned to device CPU, is_swa = 0 load_tensors: layer 1 assigned to device CPU, is_swa = 0 load_tensors: layer 2 assigned to device CPU, is_swa = 0 load_tensors: layer 3 assigned to device CPU, is_swa = 0 load_tensors: layer 4 assigned to device CPU, is_swa = 0 load_tensors: layer 5 assigned to device CPU, is_swa = 0 load_tensors: layer 6 assigned to device CPU, is_swa = 0 load_tensors: layer 7 assigned to device CPU, is_swa = 0 load_tensors: layer 8 assigned to device CPU, is_swa = 0 load_tensors: layer 9 assigned to device CPU, is_swa = 0 load_tensors: layer 10 assigned to device CPU, is_swa = 0 load_tensors: layer 11 assigned to device CPU, is_swa = 0 load_tensors: layer 12 assigned to device CPU, is_swa = 0 load_tensors: layer 13 assigned to device CPU, is_swa = 0 load_tensors: layer 14 assigned to device CPU, is_swa = 0 load_tensors: layer 15 assigned to device CPU, is_swa = 0 load_tensors: layer 16 assigned to device CPU, is_swa = 0 load_tensors: layer 17 assigned to device CPU, is_swa = 0 load_tensors: layer 18 assigned to device CPU, is_swa = 0 load_tensors: layer 19 assigned to device CPU, is_swa = 0 load_tensors: layer 20 assigned to device CPU, is_swa = 0 load_tensors: layer 21 assigned to device CPU, is_swa = 0 load_tensors: layer 22 assigned to device CPU, is_swa = 0 load_tensors: layer 23 assigned to device CPU, is_swa = 0 load_tensors: layer 24 assigned to device CPU, is_swa = 0 load_tensors: layer 25 assigned to device CPU, is_swa = 0 load_tensors: layer 26 assigned to device CPU, is_swa = 0 load_tensors: layer 27 assigned to device CPU, is_swa = 0 load_tensors: layer 28 assigned to device CPU, is_swa = 0 load_tensors: layer 29 assigned to device CPU, is_swa = 0 load_tensors: layer 30 assigned to device CPU, is_swa = 0 load_tensors: layer 31 assigned to device CPU, is_swa = 0 load_tensors: layer 32 assigned to device CPU, is_swa = 0 ``` ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version 0.9.5
GiteaMirror added the bug label 2026-04-12 19:32:40 -05:00
GiteaMirror commented 2026-04-12 19:32:41 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 7, 2025):

time=2025-07-07T16:46:35.915-04:00 level=DEBUG source=ggml.go:94 msg="ggml backend load all from path" path=/nfs/home2/user/.local/lib/ollama
load_backend: loaded CPU backend from /nfs/home2/user/.local/lib/ollama/libggml-cpu-icelake.so
time=2025-07-07T16:46:35.940-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.AVX512=1 CPU.0.AVX512_VBMI=1 CPU.0.AVX512_VNNI=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc)

The runner found the CPU backend but not any GPU backends. What's the output of

ls -l /nfs/home2/user/.local/lib/ollama
<!-- gh-comment-id:3046544015 --> @rick-github commented on GitHub (Jul 7, 2025): ``` time=2025-07-07T16:46:35.915-04:00 level=DEBUG source=ggml.go:94 msg="ggml backend load all from path" path=/nfs/home2/user/.local/lib/ollama load_backend: loaded CPU backend from /nfs/home2/user/.local/lib/ollama/libggml-cpu-icelake.so time=2025-07-07T16:46:35.940-04:00 level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.AVX512=1 CPU.0.AVX512_VBMI=1 CPU.0.AVX512_VNNI=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 compiler=cgo(gcc) ``` The runner found the CPU backend but not any GPU backends. What's the output of ``` ls -l /nfs/home2/user/.local/lib/ollama ```
GiteaMirror commented 2026-04-12 19:32:41 -05:00
Author
Owner
Copy Link

@vitaglianog commented on GitHub (Jul 7, 2025):

Thank you for the quick reply! The output is:

$ ls -l /nfs/home2/user/.local/lib/ollama/
total 2700456
lrwxrwxrwx 1 user user         21 Jul  2 13:32 libcublas.so.12 -> libcublas.so.12.8.4.1
-rwxr-xr-x 1 user user  116388640 Jul  7  2015 libcublas.so.12.8.4.1
lrwxrwxrwx 1 user user         23 Jul  2 13:32 libcublasLt.so.12 -> libcublasLt.so.12.8.4.1
-rwxr-xr-x 1 user user  751771728 Jul  7  2015 libcublasLt.so.12.8.4.1
lrwxrwxrwx 1 user user         20 Jul  2 13:32 libcudart.so.12 -> libcudart.so.12.8.90
-rwxr-xr-x 1 user user     728800 Jul  7  2015 libcudart.so.12.8.90
-rwxr-xr-x 1 user user     595648 Jul  2 13:23 libggml-base.so
-rwxr-xr-x 1 user user     619280 Jul  2 13:23 libggml-cpu-alderlake.so
-rwxr-xr-x 1 user user     619280 Jul  2 13:23 libggml-cpu-haswell.so
-rwxr-xr-x 1 user user     725776 Jul  2 13:23 libggml-cpu-icelake.so
-rwxr-xr-x 1 user user     606992 Jul  2 13:23 libggml-cpu-sandybridge.so
-rwxr-xr-x 1 user user     729872 Jul  2 13:23 libggml-cpu-skylakex.so
-rwxr-xr-x 1 user user     480048 Jul  2 13:23 libggml-cpu-sse42.so
-rwxr-xr-x 1 user user     475952 Jul  2 13:23 libggml-cpu-x64.so
-rwxr-xr-x 1 user user 1286539248 Jul  2 13:32 libggml-cuda.so
-rwxr-xr-x 1 user user  604949568 Jul  2 13:33 libggml-hip.so
<!-- gh-comment-id:3046558208 --> @vitaglianog commented on GitHub (Jul 7, 2025): Thank you for the quick reply! The output is: ``` $ ls -l /nfs/home2/user/.local/lib/ollama/ total 2700456 lrwxrwxrwx 1 user user 21 Jul 2 13:32 libcublas.so.12 -> libcublas.so.12.8.4.1 -rwxr-xr-x 1 user user 116388640 Jul 7 2015 libcublas.so.12.8.4.1 lrwxrwxrwx 1 user user 23 Jul 2 13:32 libcublasLt.so.12 -> libcublasLt.so.12.8.4.1 -rwxr-xr-x 1 user user 751771728 Jul 7 2015 libcublasLt.so.12.8.4.1 lrwxrwxrwx 1 user user 20 Jul 2 13:32 libcudart.so.12 -> libcudart.so.12.8.90 -rwxr-xr-x 1 user user 728800 Jul 7 2015 libcudart.so.12.8.90 -rwxr-xr-x 1 user user 595648 Jul 2 13:23 libggml-base.so -rwxr-xr-x 1 user user 619280 Jul 2 13:23 libggml-cpu-alderlake.so -rwxr-xr-x 1 user user 619280 Jul 2 13:23 libggml-cpu-haswell.so -rwxr-xr-x 1 user user 725776 Jul 2 13:23 libggml-cpu-icelake.so -rwxr-xr-x 1 user user 606992 Jul 2 13:23 libggml-cpu-sandybridge.so -rwxr-xr-x 1 user user 729872 Jul 2 13:23 libggml-cpu-skylakex.so -rwxr-xr-x 1 user user 480048 Jul 2 13:23 libggml-cpu-sse42.so -rwxr-xr-x 1 user user 475952 Jul 2 13:23 libggml-cpu-x64.so -rwxr-xr-x 1 user user 1286539248 Jul 2 13:32 libggml-cuda.so -rwxr-xr-x 1 user user 604949568 Jul 2 13:33 libggml-hip.so ```
GiteaMirror commented 2026-04-12 19:32:42 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 7, 2025):

Unset ROCR_VISIBLE_DEVICES.

<!-- gh-comment-id:3046573436 --> @rick-github commented on GitHub (Jul 7, 2025): Unset `ROCR_VISIBLE_DEVICES`.
GiteaMirror commented 2026-04-12 19:32:43 -05:00
Author
Owner
Copy Link

@vitaglianog commented on GitHub (Jul 7, 2025):

If I run the server with OLLAMA_DEBUG='1' ROCR_VISIBLE_DEVICES='' ollama serve
the runner is still not using the cuda backend. However the log confirms there are no ROCR_VISIBLE_DEVICES:

time=2025-07-07T17:36:45.640-04:00 level=INFO source=routes.go:1235 msg="server config" env="map[CUDA_VISIBLE_DEVICES:0,1 GPU_DEVICE_ORDINAL:0,1 HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:4096 OLLAMA_DEBUG:DEBUG OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/nobackup1/user/ollama-models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:true ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"

[...]
time=2025-07-07T17:37:12.790-04:00 level=DEBUG source=server.go:291 msg="compatible gpu libraries" compatible=[]

ollama.log

<!-- gh-comment-id:3046598298 --> @vitaglianog commented on GitHub (Jul 7, 2025): If I run the server with `OLLAMA_DEBUG='1' ROCR_VISIBLE_DEVICES='' ollama serve` the runner is still not using the cuda backend. However the log confirms there are no `ROCR_VISIBLE_DEVICES`: ``` time=2025-07-07T17:36:45.640-04:00 level=INFO source=routes.go:1235 msg="server config" env="map[CUDA_VISIBLE_DEVICES:0,1 GPU_DEVICE_ORDINAL:0,1 HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:4096 OLLAMA_DEBUG:DEBUG OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/nobackup1/user/ollama-models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:true ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]" [...] time=2025-07-07T17:37:12.790-04:00 level=DEBUG source=server.go:291 msg="compatible gpu libraries" compatible=[] ``` [ollama.log](https://github.com/user-attachments/files/21110496/ollama.log)
GiteaMirror commented 2026-04-12 19:32:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 7, 2025):

Setting it to the empty string is not enough, it needs to be unset.

$ unset ROCR_VISIBLE_DEVICES
$ OLLAMA_DEBUG='1' ollama serve
<!-- gh-comment-id:3046603326 --> @rick-github commented on GitHub (Jul 7, 2025): Setting it to the empty string is not enough, it needs to be unset. ```console $ unset ROCR_VISIBLE_DEVICES $ OLLAMA_DEBUG='1' ollama serve ```
GiteaMirror commented 2026-04-12 19:32:44 -05:00
Author
Owner
Copy Link

@vitaglianog commented on GitHub (Jul 7, 2025):

Thank you very much! By doing this the runner actually picks up the CUDA devices. I assume this is an artifact of the slurm cluster having other nodes with ROCR devices.
Probably it would be helpful for the runner to prioritize CUDA_DEVICES if they are found.

<!-- gh-comment-id:3046610634 --> @vitaglianog commented on GitHub (Jul 7, 2025): Thank you very much! By doing this the runner actually picks up the CUDA devices. I assume this is an artifact of the slurm cluster having other nodes with ROCR devices. Probably it would be helpful for the runner to prioritize `CUDA_DEVICES` if they are found.
GiteaMirror commented 2026-04-12 19:32:44 -05:00
Author
Owner
Copy Link

@Noonlord commented on GitHub (Jul 23, 2025):

Thanks a lot! Unsetting ROCR_VISIBLE_DEVICES also fixed the issue for us, in an HPC cluster. Though we didn't have any AMD GPUs in the cluster at all!

<!-- gh-comment-id:3106350729 --> @Noonlord commented on GitHub (Jul 23, 2025): Thanks a lot! Unsetting ROCR_VISIBLE_DEVICES also fixed the issue for us, in an HPC cluster. Though we didn't have any AMD GPUs in the cluster at all!
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#7472
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?