github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #15752] CUDA_VISIBLE_DEVICES=-1 incorrectly suppresses ROCm GPU detection #72101

New Issue
Open
opened 2026-05-05 03:28:55 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-05-05 03:28:55 -05:00
Owner
Copy Link

Originally created by @galuszkak on GitHub (Apr 22, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15752

What is the issue?

Setting CUDA_VISIBLE_DEVICES=-1 to hide NVIDIA GPUs unexpectedly causes the AMD ROCm GPU to disappear from inference compute as well, leaving only CPU available.

Steps to reproduce:

  1. Start ollama serve without restrictions — both CUDA and ROCm GPUs are discovered:

    inference compute ... library=CUDA ... name=CUDA0 description="NVIDIA GeForce RTX 5070 Laptop GPU"
inference compute ... library=ROCm ... name=ROCm0 description="AMD Radeon 890M Graphics"

  2. Stop the server, then run:

    CUDA_VISIBLE_DEVICES=-1 ollama serve

  3. Observe that only CPU is listed under inference compute; the ROCm GPU is no longer detected.

Expected behavior:

CUDA_VISIBLE_DEVICES should only affect CUDA-visible devices. The AMD iGPU (Radeon 890M) should still be discovered via ROCm and available for inference when this variable is set.

Actual behavior:

The ROCm GPU is completely omitted from the compute list, forcing CPU-only inference despite the AMD iGPU being present and functional.

Relevant log output

galuszkak@Thunder:~$ ollama serve 
time=2026-04-22T20:37:41.008+02:00 level=INFO source=routes.go:1752 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GGML_VK_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:0 OLLAMA_DEBUG:INFO OLLAMA_DEBUG_LOG_REQUESTS:false OLLAMA_EDITOR: OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 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:/home/galuszkak/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NO_CLOUD:false OLLAMA_NUM_PARALLEL:1 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_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false OLLAMA_VULKAN:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"
[...]
time=2026-04-22T20:37:42.087+02:00 level=INFO source=types.go:42 msg="inference compute" id=GPU-0eee6825-5aa8-a022-085a-6c135272070d filter_id="" library=CUDA compute=12.0 name=CUDA0 description="NVIDIA GeForce RTX 5070 Laptop GPU" libdirs=ollama,cuda_v13 driver=13.0 pci_id=0000:c1:00.0 type=discrete total="8.0 GiB" available="7.2 GiB"
time=2026-04-22T20:37:42.087+02:00 level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=ROCm compute=gfx1150 name=ROCm0 description="AMD Radeon 890M Graphics" libdirs=ollama,rocm driver=70253.21 pci_id=0000:c2:00.0 type=iGPU total="35.6 GiB" available="35.4 GiB"
time=2026-04-22T20:37:42.087+02:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="43.6 GiB" default_num_ctx=32768


and here with CUDA_VISIBLE_DEVICES=-1:

galuszkak@Thunder:~$ CUDA_VISIBLE_DEVICES=-1 ollama serve 
time=2026-04-22T20:37:54.239+02:00 level=INFO source=routes.go:1752 msg="server config" env="map[CUDA_VISIBLE_DEVICES:-1 GGML_VK_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:0 OLLAMA_DEBUG:INFO OLLAMA_DEBUG_LOG_REQUESTS:false OLLAMA_EDITOR: OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 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:/home/galuszkak/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NO_CLOUD:false OLLAMA_NUM_PARALLEL:1 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_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false OLLAMA_VULKAN:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"
[...]
time=2026-04-22T20:37:54.241+02:00 level=INFO source=runner.go:67 msg="discovering available GPUs..."
time=2026-04-22T20:37:54.241+02:00 level=WARN source=runner.go:485 msg="user overrode visible devices" CUDA_VISIBLE_DEVICES=-1
time=2026-04-22T20:37:54.242+02:00 level=WARN source=runner.go:489 msg="if GPUs are not correctly discovered, unset and try again"
time=2026-04-22T20:37:54.242+02:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 46377"
time=2026-04-22T20:37:54.318+02:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 36391"
time=2026-04-22T20:37:54.394+02:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 41067"
time=2026-04-22T20:37:54.576+02:00 level=INFO source=runner.go:106 msg="experimental Vulkan support disabled.  To enable, set OLLAMA_VULKAN=1"
time=2026-04-22T20:37:54.576+02:00 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="23.3 GiB" available="20.4 GiB"
time=2026-04-22T20:37:54.576+02:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="0 B" default_num_ctx=4096

OS

Linux

GPU

Nvidia, AMD

CPU

AMD

Ollama version

0.21.0

Originally created by @galuszkak on GitHub (Apr 22, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15752 ### What is the issue? Setting `CUDA_VISIBLE_DEVICES=-1` to hide NVIDIA GPUs unexpectedly causes the AMD ROCm GPU to disappear from inference compute as well, leaving only CPU available. Steps to reproduce: 1. Start `ollama serve` without restrictions — both CUDA and ROCm GPUs are discovered: ``` inference compute ... library=CUDA ... name=CUDA0 description="NVIDIA GeForce RTX 5070 Laptop GPU" inference compute ... library=ROCm ... name=ROCm0 description="AMD Radeon 890M Graphics" ``` 2. Stop the server, then run: ``` CUDA_VISIBLE_DEVICES=-1 ollama serve ``` 3. Observe that only CPU is listed under inference compute; the ROCm GPU is no longer detected. **Expected behavior:** `CUDA_VISIBLE_DEVICES` should only affect CUDA-visible devices. The AMD iGPU (Radeon 890M) should still be discovered via ROCm and available for inference when this variable is set. **Actual behavior:** The ROCm GPU is completely omitted from the compute list, forcing CPU-only inference despite the AMD iGPU being present and functional. ### Relevant log output ```shell galuszkak@Thunder:~$ ollama serve time=2026-04-22T20:37:41.008+02:00 level=INFO source=routes.go:1752 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GGML_VK_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:0 OLLAMA_DEBUG:INFO OLLAMA_DEBUG_LOG_REQUESTS:false OLLAMA_EDITOR: OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 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:/home/galuszkak/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NO_CLOUD:false OLLAMA_NUM_PARALLEL:1 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_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false OLLAMA_VULKAN:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]" [...] time=2026-04-22T20:37:42.087+02:00 level=INFO source=types.go:42 msg="inference compute" id=GPU-0eee6825-5aa8-a022-085a-6c135272070d filter_id="" library=CUDA compute=12.0 name=CUDA0 description="NVIDIA GeForce RTX 5070 Laptop GPU" libdirs=ollama,cuda_v13 driver=13.0 pci_id=0000:c1:00.0 type=discrete total="8.0 GiB" available="7.2 GiB" time=2026-04-22T20:37:42.087+02:00 level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=ROCm compute=gfx1150 name=ROCm0 description="AMD Radeon 890M Graphics" libdirs=ollama,rocm driver=70253.21 pci_id=0000:c2:00.0 type=iGPU total="35.6 GiB" available="35.4 GiB" time=2026-04-22T20:37:42.087+02:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="43.6 GiB" default_num_ctx=32768 and here with CUDA_VISIBLE_DEVICES=-1: galuszkak@Thunder:~$ CUDA_VISIBLE_DEVICES=-1 ollama serve time=2026-04-22T20:37:54.239+02:00 level=INFO source=routes.go:1752 msg="server config" env="map[CUDA_VISIBLE_DEVICES:-1 GGML_VK_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:0 OLLAMA_DEBUG:INFO OLLAMA_DEBUG_LOG_REQUESTS:false OLLAMA_EDITOR: OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 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:/home/galuszkak/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NO_CLOUD:false OLLAMA_NUM_PARALLEL:1 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_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false OLLAMA_VULKAN:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]" [...] time=2026-04-22T20:37:54.241+02:00 level=INFO source=runner.go:67 msg="discovering available GPUs..." time=2026-04-22T20:37:54.241+02:00 level=WARN source=runner.go:485 msg="user overrode visible devices" CUDA_VISIBLE_DEVICES=-1 time=2026-04-22T20:37:54.242+02:00 level=WARN source=runner.go:489 msg="if GPUs are not correctly discovered, unset and try again" time=2026-04-22T20:37:54.242+02:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 46377" time=2026-04-22T20:37:54.318+02:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 36391" time=2026-04-22T20:37:54.394+02:00 level=INFO source=server.go:444 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --port 41067" time=2026-04-22T20:37:54.576+02:00 level=INFO source=runner.go:106 msg="experimental Vulkan support disabled. To enable, set OLLAMA_VULKAN=1" time=2026-04-22T20:37:54.576+02:00 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="23.3 GiB" available="20.4 GiB" time=2026-04-22T20:37:54.576+02:00 level=INFO source=routes.go:1860 msg="vram-based default context" total_vram="0 B" default_num_ctx=4096 ``` ### OS Linux ### GPU Nvidia, AMD ### CPU AMD ### Ollama version 0.21.0
GiteaMirror added the bug label 2026-05-05 03:28:55 -05:00
GiteaMirror commented 2026-05-05 03:28:59 -05:00
Author
Owner
Copy Link

@galuszkak commented on GitHub (Apr 25, 2026):

After doing some investigation I found this in ROCm documentation:
https://rocm.docs.amd.com/en/latest/conceptual/gpu-isolation.html#cuda-visible-devices

CUDA_VISIBLE_DEVICES
Provided for CUDA compatibility, has the same effect as HIP_VISIBLE_DEVICES on the AMD platform.

Unsure now what to do with this - it looks like I should select on CUDA_VISIBLE_DEVICES also AMD devices? Not sure if this bug then should be closed because of that - or at least put note in documentation so people are also aware of this behaviour as it's not easy to find.

<!-- gh-comment-id:4317828675 --> @galuszkak commented on GitHub (Apr 25, 2026): After doing some investigation I found this in ROCm documentation: https://rocm.docs.amd.com/en/latest/conceptual/gpu-isolation.html#cuda-visible-devices > CUDA_VISIBLE_DEVICES > Provided for CUDA compatibility, has the same effect as HIP_VISIBLE_DEVICES on the AMD platform. Unsure now what to do with this - it looks like I should select on CUDA_VISIBLE_DEVICES also AMD devices? Not sure if this bug then should be closed because of that - or at least put note in documentation so people are also aware of this behaviour as it's not easy to find.
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#72101
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?