github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 13:19:40 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #14123] RTX 3060TIFE & GTX 960 2DG5 Cuda Error question. #9214

New Issue
Closed
opened 2026-04-12 22:04:35 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 22:04:35 -05:00
Owner
Copy Link

Originally created by @jekv2 on GitHub (Feb 6, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14123

Windows 10 22H2.

Having both RTX 3060TIFE & GTX 960 2DG5 enabled, I get

Error 500 Internal Server Error: llama runner process has terminated: CUDA error

I asked chatgpt.com, results are below and correct, I disabled the 960, then able to to use Ollama. So because they they both use different compute versions.

So my question is, absolutely cannot combine the two GPU's for the VRAM for one Model on Ollama?

two different gpu's rtx 3060 and a gtx 960

Ahhh, that explains it 😬 — mixed GPUs is the smoking gun here.

An RTX 3060 (Ampere, compute 8.6) + GTX 960 (Maxwell, compute 5.2) is very likely to crash CUDA-based LLaMA runners.

Here’s what’s going on and how to fix it.

Why this combo breaks things

CUDA apps usually:

Pick GPU 0 automatically

Or try to initialize all visible GPUs

The GTX 960 is ancient by CUDA standards:

Different compute capability

Different kernel support

Often not supported by modern llama.cpp / Ollama CUDA builds

So when the runner touches the 960 → boom, CUDA error, process dies, 500 error.

Originally created by @jekv2 on GitHub (Feb 6, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14123 Windows 10 22H2. Having both RTX 3060TIFE & GTX 960 2DG5 enabled, I get Error 500 Internal Server Error: llama runner process has terminated: CUDA error I asked chatgpt.com, results are below and correct, I disabled the 960, then able to to use Ollama. So because they they both use different compute versions. **So my question is, absolutely cannot combine the two GPU's for the VRAM for one Model on Ollama?** ---------------- two different gpu's rtx 3060 and a gtx 960 Ahhh, that explains it 😬 — mixed GPUs is the smoking gun here. An RTX 3060 (Ampere, compute 8.6) + GTX 960 (Maxwell, compute 5.2) is very likely to crash CUDA-based LLaMA runners. Here’s what’s going on and how to fix it. Why this combo breaks things CUDA apps usually: Pick GPU 0 automatically Or try to initialize all visible GPUs The GTX 960 is ancient by CUDA standards: Different compute capability Different kernel support Often not supported by modern llama.cpp / Ollama CUDA builds So when the runner touches the 960 → boom, CUDA error, process dies, 500 error.
GiteaMirror commented 2026-04-12 22:04:36 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 6, 2026):

Ollama server logs and the output of nvidia-smi may help.

<!-- gh-comment-id:3862394263 --> @rick-github commented on GitHub (Feb 6, 2026): Ollama server logs and the output of `nvidia-smi` may help.
GiteaMirror commented 2026-04-12 22:04:37 -05:00
Author
Owner
Copy Link

@jekv2 commented on GitHub (Feb 7, 2026):

Ollama server logs and the output of nvidia-smi may help.

Ok, I am not sure what I have done to have it fixed.

This is awesome. I did enable airplane mode and done a bit of disabling stuff with shutupwindows app and sexy.privacy.

I am not receiving the cuda error now for some reason and vram on both gpu's are being utilized.

RTX 3060TIFE-8GB/VRAM & GTX 960 2DG5-2GB/VRAM, Ollama utilizing 11GB of system 32GB overclocked and fine tuned ram.

Runing qwen3:30b model, and it's fast.

So, 10GB vram + the 11from system, RTX & GTX GPU utilization jumps up "thumbs up".

ScreenShot.

Image
<!-- gh-comment-id:3863512051 --> @jekv2 commented on GitHub (Feb 7, 2026): > Ollama server logs and the output of `nvidia-smi` may help. Ok, I am not sure what I have done to have it fixed. This is awesome. I did enable airplane mode and done a bit of disabling stuff with shutupwindows app and sexy.privacy. I am not receiving the cuda error now for some reason and vram on both gpu's are being utilized. RTX 3060TIFE-8GB/VRAM & GTX 960 2DG5-2GB/VRAM, Ollama utilizing 11GB of system 32GB overclocked and fine tuned ram. Runing qwen3:30b model, and it's fast. So, 10GB vram + the 11from system, RTX & GTX GPU utilization jumps up "thumbs up". ScreenShot. <img width="1875" height="577" alt="Image" src="https://github.com/user-attachments/assets/1c83a058-28fb-4ea5-b94b-98ae9fdd88f3" />
GiteaMirror commented 2026-04-12 22:04:38 -05:00
Author
Owner
Copy Link

@jekv2 commented on GitHub (Feb 7, 2026):

I got a significant improvement adding this GTX, I am now able to run gpt-oss:20b @ high thinking level "thumbs up"

Ima state that everything is overclocked aswell, fine tuned, CPU/Memory/GPU's-GPU-Memory.

<!-- gh-comment-id:3863526109 --> @jekv2 commented on GitHub (Feb 7, 2026): I got a significant improvement adding this GTX, I am now able to run gpt-oss:20b @ high thinking level "thumbs up" Ima state that everything is overclocked aswell, fine tuned, CPU/Memory/GPU's-GPU-Memory.
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
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#9214
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?