github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #9165] Inconsistent results when mixing multiple nvidia gpu #5965

New Issue
Open
opened 2026-04-12 17:18:58 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 17:18:58 -05:00
Owner
Copy Link

Originally created by @galvanoid on GitHub (Feb 17, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9165

What is the issue?

I have this setup:

2 x rtx3060
1 x gtx1060

So far everything has worked perfectly, but I updated the Nvidia drivers to version 565 and I have the following problem.

If I load a model that fits one or both rtx3060s, it works fine, with consistent responses (especially in RAG).

But if I load a model that occupies both rtx3060s together with the gtx1060 (different model or the same one with more quants), then the responses are inconsistent and incoherent.

If I update the Nvidia drivers to version 570 (latest), the result is the same.

Note that every time I update the drivers, the CUDA version is also updated, so I don't know if it's something related to the drivers or the CUDA version.

However, when downgrading the driver to version 535 or 550, the responses are correct and coherent.

As I said, this is especially noticeable when using RAG (document collections).

Regards.

Relevant log output

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.5.11

Originally created by @galvanoid on GitHub (Feb 17, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9165 ### What is the issue? I have this setup: 2 x rtx3060 1 x gtx1060 So far everything has worked perfectly, but I updated the Nvidia drivers to version 565 and I have the following problem. If I load a model that fits one or both rtx3060s, it works fine, with consistent responses (especially in RAG). But if I load a model that occupies both rtx3060s together with the gtx1060 (different model or the same one with more quants), then the responses are inconsistent and incoherent. If I update the Nvidia drivers to version 570 (latest), the result is the same. Note that every time I update the drivers, the CUDA version is also updated, so I don't know if it's something related to the drivers or the CUDA version. However, when downgrading the driver to version 535 or 550, the responses are correct and coherent. As I said, this is especially noticeable when using RAG (document collections). Regards. ### Relevant log output ```shell ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.5.11
GiteaMirror added the bug label 2026-04-12 17:18:58 -05:00
GiteaMirror commented 2026-04-12 17:19:00 -05:00
Author
Owner
Copy Link

@dnielso5 commented on GitHub (Jan 17, 2026):

Following as I have a 3060 and 1060 in my truenas and upgrading to 25.10 updates the driver to 570.

<!-- gh-comment-id:3762719800 --> @dnielso5 commented on GitHub (Jan 17, 2026): Following as I have a 3060 and 1060 in my truenas and upgrading to 25.10 updates the driver to 570.
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#5965
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?