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 #5504] 0xc0000409 CUDA error | was working fine before - OOM crash #49953

New Issue
Closed
opened 2026-04-28 13:32:41 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-28 13:32:41 -05:00
Owner
Copy Link

Originally created by @gaduffl on GitHub (Jul 5, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/5504

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

Ollama was working fine with all small models I tested so far (4GB VRAM).
After upgrading to 0.1.48, I get a CUDA error with all models, e.g. Llama3 8B:

Error: llama runner process has terminated: exit status 0xc0000409 CUDA error"

This model was running perfectly fine before.
server.log

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.1.48

Originally created by @gaduffl on GitHub (Jul 5, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/5504 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? Ollama was working fine with all small models I tested so far (4GB VRAM). After upgrading to 0.1.48, I get a CUDA error with all models, e.g. Llama3 8B: _Error: llama runner process has terminated: exit status 0xc0000409 CUDA error"_ This model was running perfectly fine before. [server.log](https://github.com/user-attachments/files/16113333/server.log) ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.1.48
GiteaMirror added the memorybugwindows labels 2026-04-28 13:32:44 -05:00
GiteaMirror commented 2026-04-28 13:32:49 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jul 9, 2024):

Digging into this a bit, it looks like there's a skew between what the management library and driver libraries are reporting on windows when the GPU is the primary display device. I believe the management library is unaware of some amount of allocation on the card (probably OS level), and as a result, our recent changes to favor the management library to enable concurrency scheduling leads us to over allocate. I'll get a PR up soon to record the skew between these two and remember that overhead so we can set that VRAM aside which should resolve this.

<!-- gh-comment-id:2218136775 --> @dhiltgen commented on GitHub (Jul 9, 2024): Digging into this a bit, it looks like there's a skew between what the management library and driver libraries are reporting on windows when the GPU is the primary display device. I believe the management library is unaware of some amount of allocation on the card (probably OS level), and as a result, our recent changes to favor the management library to enable concurrency scheduling leads us to over allocate. I'll get a PR up soon to record the skew between these two and remember that overhead so we can set that VRAM aside which should resolve this.
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 memory windows
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#49953
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?