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 #4979] 0xc0000409 error with llava-phi3 #65185

New Issue
Closed
opened 2026-05-03 19:57:07 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-05-03 19:57:07 -05:00
Owner
Copy Link

Originally created by @razvanab on GitHub (Jun 11, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4979

What is the issue?

I get the 0xc0000409 error with llava-phi3 when OLLAMA_FLASH_ATTENTION is enabled.

RAM = 32GB

GPU = GTX 1060
VRAM = 6GB

OS

Windows

GPU

Nvidia

CPU

AMD

Ollama version

0.1.42

Originally created by @razvanab on GitHub (Jun 11, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4979 ### What is the issue? I get the 0xc0000409 error with llava-phi3 when OLLAMA_FLASH_ATTENTION is enabled. RAM = 32GB GPU = GTX 1060 VRAM = 6GB ### OS Windows ### GPU Nvidia ### CPU AMD ### Ollama version 0.1.42
GiteaMirror added the bug label 2026-05-03 19:57:07 -05:00
GiteaMirror commented 2026-05-03 19:57:13 -05:00
Author
Owner
Copy Link

@AncientMystic commented on GitHub (Jun 11, 2024):

Pascal gpus do not officially support flash attention due to their lack of tensor cores and require a FP32 non tensor core vector kernel which has been recently updated for support in llama.cpp and i am not sure if ollama is using a recent enough version of llama.cpp as it should provide almost a 2x boost in tokens per second, which i notice no difference on my tesla p4

Not sure if that is the problem here at all but it could be apart of it

<!-- gh-comment-id:2160606522 --> @AncientMystic commented on GitHub (Jun 11, 2024): Pascal gpus do not officially support flash attention due to their lack of tensor cores and require a FP32 non tensor core vector kernel which has been recently updated for support in llama.cpp and i am not sure if ollama is using a recent enough version of llama.cpp as it should provide almost a 2x boost in tokens per second, which i notice no difference on my tesla p4 Not sure if that is the problem here at all but it could be apart of it
GiteaMirror commented 2026-05-03 19:57:16 -05:00
Author
Owner
Copy Link

@razvanab commented on GitHub (Jun 11, 2024):

I see. Thank you.

<!-- gh-comment-id:2160614611 --> @razvanab commented on GitHub (Jun 11, 2024): I see. Thank you.
GiteaMirror commented 2026-05-03 19:57:18 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jun 18, 2024):

It sounds like we can close this issue. Eventually our goal is to get flash attention turned on automatically based on if the GPU can support it, but until then, the experimental env var can be used to enable manually.

<!-- gh-comment-id:2176888963 --> @dhiltgen commented on GitHub (Jun 18, 2024): It sounds like we can close this issue. Eventually our goal is to get flash attention turned on automatically based on if the GPU can support it, but until then, the experimental env var can be used to enable manually.
GiteaMirror commented 2026-05-03 19:57:18 -05:00
Author
Owner
Copy Link

@AncientMystic commented on GitHub (Jun 18, 2024):

It sounds like we can close this issue. Eventually our goal is to get flash attention turned on automatically based on if the GPU can support it, but until then, the experimental env var can be used to enable manually.

It does seem to be solved and it would be really nice to see the fp32 vector kernel flash attention supported on ollama (doesn't seem to be yet but from my understanding it is now in the last month on llama.cpp) as many of us are still on pascal which also has crippled fp16 performance, that plus the recently added kv quant should really benefit pascels

<!-- gh-comment-id:2176920624 --> @AncientMystic commented on GitHub (Jun 18, 2024): > It sounds like we can close this issue. Eventually our goal is to get flash attention turned on automatically based on if the GPU can support it, but until then, the experimental env var can be used to enable manually. It does seem to be solved and it would be really nice to see the fp32 vector kernel flash attention supported on ollama (doesn't seem to be yet but from my understanding it is now in the last month on llama.cpp) as many of us are still on pascal which also has crippled fp16 performance, that plus the recently added kv quant should really benefit pascels
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#65185
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?