github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-23 06:01:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #15918] Silent CPU fallback after driver update: CUDA forward compatibility error causes 100% CPU usage with no user warning #87826

New Issue
Open
opened 2026-05-10 06:25:52 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-10 06:25:52 -05:00
Owner
Copy Link

Originally created by @0xpietri on GitHub (May 1, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15918

What is the issue?

After updating NVIDIA drivers today, ollama silently fell back to CPU-only inference instead of failing fast or warning the user. The model (qwen3.5:9b, Q4_K_M, 6.1 GiB) loaded entirely on CPU, consuming 766% CPU and 16 GB RAM for ~38 minutes, causing the system to reach 101°C, while the API kept returning 500 errors after 60s timeouts.

The fallback happened 3 times at 30-minute intervals (12:28, 12:58, 13:28) triggered by an external tool (OpenAI Codex) calling POST /api/chat. No warning was shown anywhere — ollama appeared to be running normally.

Expected behavior: ollama should either refuse to load the model with a clear error, or emit a visible WARNING log when falling back to CPU due to a CUDA initialization failure.

Ollama version: 0.20.6
NVIDIA driver updated today (NVML library version: 580.159)
nvidia-smi itself fails with "Driver/library version mismatch" — kernel module still loaded from previous driver version, reboot pending.

Relevant log output

ggml_cuda_init: failed to initialize CUDA: forward compatibility was attempted on non supported HW
time=... source=ggml.go:494 msg="offloaded 0/33 layers to GPU"
time=... source=device.go:245 msg="model weights" device=CPU size="6.1 GiB"
time=... source=device.go:256 msg="kv cache" device=CPU size="9.2 GiB"
time=... source=device.go:267 msg="compute graph" device=CPU size="1.5 GiB"
time=... source=device.go:272 msg="total memory" size="16.8 GiB"
[GIN] 2026/05/01 - 13:29:04 | 500 | 59.999378025s | 192.168.0.38 | POST "/api/chat"

OS

No response

GPU

No response

CPU

No response

Ollama version

No response

Originally created by @0xpietri on GitHub (May 1, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15918 ### What is the issue? After updating NVIDIA drivers today, ollama silently fell back to CPU-only inference instead of failing fast or warning the user. The model (qwen3.5:9b, Q4_K_M, 6.1 GiB) loaded entirely on CPU, consuming 766% CPU and 16 GB RAM for ~38 minutes, causing the system to reach 101°C, while the API kept returning 500 errors after 60s timeouts. The fallback happened 3 times at 30-minute intervals (12:28, 12:58, 13:28) triggered by an external tool (OpenAI Codex) calling POST /api/chat. No warning was shown anywhere — ollama appeared to be running normally. Expected behavior: ollama should either refuse to load the model with a clear error, or emit a visible WARNING log when falling back to CPU due to a CUDA initialization failure. Ollama version: 0.20.6 NVIDIA driver updated today (NVML library version: 580.159) nvidia-smi itself fails with "Driver/library version mismatch" — kernel module still loaded from previous driver version, reboot pending. ### Relevant log output ```shell ggml_cuda_init: failed to initialize CUDA: forward compatibility was attempted on non supported HW time=... source=ggml.go:494 msg="offloaded 0/33 layers to GPU" time=... source=device.go:245 msg="model weights" device=CPU size="6.1 GiB" time=... source=device.go:256 msg="kv cache" device=CPU size="9.2 GiB" time=... source=device.go:267 msg="compute graph" device=CPU size="1.5 GiB" time=... source=device.go:272 msg="total memory" size="16.8 GiB" [GIN] 2026/05/01 - 13:29:04 | 500 | 59.999378025s | 192.168.0.38 | POST "/api/chat" ``` ### OS _No response_ ### GPU _No response_ ### CPU _No response_ ### Ollama version _No response_
GiteaMirror added the bug label 2026-05-10 06:25:52 -05:00
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#87826
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?