github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 17:49:24 -05:00
Code Issues 14.6k Packages Projects Releases 122 Wiki Activity

[PR #1834] [MERGED] Detect very old CUDA GPUs and fall back to CPU #10699

New Issue
Closed
opened 2026-04-12 23:07:59 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-12 23:07:59 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/1834
Author: @dhiltgen
Created: 1/7/2024
Status: ✅ Merged
Merged: 1/7/2024
Merged by: @dhiltgen

Base: main ← Head: old_cuda

📝 Commits (1)

  • d74ce6b Detect very old CUDA GPUs and fall back to CPU

📊 Changes

3 files changed (+74 additions, -2 deletions)

View changed files

📝 gpu/gpu.go (+15 -1)
📝 gpu/gpu_info_cuda.c (+51 -1)
📝 gpu/gpu_info_cuda.h (+8 -0)

📄 Description

If we try to load the CUDA library on an old GPU, it panics and crashes the server. This checks the compute capability before we load the library so we can gracefully fall back to CPU mode.

Prior to version 0.1.18, the fallback behavior resulted from the subprocess runner crashing. Example from an old GPU:

ggml_init_cublas: found 1 CUDA devices:
  Device 0: NVIDIA GeForce GTX 765M, compute capability 3.0

cuBLAS error 3 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:5552
current device: 0
2024/01/06 21:48:54 llama.go:320: llama runner exited with error: exit status 1

In 0.1.18 without this fix, the server crashes with a panic due to the assert in llama.cpp.

With this fix on the same system we now detect and fallback to CPU mode:

2024/01/06 21:52:17 shim_ext_server.go:142: Dynamic LLM variants [cuda rocm]
2024/01/06 21:52:17 gpu.go:37: Detecting GPU type
2024/01/06 21:52:17 gpu.go:56: Nvidia GPU detected
2024/01/06 21:52:17 gpu.go:89: CUDA GPU is too old. Falling back to CPU mode. Compute Capability detected: 3.0
2024/01/06 21:52:17 routes.go:953: no GPU detected
...

Example output on a newer supported GPU which correctly runs on the GPU:

2024/01/06 21:55:11 shim_ext_server.go:142: Dynamic LLM variants [cuda rocm]
2024/01/06 21:55:11 gpu.go:37: Detecting GPU type
2024/01/06 21:55:11 gpu.go:56: Nvidia GPU detected
2024/01/06 21:55:11 gpu.go:86: CUDA Compute Capability detected: 7.5

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/1834 **Author:** [@dhiltgen](https://github.com/dhiltgen) **Created:** 1/7/2024 **Status:** ✅ Merged **Merged:** 1/7/2024 **Merged by:** [@dhiltgen](https://github.com/dhiltgen) **Base:** `main` ← **Head:** `old_cuda` --- ### 📝 Commits (1) - [`d74ce6b`](https://github.com/ollama/ollama/commit/d74ce6bd4f78f8a06c97bf9b24485211c48a41d8) Detect very old CUDA GPUs and fall back to CPU ### 📊 Changes **3 files changed** (+74 additions, -2 deletions) <details> <summary>View changed files</summary> 📝 `gpu/gpu.go` (+15 -1) 📝 `gpu/gpu_info_cuda.c` (+51 -1) 📝 `gpu/gpu_info_cuda.h` (+8 -0) </details> ### 📄 Description If we try to load the CUDA library on an old GPU, it panics and crashes the server. This checks the compute capability before we load the library so we can gracefully fall back to CPU mode. Prior to version 0.1.18, the fallback behavior resulted from the subprocess runner crashing. Example from an old GPU: ``` ggml_init_cublas: found 1 CUDA devices: Device 0: NVIDIA GeForce GTX 765M, compute capability 3.0 cuBLAS error 3 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:5552 current device: 0 2024/01/06 21:48:54 llama.go:320: llama runner exited with error: exit status 1 ``` In 0.1.18 without this fix, the server crashes with a panic due to the assert in llama.cpp. With this fix on the same system we now detect and fallback to CPU mode: ``` 2024/01/06 21:52:17 shim_ext_server.go:142: Dynamic LLM variants [cuda rocm] 2024/01/06 21:52:17 gpu.go:37: Detecting GPU type 2024/01/06 21:52:17 gpu.go:56: Nvidia GPU detected 2024/01/06 21:52:17 gpu.go:89: CUDA GPU is too old. Falling back to CPU mode. Compute Capability detected: 3.0 2024/01/06 21:52:17 routes.go:953: no GPU detected ... ``` Example output on a newer supported GPU which correctly runs on the GPU: ``` 2024/01/06 21:55:11 shim_ext_server.go:142: Dynamic LLM variants [cuda rocm] 2024/01/06 21:55:11 gpu.go:37: Detecting GPU type 2024/01/06 21:55:11 gpu.go:56: Nvidia GPU detected 2024/01/06 21:55:11 gpu.go:86: CUDA Compute Capability detected: 7.5 ``` --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-12 23:07:59 -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 pull-request
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#10699
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?