Repository navigation
|
I used one 3090 previously, but now I have purschased two aftermarket sxm2 tesla v100 16gb with sxm2 to pci-e adapters, these by themselves seem to work just fine (in llama.cpp (when used for non-moe models and one at a time), comfy.ui, stress tests, torch model training, etc.). But in llama.cpp I have encountered two different problems.
|
Answered by
pl752
Feb 16, 2026
Replies: 1 comment
|
I have found solution to these problems: it is to force mmq pipeline usage (build with |
0 replies
Answer selected by
pl752
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I have found solution to these problems: it is to force mmq pipeline usage (build with
-DGGML_CUDA_FORCE_MMQ=1), if it didn't help for the first disable peer copy (will also disable pipeline parallelism, build with-DGGML_CUDA_NO_PEER_COPY=ON)