James Leong comments

Results 6 comments of


                                            James Leong

Segfault when preloading scuda client libscuda_12.2.so even though the server is running

Encountered the same problem, has anyone solved this problem?

Thread safety in Connection Pool - connections growing

if you want to limit the number of connections, you can set the `max_connections` parameter when initializing the ConnectionPool object.

RuntimeError was encountered while calling torch._C._cuda_init()

> Are you able to reproduce this issue without pytorch? You may have to write > > a small program and call cudaInit() yourself. > > > > I suspect...

RuntimeError was encountered while calling torch._C._cuda_init()

I'm afraid I have to say that the project doesn't support pytorch at present. @dyunwei

RuntimeError was encountered while calling torch._C._cuda_init()

First, it is necessary to analyze the CUDA-related interfaces used by PyTorch, and then implement the RPC call process for these interfaces in these two files: [manual_server.cpp](https://github.com/kevmo314/scuda/blob/main/codegen/manual_server.cpp), [manual_client.cpp](https://github.com/kevmo314/scuda/blob/main/codegen/manual_client.cpp). I previously...

RuntimeError was encountered while calling torch._C._cuda_init()

I think the GPU-dependent methods in cudnn should also be intercepted, but the methods that should be executed locally indeed should not be sent to the remote. @violinY