njw1123

Results 5 comments of njw1123

symbol lookup error: /usr/local/lib/ucx/libuct_cuda.so.0: undefined symbol: nvmlDeviceGetGpuFabricInfo

> If you build UCX from sources, you can also try to use this patch [#10680](https://github.com/openucx/ucx/pull/10680). It should work without changing the driver. I have already tried this version, but...

symbol lookup error: /usr/local/lib/ucx/libuct_cuda.so.0: undefined symbol: nvmlDeviceGetGpuFabricInfo

Is there any other solution besides reinstalling the driver?

Does DeepEP use SMs for its normal and low-latency kernels?

> > Does that mean DeepSeek training uses normal kernel and the 20 SMs are used because of NVLink forwarding? > > Yes. Normal kernels are designed for training and...

Does DeepEP use SMs for its normal and low-latency kernels?

> > Does this mean that whether SMs are used is independent of whether IBRC or IBGDA is used — that is, pure IBRC or IBGDA communication does not occupy...

[Question]: Does it support one-sided GPU-to-GPU communication across nodes using GPUDirect RDMA?

> Yes, it is supported Thanks for your reply