Leo Fang

I feel something is wrong in your container in a way that I haven't fully understood. CuPy lazy-loads all CUDA libraries so `import cupy` does not trigger the loading of...

Ah, good finding, I forgot there's the preload logic... Could you file a bug in CuPy's issue tracker? I think the preload logic needs to happen _as part of_ the...

If you also install CuPy from conda-forge, library loading should already be taken care for you. conda knows where to find shared libraries. I would like to know better what...

> Although I use a conda environment, I install my dependencies from pypi.org using pip ... This is the kind of information that would have helped us understand the issue...

This issue is actually very tricky because the modern way of exposing peer access support is again tied to the mempool/VMM APIs, such as `cuMemPoolSetAccess`/`cuMemSetAccess`, same as IPC (#103). We'll...

What's more fun: In the `test_underscore_prefix.pyx` file if we do this instead of printing `sizeof(_anon_pod1)` ```cython # distutils: language = c++ # distutils: extra_compile_args = -std=c++17 def test(): print(sizeof(obj_t)) ```...

We came up with a solution for Case 1: 1. declare `_anon_pod1` with `cdef` instead of `ctypedef` 2. get size of `_anon_pod1` with `sizeof((NULL).union_member)` instead of `sizeof(_anon_pod1)` No.2 is understandable...

cc @rwgk for vis

xref: https://github.com/NVIDIA/cuda-python/actions/runs/14048031704?pr=517

We've observed no more network issues lately! According to @ajschmidt8: > Most likely moving the V100s from RDS Lab to NVKS resolved the network issues. > The NVKS cluster is...