Leo Fang
> In runtime we rely on several aspects of CTK and their discovery strategy is a bit cluttered: > > * CTK shared libraries > * Fully rely on ``cuda-pathfinder``'s...
As a follow-up of #9444, we should set up CI to test pure-wheel environments (no CTK pre-installed in either the VM or the container), in order to ensure the pathfinder...
We added Jitify (https://github.com/cupy/cupy/pull/4228) 5 years ago because we had to use NVRTC to compile Thrust/CUB code, and Thrust/CUB included a ton of system headers that choked NVRTC and had...
Close #5349. - Clean up outdated test skip conditions with the upcoming (v14) CUDA & ROCm support matrices in mind - Add an adaptor for hooking a `cuda.core.{experimental.}DeviceMemoryResource` with `MemoryAsyncPool`...
Building CuPy from source can be painful. pixi will make it a lot easier.
For a long time, CuPy being a well-known brand leads to the muscle memory shared among Python users trying to `pip install cupy`. It is rare that a newcomer would...
> We should probably document the limitation of `@cupy.fuse`. I'll create a separate tracking issue. _Originally posted by @leofang in [#8689](https://github.com/cupy/cupy/issues/8689#issuecomment-3453957818)_
ACH (Accelerated Computing Hub, https://github.com/NVIDIA/accelerated-computing-hub) is where we at NVIDIA publish CUDA C++ and Python tutorials, workshop and course materials. @brycelelbach @nv-kriehl and others have been leading many well-received workshops,...
For new Python projects, we encourage them to just use `cuda.core..Stream`. For existing Python projects, transitioning to `cuda.core` may or may not be desired. As a result, we encourage projects...
Close #9358.