New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
CUDA 11.2 Support #4443
Comments
cudaMallocAsync and cudaFreeAsync in CUDA 11.2: |
File name of some libraries have been changed between 11.1 and 11.2 (e.g. |
From https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#cuda-general-new-features:
Looks promising! |
Other potentially useful additions or notes:
|
Stream Ordered Memory Allocator |
In addition to texture reference, I saw several additional deprecation warnings from cuSPARSE and Profiler with CUDA 11.2 (the log is interleaving due to the parallel build). Also, the built-in Thrust/CUB complains that output
|
Hi, Current if we want a working cupy with CUDA 11.2, we need compile from scratch. The compilation of cupy is very slow. Would it be possible to make a minimal pipy release of cupy_cuda112 sooner? This will help the community try/adopt CUDA 11.2 faster. From this threads, I do not see any big problems. All issues are to support new CUDA 11.2 features or warnings that can be ignore (but aren't ideal to have). Did I miss something? |
@nouiz CUDA provides a strong guarantee for forward compatibility since CUDA 11.0, meaning that applications built on CUDA 11.x can run on 11.y with y>=x, so in theory you could do |
Good point. Thanks. |
Hi, The above exception was the direct cause of the following exception: Traceback (most recent call last): If you are using wheel distribution (cupy-cudaXX), make sure that the version of CuPy you installed matches with the version of CUDA on your host. If you are building CuPy from source, please check your environment, uninstall CuPy and reinstall it with: Check the Installation Guide for details: original error: libcublas.so.11: cannot open shared object file: No such file or directory please let me know if there is a fix that could be applied here. Thank you! |
@eghbalhosseini you currently must compile cupy from source to have 11.2 support; the e.g. cupy-cuda111 wheel has the library version compiled in. |
@takagi Could you please work on this? |
btw CUDA 11.2.1 is out... |
Tasks:
cupy/core/include/cupy/_cuda/
, then updatecupy/core/core.pyx
to use it. Support CUDA 11.2 #4702cupyx/tools/install_library.py
Add cudnn and cutensor for CUDA 11.2 ininstall_library
#4680install_library
#4703README.md
. Add CUDA 11.2 to docs #4705Additional tasks:
The text was updated successfully, but these errors were encountered: