Add RTX 50 series GPU compatibility via PyTorch 2.7 and CUDA 12.8#369
Add RTX 50 series GPU compatibility via PyTorch 2.7 and CUDA 12.8#369wualbert wants to merge 2 commits intoNVlabs:mainfrom
Conversation
|
There are currently no Docker images that can run on RTX 50 series GPU. The one built for RTX 40 series card (see #27) does not work on RTX 50 series due to the issues addressed in this PR. |
|
Thank you! It works on RTX 5090 in WSL with docker. Except that I still need to import |
|
Hi, I have followed your instruction. However, there is an error says "No module named 'torch'" when i tried to bash build_all_conda.sh to install mycpp. I am sure theres pytorch 2.7.1 with cuda 12.8 on my 5090D PC. And the detail of bugs is listed below. Could you please let me know why and what should i do? This warning is for project developers. Use -Wno-dev to suppress it. -- Found Boost: /home/dell/miniconda3/envs/foundationpose/lib/cmake/Boost-1.85.0/BoostConfig.cmake (found version "1.85.0") found components: system program_options -- Found PythonInterp: /home/dell/miniconda3/envs/foundationpose/bin/python (found suitable version "3.9.23", minimum required is "3.8") × Getting requirements to build editable did not run successfully. note: This error originates from a subprocess, and is likely not a problem with pip. |
|
@GXYsteven Hi, have you sovled this problem? I encountered the same issue when using pytorch 2.9 and cuda 13.0. |
You need to install torch in your environment and then run without build isolation. |
This PR addresses multiple compatibility issues that prevent FoundationPose from building and running on RTX 50 series (sm_120 architecture) GPUs with modern PyTorch versions.
bundlesdf/mycuda/common.cuto fix compilation errors with PyTorch 2.7+bundlesdf/mycuda/setup.pyto meet modern PyTorch requirementsmycpp/src/app/pybind_api.cppby changing%dto%zuforsize_tvaluesrequirements.txtto use PyTorch 2.7.1 with CUDA 12.8 for RTX 50 series compatibilitymycppimport path inUtils.pyfrommycpp.build.mycpptomycppreadme.md