eqy
November 9, 2021, 6:07pm
5
Ah, sorry, I misread your question. I think the difference might be that CUDA 11 will support more GPU architectures; there are corresponding kernels for the newer architectures with the newer CUDA version e.g.,
NIT: TORCH_CUDA_ARCH_LIST will be used while building from source and won’t change the shipped compute capabilities in the binaries, which you can get via print(torch.cuda.get_arch_list()).
That’s expected, since Ampere GPUs need CUDA>=11.0
No, the 3070 uses sm_86, which is natively supported in CUDA>=11.1 and is binary compatible to sm_80, so would already work in CUDA=11.0.
In any case, good to hear it’s working now
As far as the file split, I think that might be just an artifact of a tweak to the build process, but I’m not very knowledgeable on the details here.
option(BUILD_JNI "Build JNI bindings" OFF)
option(BUILD_MOBILE_AUTOGRAD "Build autograd function in mobile build (in development)" OFF)
cmake_dependent_option(
INSTALL_TEST "Install test binaries if BUILD_TEST is on" ON
"BUILD_TEST" OFF)
option(USE_CPP_CODE_COVERAGE "Compile C/C++ with code coverage flags" OFF)
option(COLORIZE_OUTPUT "Colorize output during compilation" ON)
option(USE_ASAN "Use Address Sanitizer" OFF)
option(USE_TSAN "Use Thread Sanitizer" OFF)
option(USE_CUDA "Use CUDA" ON)
# BUILD_SPLIT_CUDA must also be exported as an environment variable before building, with
# `export BUILD_SPLIT_CUDA=1` because cpp_extension.py can only work properly if this variable
# also exists in the environment.
# This option is incompatible with CUDA_SEPARABLE_COMPILATION.
cmake_dependent_option(
BUILD_SPLIT_CUDA "Split torch_cuda library into torch_cuda_cu and torch_cuda_cpp" OFF
"USE_CUDA AND NOT CUDA_SEPARABLE_COMPILATION" OFF)
option(USE_FAST_NVCC "Use parallel NVCC build" OFF)
option(USE_ROCM "Use ROCm" ON)
option(CAFFE2_STATIC_LINK_CUDA "Statically link CUDA libraries" OFF)
cmake_dependent_option(