-
Notifications
You must be signed in to change notification settings - Fork 210
Description
我的显卡是 Nvidia A10 24G,通过 conda create 进行部署,执行最后一条命令 pip install turbodiffusion --no-build-isolation 后,运行报错。
conda create -n turbodiffusion python=3.12
conda activate turbodiffusion
pip install turbodiffusion --no-build-isolation
报错信息为:
DCUTLASS_DEBUG_TRACE_LEVEL=0 -DNDEBUG -Xcompiler -fPIC -DEXECMODE=0 -gencode arch=compute_120a,code=sm_120a -gencode arch=compute_90,code=sm_90 -gencode arch=compute_89,code=sm_89 -gencode arch=compute_80,code=sm_80 --threads 4 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=turbo_diffusion_ops -D_GLIBCXX_USE_CXX11_ABI=0
nvcc fatal : Unsupported gpu architecture 'compute_120a'
既然宣传的 RTX-5090 都能运行,为什么会出现 nvcc fatal : Unsupported gpu architecture 'compute_120a' 的提示呢,我修改了
./ops/cutlass/CMakeLists.txt: list(APPEND CUTLASS_NVCC_ARCHS_SUPPORTED 100 100a 120 120a 121 121a)
./ops/cutlass/customConfigs.cmake: set(PROFILER_ARCH_LIST 100a 100f 103a 120a 120f 121a)
修改为了 80,89和90, 是应该这么修改吗,如何解决上面的报错。