【异常问题】nvcc fatal : Unsupported gpu architecture ‘compute_86‘ 解决方案_nvcc fatal : unsupported gpu architecture 'compute

异常问题：nvcc fatal : Unsupported gpu architecture ‘compute_86’ 解决方案

（1）报错信息

/usr/local/cuda/bin/nvcc -DWITH_CUDA -I/home/shanshan/anaconda3/envs/SOLOv2/lib/python3.7/site-packages/torch/include -
I/home/shanshan/anaconda3/envs/SOLOv2/lib/python3.7/site-packages/torch/include/torch/csrc/api/include -
I/home/shanshan/anaconda3/envs/SOLOv2/lib/python3.7/site-packages/torch/include/TH -
I/home/shanshan/anaconda3/envs/SOLOv2/lib/python3.7/site-packages/torch/include/THC -I/usr/local/cuda/include -
I/home/shanshan/anaconda3/envs/SOLOv2/include/python3.7m -c mmdet/ops/nms/src/nms_kernel.cu -o build/temp.linux-x86_64-
3.7/mmdet/ops/nms/src/nms_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -
D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options '-fPIC' -D__CUDA_NO_HALF_OPERATORS__ -
D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -DTORCH_API_INCLUDE_EXTENSION_H -
DTORCH_EXTENSION_NAME=nms_cuda -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_86,code=sm_86 -std=c++14

nvcc fatal   : Unsupported gpu architecture 'compute_86'
error: command '/usr/local/cuda/bin/nvcc' failed with exit status 1
1
2
3
4
5
6
7
8
9
10
11
12

（2）系统环境

• GPU： RTX 3080ti
• Linux： Ubuntu 18.04
• Pytorch: 1.7.0
• Python: 3.7.9
• CUDA version: V11.0.194
• GCC version: 5.2.0

（3）原因分析

应该是硬件能够支持的算力比较高，能达到8.6，但是cuda11.0支持不了这么高的算力。

（4）解决方案

• 设置环境变量，降低算力要求

# 清除TORCH_CUDA_ARCH_LIST
unset TORCH_CUDA_ARCH_LIST
# 设置cuda算力为8.0
export TORCH_CUDA_ARCH_LIST="8.0"
1
2
3
4