【nvidia-B200】生产报错common.h:14:10: fatal error: mpi.h: No such file or directory

目录

一、解决 “nccl-tests 目录已存在” 问题

二、解决 “mpi.h: No such file or directory” 问题（核心错误）

---

报错日志：

# cat nccl.sh cd /home/test
git clone https://github.com/NVIDIA/nccl-tests.git
cd nccl-tests  && make MPI=1 MPI_HOME=/usr/lib/x86_64-linux-gnu/openmpi/
#./build/all_reduce_perf -b 8 -e 512M -f 2 -g 8
/home/test/nccl-tests/build/all_reduce_perf -b 8 -e 512M -f 2 -g 8
chown -R test:test /home/test/nccl-tests
root@b1-g25:/home/test# bash nccl.sh 
fatal: destination path 'nccl-tests' already exists and is not an empty directory.
make -C src build BUILDDIR=/home/test/nccl-tests/build
make[1]: Entering directory '/home/test/nccl-tests/src'
Compiling  all_reduce.cu                       > /home/test/nccl-tests/build/all_reduce.o
nvcc warning : Support for offline compilation for architectu