Linux搭建Linpack环境测试GPU性能

一、相关软件安装：

1、GPU驱动：https://blog.csdn.net/liu_feng_zi_/article/details/107232252

2、CUDA：https://blog.csdn.net/liu_feng_zi_/article/details/107234271

3、MKL数学库：https://blog.csdn.net/liu_feng_zi_/article/details/107427102

4、OpenMPI：https://blog.csdn.net/liu_feng_zi_/article/details/107429347

5、平台：CentOS Linux release 7.6.1810 (Core) 

6、编译器：系统自带的GNU编译器

二、下载HPL包：

https://developer.nvidia.com/computeworks-developer-exclusive-downloads

三、环境搭建：

1、将压缩包上传到服务器并解压

tar zxf hpl-2.0_FERMI_v15.tgz 

 2、进入解压后的文件夹，修改Make.CUDA的参数

TOPdir = /Linpack/hpl-2.0_FERMI_v15 #hpl所在的路径目录
..............................................................

MPdir  = /usr/local/openmpi         #openmpi的安装目录
MPinc = -I$(MPdir)/include          
MPlib = $(MPdir)/lib/libmpi.so
..............................................................

LAdir 	=/opt/intel/mkl/lib/intel64 #MKL库文件目录

..............................................................

CC   =/usr/local/openmpi/bin/mpicc

..............................................................

3、编译

make arch= CUDA clean_arch_all
make arch= CUDA

4、完成

/hpl/bin/CUDA目录下生成两个文件：HPL.dat和xhpl