Performance test with Optimized LINPACK Benchmark

[choi__minseok]

Hi, I evaluated my embedded board, which has Xeon D 1.7GHz(8-cores, 12MB cache and 32GB DDR). The tool is Intel Optimized LINPACK Benchmark in MKL 2018. I expected about 400 GFLOPS but the result was about 145 GFLOPS. I would like to know the result is reasonable or not, and then how to get maximum performance. Thank you.

[Ying_H_Intel]

Hi Choi,

Could you please share some information like the OS etc.  and  how you expect about 400GFlops?

is it the Peak double floating points GLOPS 1.7G *8 * 256/64 x2(add+muti)x2(FMA)  = 217.6GFLP?   https://en.wikipedia.org/wiki/FLOPS

We had some explanation about how to use the benchmark in MKL developer guide

https://software.intel.com/en-us/mkl-linux-developer-guide-intel-optimized-linpack-benchmark-for-linux

and some MKL benchmark   https://software.intel.com/en-us/mkl/features/benchmarks  about the performance of Xeon.

Best Regards,

Ying

[choi__minseok]

HI Ying,

My OS is CentOS 7.1(64-bit) and CPU is Xeon D-1539.

I expect based on the following.

409.6 GFLOPS = 1.6GHz  x 8 cores x 8(single precision data) x 2 FMAs x 2 dispatch ports

Thank you.

[Ying_H_Intel]

Get it. The benchmark return double precision data by default

Best Regards,

Ying

[Hanneken__Lukas]

Ying H. (Intel) wrote:

We had some explanation about how to use the benchmark in MKL developer guide

https://software.intel.com/en-us/mkl-linux-developer-guide-intel-optimized-linpack-benchmark-for-linux

Hey Ying,

I know it's an old thread, but I guess here is my best shot. I want to read your guide since I'm investigating Performance Variation and the Intel HPL Linpack seems a fitting benchmark. But The links only says "Access Denied" for me.

Can I somehow get acces or find another guidance how to use it correctly? 

Regards,

Lukas