I am a user of VASP. And I noticed that a comparation of performances between 2K80 and 2P100 on NVIDIA website(As shown on http://www.nvidia.com/object/tesla-p100.html ). I compiled vasp and tested performances between 2K80 and 2P100 on my server. The data I’ve got showed that performance of the same workload on 2P100 was lower than 2K80. This is different form yours.
The performance of VASP I’ve got is following:
2 × K80 elapsed : 242 seconds
2 × P100 elapsed : 319 seconds
I really want to know why there is such a difference and how do you compile is and what examples are you testing!
Except your response.