Re: [AMBER] Fixed the problem of AMBER20 run with pmemd.cuda with CUDA11.1 and 3080 GPU card

From: liu kai <>
Date: Sun, 3 Jan 2021 02:35:20 +0000

Dear All,

 I have recompiled the source code by the legend way again.
 This time, I only kept the sm80flags and removed the sm86flags.
 It's still work.
 According to Ross's comment, thus it may be not necessary to include the sm86flag at current moment.

 I have a small system, including 40078 atoms by NPT (ntt=3, gamma_ln=5) ensemble for 5 ns production run.
 I used intel complier2017 + Nvidia cuda 11.1 for compile.
 I extracted the data form the summary of the calculation as below:
  Nvidia 2080ti + CUDA 10.2 221.96 ns/day
  Nvidia 2080ti + CUDA 11.1 246.85 ns/day
  Nvida 3080 + CUDA 11.1 270.57 ns/day
  So, it may provide about 10% enhancement roughly.

 For the GPU performance, it think you can visit the benchmark test.


AMBER mailing list
Received on Sat Jan 02 2021 - 19:00:02 PST
Custom Search