General

Profile

Jonathan Vincent

  • Registered on: 12/08/2017
  • Last connection: 10/18/2019

Issues

Activity

09/07/2019

01:14 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
To answer the more detailed questions
* No the current gerrit code does not support all configurations. I will fix...

09/05/2019

12:33 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
The spreadsheet is at https://docs.google.com/spreadsheets/d/1Cxu-4KWu8YZDDxE8gjLB7ln3-65aYrBqa0mHjMsWLCE/edit?usp=sh...

09/04/2019

03:37 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
There is not an inherent need for 4 threads per atom.
There is an issue with the data format, where it interleave...

08/21/2019

06:08 PM GROMACS Task #3031: evaluate the impact of particle order on PME
Ran this with the water boxes using either 4 tMPI ranks (3 PP and 1 PME) or 2 tMPI ranks (1 PP and 1 PME). Uisng a se...
05:10 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
OK so I ran some on the RTX 2080 as well.
Looking at the total time for the spline_and_spread kernel plus the gath...

08/16/2019

12:42 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
One solution would be to run with 16 threads, and fall back to the save/reload for small sizes.
From my code it w...

08/15/2019

12:11 AM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
Ok those is a good points.
Yes there is DD on the PP side as we have 3 PP ranks and 1 PME rank.
Re-ran with 16...
04:12 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
Water boxes
Command gmx mdrun -ntmpi 4 -ntomp 10 -pme gpu -nb gpu -pin on -nsteps 1000 -v -npme 1 -notunepme
V100...

08/14/2019

03:02 PM GROMACS Task #2792: Improvement of PME gather and spread CUDA kernels
To pass the unit tests it is necessary to write out the theta and grid lines information. Similarly for the gather it...

07/10/2019

11:50 AM GROMACS Task #3031: evaluate the impact of particle order on PME
What is the change/patch that implements DD sorting?
I should look at this as well.

Also available in: Atom