NAMD 3.0

[bhavranek]

Hello,
In your nature protocol paper did you use NAMD 3.0 GPU version to run the simulations? Did you use "CUDASOAintegrate on" in the input file which is required for high performance? The reason I ask is because in the SI it looks like you used 32 CPU cores and 2 GTX 2080Ti. However, when using "CUDASOAintegrate on" with NAMD 3.0 I get the following:

REPLICA 1 FATAL ERROR: CUDASOA integration is incompatible with the following options:
minimization; pressure profiling; Berendsen pressure;
multigrator; Lowe-Andersen; fixed atoms; GBIS; LCPO;
zero momentum; TCL forces; Colvars;
temperature coupling, rescaling, or reassignment;
water models other than TIP3; lonepairs or Drude.

How was your NAMD 3.0 able to run using multiple GPUs with colvars when it says it is not yet supported? Or do I have to compile with the NAMD git version of Colvars to get that functionality? Or did you not run with "CUDASOAintegrate on" ?

[fhh2626]

Hi @bhavranek ,

The new integrator is not compatible with Colvars. I am afraid you must remove "CUDASOAintegrate on" in your conf file.

Good luck,
Haohao

[bhavranek]

Do you have a sample SLURM or PBS script so I can see how you have been allocating GPU and CPU in your runs? Similar to what was done in your Nature Protocol paper using 32 CPUs and 2 GPUs.