On Mon, Feb 10, 2025, TUSHAR GUPTA via AMBER wrote:
>I hope this email finds you well. I have been running MD simulations using
>AMBER on a CPU and have obtained good results. However, when I shift to
>CUDA (GPU) simulations, I am encountering the following issues:
>1. The water box appears broken in VMD.
>2. Some water molecules cluster together unnaturally.
>3. The overall system visualization differs significantly from
>the CPU-generated trajectory.
>I have followed the same input parameters and input files for both CPU and
>GPU runs using examples in amber tutorial. My simulation setup includes:
>• Software: AMBER22
>• Force Field: [ff14SB + TIP3P]
>• Processor: 13th gen i9 Intel 13900k
>Cuda NVIDIA GPU RTX4090
>I would appreciate any insights or troubleshooting steps you can
>suggest. Please let me know if you need additional information or input
>files. I have attached two images one for CUDA and one from CPU.
Did you image the trajectory in cpptraj before visualizing in VMD? The
"autoimage" command is generally all you need.
It looks like you are comparing a sander CPU run with a pmemd.cuda GPU run
-- is that correct? It's probably safest to compare pmemd with pmemd.cuda,
so you can make sure that everything is the same except for the executable
name.
The GPU result looks non-periodic, whereas the CPU result looks periodic.
Can you double-check your GPU runs to make sure that periodicity is turned
on?
....regards...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Feb 10 2025 - 08:00:02 PST
