From: Josh Vermaas (joshua.vermaas_at_gmail.com)
Date: Mon Oct 05 2020 - 16:26:29 CDT
What I think the admins are asking about is if they need ECC memory, which
would require more expensive GPUs. NAMD will run just fine on non-ECC RAM,
and indeed that is usually the most cost effective. If the GPU memory is
faulty, you'd get frequent atoms moving too fast errors, but that just
tells you to replace the GPUs.
On Mon, Oct 5, 2020, 3:21 PM Bassam Haddad <bhaddad_at_pdx.edu> wrote:
> Hello NAMDers,
> The HPC administrator for our cluster at my university had questions for
> me regarding some of the hardware specifications for running MD
> simulations. I was unable to find a good answer for him with regard to the
> "lossiness" of simulations, and the tolerance for high-accuracy GPU-RAM.
> This was the question posed:
> - Are your processes or application "lossy," i.e. don't require strict
> accuracy? (relates to the onboard GPU RAM requirements)
> Thank you for your time on this matter.
> Best regards,
This archive was generated by hypermail 2.1.6 : Thu Dec 31 2020 - 23:17:14 CST