Re: [AMBER] Sander.mpi haults at 999 steps

From: Montalvillo, Fernando via AMBER <amber.ambermd.org>
Date: Mon, 4 Mar 2024 18:37:19 +0000

Yes, I had this option (#SBATCH --gres=gpu:1) with the gpu value set to 4 but did not work very well.

Awesome! I will send another question with that Keyword!

Thank you so much Carlos,
Fernando
________________________________
From: Carlos Simmerling <carlos.simmerling.stonybrook.edu>
Sent: Monday, March 4, 2024 11:29 AM
To: Montalvillo, Fernando <Fernando.Montalvillo.UTDallas.edu>
Cc: AMBER Mailing List <amber.ambermd.org>
Subject: Re: [AMBER] Sander.mpi haults at 999 steps

ah, I see, I didn't see that you are using QUICK.
a typical gpu request on slurm for my cluster looks like this: #SBATCH --gres=gpu:1
but I'm not sure about what quick would want, and whether MPI is supported. If you don't get others that use it here, you might post with QUICK in the email subject to make it more likely that QUICK users/developers would read it.

On Mon, Mar 4, 2024 at 12:01 PM Montalvillo, Fernando <Fernando.Montalvillo.utdallas.edu<mailto:Fernando.Montalvillo.utdallas.edu>> wrote:
Hi Carlos,

Thank for your response. Maybe I wasn't clear, but the executable is for sure cuda-GPU accelerated as reported in the manual and in the paper by Dr. Cruzeiro and et al. https://pubs.acs.org/doi/10.1021/acs.jcim.1c00169. Also, I don't fully understand how sander.quick.cuda.MPI works. Maybe not all of the simulation is run in GPUs, meaning the MM part is run with sander on CPUs and only the QM part is run with QUICK on GPUs. Anyhow, the executable can request GPUs, but I don't seem to be able to. Does anyone know what I would need to specify in my slurm file or my command line to request GPUS and not only CPUs?

Regards,
Fernando

________________________________
From: Carlos Simmerling <carlos.simmerling.stonybrook.edu<mailto:carlos.simmerling.stonybrook.edu>>
Sent: Saturday, March 2, 2024 5:56 AM
To: Montalvillo, Fernando <Fernando.Montalvillo.UTDallas.edu>; AMBER Mailing List <amber.ambermd.org<mailto:amber.ambermd.org>>
Subject: Re: [AMBER] Sander.mpi haults at 999 steps

Sander does not work on gpus.

On Fri, Mar 1, 2024, 10:42 PM Montalvillo, Fernando via AMBER <amber.ambermd.org<mailto:amber.ambermd.org>> wrote:
Yes, the error output was empty and the .out output looked fine and just told that the mpi was being run correctly.

I actually figured it out myself but did not know how to answer myself in the amber list. I tried doing the pmemd version of the tutorial instead and the output file also stopped also at the 999th step. However, the mdout file had more information. It said there was a problem regarding mbar. It turns out some of the options of mbar that were in the input file were deprecated since amber16 or 18, I don't remember what it said in the manual.

It would be helpful if the error or mdout file would print if there is deprecated options but I totally understand this might be difficult to implement in the amber suite with all the packages there are.


The reason I wanted to learn sander TI is because now that sander.quick.cuda.MPI has been released, in theory, TI using QM/MM should be feasible and more easy to set up. The problem that I am facing right now is requesting multiple GPUs with sander.quick.cuda.MPI. I have only managed to request CPUs. Does anyone have any idea how to request GPUs for parallel simulations?

Thank you for your answer,
Fernando
________________________________
From: David A Case <dacase1.gmail.com<mailto:dacase1.gmail.com>>
Sent: Friday, March 1, 2024 7:29 PM
To: Montalvillo, Fernando <Fernando.Montalvillo.UTDallas.edu>; AMBER Mailing List <amber.ambermd.org<mailto:amber.ambermd.org>>
Subject: Re: [AMBER] Sander.mpi haults at 999 steps

On Tue, Feb 27, 2024, Montalvillo, Fernando via AMBER wrote:

>I installed Amber20 with AmberTools23 (i don't know if that is a
>problem).

I'm not sure if this is relevant to your problem,, but Amber20 should not be
installed with AmberTools23. I suggest you upgrade to Amber22, which is
compatible with AmberTools23.

>
>#!/bin/bash
>#SBATCH --job-name=ti_ligands
>#SBATCH --nodes=1
>#SBATCH --ntasks=16

    Try a job with ntasks=2, just to see if anything changes

>#SBATCH --partition=torabifard
>#SBATCH --time=48:00:00
>#SBATCH --output=ti_complex_%j.out
>#SBATCH --error=ti_complex_%j.err

Did you examine the .out and .err files?

....dac

_______________________________________________
AMBER mailing list
AMBER.ambermd.org<mailto:AMBER.ambermd.org>
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 04 2024 - 11:00:02 PST
Custom Search