Re: [AMBER] error installing Amber12-gpu version

From: Vijay Manickam Achari <vjrajamany.yahoo.com>
Date: Wed, 18 Apr 2012 15:49:41 +0100 (BST)

Dear Jan-Philip,

Thanks for your explanation. I feel it is informative for me.

Regards

 
Vijay Manickam Achari
(Phd Student c/o Prof Rauzah Hashim)
Chemistry Department,
University of Malaya,
Malaysia
vjramana.gmail.com


________________________________
 From: Jan-Philip Gehrcke <jgehrcke.googlemail.com>
To: AMBER Mailing List <amber.ambermd.org>
Sent: Wednesday, 18 April 2012, 16:18
Subject: Re: [AMBER] error installing Amber12-gpu version
 
Vijay,

you need the following information:

- you can run jobs either on GPU *or* CPU
- 1 GPU job also consumes 1 CPU core by 100 %
- if your CPUs are Intel, then I am pretty sure you have only two
hexacores in your machine, i.e. 12 real cores (the factor 2 comes from
"Hyperthreading" -- you can read about this elsewhere).

Say you have N CPU cores in your machine. If you run G independent jobs
on the GPUs, then also G CPU cores are occupied. I would leave these G
cores entirely for the GPU jobs, i.e. all other jobs/processes on the
machine should not occupy more than N-G cores of your machine (which is
8 in your case).

In other words, while running 4 GPU jobs, don't hesitate to use 8 cores
of your machine for whatever you like (a normal pmemd simulation, for
instance).

On the other hand, if you run 4 GPU jobs and at the same time 24 other
processes on your machine each putting as much load on the CPUs as
possbile, the GPU jobs will suffer I believe.

Let's ask the developers:

A pmemd.cuda job consumes as much of 1 CPU core as it gets. If there is
no competition, it just takes 100 %. As I understand, this is an I/O
loop that is not event-based and therefore a significant part of these
100 % could actually be "wasted". How much of CPU power does a GPU job
really need? When do you expect the GPU job performance starting to suffer?


Regards,

Jan-Philip


On 04/18/2012 04:37 AM, Vijay Manickam Achari wrote:
> Thanks for Dac and Jason.
> I could install the AMBER12 GPU version successfully.
>
> Well now I have another question to ask.
> Our GPU simulation box have 2 CPUs (each with 12 cores, so total 24 cores) and 4 units of GPU.
>
> What I want to know is how to submit job with choosing lets say 12 cores of cpus and 2 units of GPU? We dont use PBS or any other job scheduler package yet. I would like to know how to submit job without scheduler?
>
> Thanks in advance.
>
>
>
> Vijay Manickam Achari
> (Phd Student c/o Prof Rauzah Hashim)
> Chemistry Department,
> University of Malaya,
> Malaysia
> vjramana.gmail.com
>
>

_______________________________________________
AMBER mailing list
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 Wed Apr 18 2012 - 08:00:03 PDT
Custom Search