Re: iDataPlex

From: Francesco Pietra (chiendarret_at_gmail.com)
Date: Tue Mar 18 2014 - 04:03:29 CDT

Hello Andrew:

I waited too long for help by namd developers about restarting t-remd with
colvar. No help came and I am now short of time , with deadline of my grant
on 24 April. Cristiano has therefore proposed to move the unused hours, if
any on April 22, on my request, to either eurora or plx. OK. Actually,
Alessandro Grottesi tried to run namd t-remd on eurora, without success.
However, I don't know if he "activated" the GPUs correctly.

So far, I carried out t-remd on fermi with my peptides, under either GB
conditions (longdebug 64 nodes) or under explicit water (to have exchange
between the replicas, 128 replicas were needed, which required 256 nodes).
I got some folding, not complete, with 24hr jobs in either case. As
restarting with colvars failed, I selected a conformation, restarting from
scratch from that. Which is not the best.

The shorter peptide was handled best. I completed the folding with namd2.9
on my single-node machine with MacCammon accelMD. (I have GPUs too, but
accelMD does not run on GPUs with namd cuda, and I have no amber version
for GPU to try)

The longer peptide is still under modeling, apparently much more difficult
for accelMD. I do not expect a Nobel from this work, but it will be quite
significant to clarify the structure, where experimental science failed,
for a protein of great importance.

All that in view of choosing between PLX and EURORA. Probably PLX is less
expensive.

Thanks
francesco pietra

On Tue, Mar 18, 2014 at 9:15 AM, Andrew Emerson <a.emerson_at_cineca.it> wrote:

> Dear Francesco
>
> I assume you are referring to Cineca's PLX system?
> I can help you since I have some done benchmarks there, getting a speed up
> of about 3x compared to non-accelerated code.
> You can contact me off list.
>
> best regards
> Andrew
> On 17/03/2014 19:12, Francesco Pietra wrote:
>
>> Hello;
>> Any experience with running namd on IBM
>> iDataPlex DX360M3? I am particularly interested for namd CUDA.
>>
>> Thanks
>>
>> francesco pietra
>>
>
>
> --
> Dr. Andrew Emerson
> Supercomputing Applications and Innovation Dept (SCAI)
> CINECA
> Via Magnanelli, 6/3 40033 Casalecchio di Reno (BO), Italy.
> a.emerson_at_cineca.it
> 0039-0516171653
> skype: aem01966
> http://hpc.cineca.it
>
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:14 CST