Re: [AMBER] amber.cuda performance on Nvidia S2050?

From: Sasha Buzko <obuzko.ucla.edu>
Date: Mon, 07 Jun 2010 11:10:32 -0700

Thanks, Scott.

As for S2050 performance, I'm just checking ahead - we don't have one of
those yet.

Sasha




Scott Le Grand wrote:
> I wouldn't expect it, but are you seeing something?
>
> Also, I'll get to the IWRAP issue sometime later today. I suspect the fix is but a few lines of code but I'm neckdeep in multi-GPU code right now. And in that case, the more independent bandwidth between GPUs the better...
>
>
>
> -----Original Message-----
> From: amber-bounces.ambermd.org [mailto:amber-bounces.ambermd.org] On Behalf Of Sasha Buzko
> Sent: Monday, June 07, 2010 09:24
> To: AMBER Mailing List
> Subject: [AMBER] amber.cuda performance on Nvidia S2050?
>
> Hi all,
> has anyone had any experience with amber.cuda using Nvidia S2050
> rack-mounted units? I wonder if there is any loss in performance due to
> each pair of Tesla cards sharing a PCI-E slot on the motherboard.
> It didn't seem to be a problem with the previous generation cards, but
> with faster Fermi cards the bottleneck could possibly shift to the
> interface. Has anyone seen such a drop in individual performance when
> all cards in S2050 execute at the same time?
> Thanks for any info
>
> Sasha
>
>
> Ross Walker wrote:
>
>> Hi Dean,
>>
>>
>>
>>> I have a quick question:
>>>
>>> I'm just curious as to the theoretical explanation for why the box
>>> clearance
>>> should be greater than your cutoff?
>>>
>>>
>> The requirement, due to the fact that AMBER implements the minimum image
>> convention within the code. In that it does not have image atoms and expects
>> each atom to only interact with itself once in the direct space calculation
>> means that the minimum box dimension must be at least the cutoff size.
>>
>>
>>
>>> If I did 10A clearance for my protein in a water box, and used 10A for
>>> my cutoff, what would be the deleterious results?
>>>
>>>
>> In this case your box is almost certainly large enough even allowing for the
>> density increasing during NPT runs. The solvatebox command in leap adds
>> sufficient water so that no single solute atom is within the buffer distance
>> of the box edge. In your case if your protein was 10 angstroms by 10
>> angstroms by 10 angstroms (pretty small!) then you would have added a
>> minimum of 10 angstroms of water on each side of your solute. This gives you
>> a box with a minimum size of 30 angstroms^3. This is 3 times larger than the
>> minimum size implied by the minimum image convention with a 10 angstrom
>> cutoff so you have significant room to spare.
>>
>> Note, the code continually checks that the minimum image convention is not
>> violated and will quit it your box becomes too small rather than running and
>> generating incorrect answers.
>>
>> I hope this helps.
>>
>> All the best
>> Ross
>>
>> /\
>> \/
>> |\oss Walker
>>
>> | Assistant Research Professor |
>> | San Diego Supercomputer Center |
>> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
>> | http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
>>
>> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
>> be read every day, and should not be used for urgent or sensitive issues.
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> 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
> -----------------------------------------------------------------------------------
> This email message is for the sole use of the intended recipient(s) and may contain
> confidential information. Any unauthorized review, use, disclosure or distribution
> is prohibited. If you are not the intended recipient, please contact the sender by
> reply email and destroy all copies of the original message.
> -----------------------------------------------------------------------------------
>
> _______________________________________________
> 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 Mon Jun 07 2010 - 11:30:03 PDT
Custom Search