The Best VGA For Gromacs Software

[BHZ-GTR]

Hi Friends

I am Need Help For The Best Video Graphics , The software used in the pharmaceutical field.

This software does not support Open Cl And I offer my university To Use R9 295X/ X2 But This is Software not support Open Cl


I did a little search , The software works faster with nvidia graphics card .

I am Choose : 2* Intel XEON E5-2697 V3 + 1* GTX 980 + C612 Chipset

and 1* Intel XEON E5-2695 V3 + X99 + 4*980 GTX For This Software Supported Multigpus

The university needs very fast processing .

What are your suggestions Friends

 

[sm625]

There is no guarantee that your particular simulation will be able to make the best use of the potential GPU compute power. But chances are, the 4*980 configuration would be significantly faster.

 

[BHZ-GTR]

Thank You

In your opinion , We use the Geforce VGA Or Quadro VGA For Fast Processing .

And We use a single processor 2* Xeon With QuadRo Better or 2* XEON With Geforce GTX 980 .

 

[Ken g6]

GROMACS - that's what Folding@Home uses, right? Are you doing single-precision or double-precision?

Location: IRAN

I'm surprised you can get any of this hardware there!

 

[BHZ-GTR]

Hi My Friend Ken g6

how Are you , Yes i am Live In IRAN .

Why are you surprised ، Yes In my country any existing piece of Hardware.

In your opinion I choose which hardware for quick processing

 

[Vesku]

Thank You

In your opinion , We use the Geforce VGA Or Quadro VGA For Fast Processing .

And We use a single processor 2* Xeon With QuadRo Better or 2* XEON With Geforce GTX 980 .

I think you are better off asking the Gromacs community. The relationship of CPU and GPU in Gromacs is that they work together at the same time, unlike many GPU Compute tasks where the GPU and CPU do their own separate tasks. GTX 980 looks to be an alright choice, only mention of Quadro or Tesla benefit is to better fit in a rack server.

http://www.gromacs.org/GPU_acceleration

If I'm understanding this page I think 2 CPU + 2 GPU may be the best configuration. The mailing list may be able to help you better especially if you are stuck choosing between 2 CPU + 1 GPU or 1 CPU + 4 GPU and are unable to consider 2 CPU + 2 GPU.

http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List

 

[BHZ-GTR]

Thank You Very Much Vesku.

Yes , I Know These associations were .

But The site ANANDTECH is considered the best place for advice

I'm confused , The Best Choose For The Processing

Tier 1 : 2* XEON E5-2697 V3 + 1*980 GTX + C612

Tier 2 : 1* XEON E5-2695 V3 + 4*980 GTX + X99

Tier 3 : I use the network For Dividing the processing

 

[Vesku]

Again, IF I've understood their GPU Acceleration page the best option would be:

2*Xeon E5-2697 V3 + 2*980 GTX + C612 - Assuming it can support a second 980

If that is not possible I think the 2*Xeon + 1*980 is likely better than 1*Xeon + 4*980.

That is if I've understood you and you must choose ONE configuration. The Gromacs GPU performance page has this note at the end:

There are some extra tricks to improve performance on nodes with lots of cores (e.g. Opteron machines with 32-64 cores and just 1-2 GPUs) - post to the mailing list if you are having problems!

http://www.gromacs.org/About_Gromacs/Benchmarks

Might help to confirm my configuration suggestion.

 

[BHZ-GTR]

Vesku My Friend , Which graphics card is better in your opinion?

Quadro OR Geforce < To take advantage of this software

 

[mv2devnull]

If you did look at the GPU_acceleration link in Vesku's posts, you did saw some benchmark results.

If you do google search with "site:mailman-1.sys.kth.se gmx-users gtx tesla" or similar keywords, you will find posts by the developers of Gromacs and the answer is usually:

It depends.

Overall, GTX has nice performance/price, but if you will build a TOP-500 cluster, then there are other aspects, like cooling.

More importantly, the size and nature of the system (that you calculate) does matter a lot. What is the ratio of different subtasks? Some fit to CPU, some to GPU. You might get "optimal" setup for one run, but will any other run have exact same preferences?

 

[ggadrian]

I you can't go with dual xeon plus 2 GTX980 you might look at 2 GTX970.

But budget aside, it just looks like a job for a dual xeon and many tesla cards, specially if it's for a university and not for you at home.

 

[Vesku]

GeForce will provide similar performance to equivalent Quadro and Tesla cards, this software is apparently FP32. If you are budget focused I'd stick with 980s. As ggadrian mentioned the typical well funded, not budget focused, University setup seems to be to run it on dense nodes of dual Xeon + 4-6 Tesla cards.

 

[BHZ-GTR]

Hello To Everyone

I have a question of your friends.

We decided to use > 2* XEON E5 2695 V3 + 4 * 980 GTX And C612 For processing the application

Question : Do we use memory's Feature {ECC} Is Supported And how the most influential in the process&#1567; Do not delay in processing makes .

This software is highly dependent on the bandwidth

Memory ECC Yes Or No ? Why ?

Thank You Guys

 

[at80eighty]

my understanding of ECC is you have a tradeoff between failure rate (ECC is more reliable) & speed (ECC is just slightly slower)

the guys at http://forums.anandtech.com/forumdisplay.php?f=47 might give you a better answer though