<html><head></head><body><div style="font-family: Verdana;font-size: 12.0px;"><div>
<div>Dear Dr. Starlight,</div>
<div> </div>
<div>Dont know the answere to all, but funny I was looking at performance on varied others web sites. I use a core i7 970, but it seems their newest chip is almost the same as the i7 in performance (thier newer chips dont scale past 12 cores for some internal chip based design, they put out a 60 core chip for 1K and it doesnt even reach 1 terflop), and the newer GPU's suffer from board IO limits due the PCIe being maxed out.</div>
<div> </div>
<div>As far as boards, home PC (im in Switzerland so have been mind distroyed as far as prices), there's a 4 CPU 10 GPU tyan board for 4K US (if you dig enough maybe 3,500) which with ram, chips and a single Nvidia would run up to 5-6 K US$. The i7 970's can be baught for 30-50 US$ a pop now if you look becuase intel overshot its abilities and gave off the rights to several bulk manufacturers (all LG 1366 socket chips even i5's) as they add more pins (2100 or something), but you have to spend a day or two on the internet looking through sellers abroad. The Nvida's are the pricy thing, however from what I have read, you dont get much more from their latest as well as mentioned, so 2 more years waiting, or something like I mentioned with a 2 year old Nvidia or two, would be the best especially for research purposes. Otherwise a single 300 US board with 4 usable PCIe's (ASUS arnt bad), ram, 1 i7 970, and 1 Nvida from 2 years back gives you around 1.3 terflops, with their lates 1.4-1.5 teraflops (integrated use of chip alone). The i7 I have I have got around 350 Gflops for some things with 8 threads. I use radeons though, although this requires user programming, and if you use ther FFT libraries, you have to sign your first child over and loose any freeware redistribution rights in any way, otherwise their supposedly supposed to have all the same CUDA like libraries by the end of the summer (Portlandgroup , openCL), but I heard the same for 2 years now.</div>
<div> </div>
<div>Sincerely,</div>
<div> </div>
<div>Stephan Watkins</div>
<div>
<div name="quote" style="margin:10px 5px 5px 10px; padding: 10px 0 10px 10px; border-left:2px solid #C3D9E5; word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;">
<div style="margin:0 0 10px 0;"><b>Gesendet:</b> Freitag, 24. Mai 2013 um 13:17 Uhr<br/>
<b>Von:</b> "James Starlight" <jmsstarlight@gmail.com><br/>
<b>An:</b> "Discussion list for GROMACS users" <gmx-users@gromacs.org><br/>
<b>Betreff:</b> [gmx-users] GPU-based workstation</div>
<div name="quoted-content">Dear Gromacs Users!<br/>
<br/>
<br/>
I'd like to build new workstation for performing simulation on GPU with<br/>
Gromacs 4.6 native cuda support.<br/>
Recently I've used such setup with Core i5 cpu and nvidia 670 GTX video<br/>
and obtain good performance ( ~ 20 ns\day for typical 60.000 atom system<br/>
with SD integrator)<br/>
<br/>
<br/>
Now I'd like to build multi-gpu wokstation.<br/>
<br/>
My question - How much GPU would give me best performance on the typical<br/>
home-like workstation. What algorithm of Ncidia GPU integration should I<br/>
use (e.g SLI etc) ?<br/>
<br/>
<br/>
Thanks for help,<br/>
<br/>
<br/>
James<br/>
--<br/>
gmx-users mailing list gmx-users@gromacs.org<br/>
<a href="http://lists.gromacs.org/mailman/listinfo/gmx-users" target="_blank">http://lists.gromacs.org/mailman/listinfo/gmx-users</a><br/>
* Please search the archive at <a href="http://www.gromacs.org/Support/Mailing_Lists/Search" target="_blank">http://www.gromacs.org/Support/Mailing_Lists/Search</a> before posting!<br/>
* Please don't post (un)subscribe requests to the list. Use the<br/>
www interface or send it to gmx-users-request@gromacs.org.<br/>
* Can't post? Read <a href="http://www.gromacs.org/Support/Mailing_Lists" target="_blank">http://www.gromacs.org/Support/Mailing_Lists</a></div>
</div>
</div>
</div></div></body></html>