<html>
<head>
<style>
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
FONT-SIZE: 10pt;
FONT-FAMILY:Tahoma
}
</style>
</head>
<body class='hmmessage'>
> Does this mean, that you have single socket quad core CPU/machines, and you <br>> have 5 nodes in your cluster? What is the interconnect between the machines?<br><br>Yes, my mashines is a single socket on system board. My cpu is intel Core TM II Q6600.<br>The network between my cluster is only 1Gbit ethernet using TCP/IP.<br><br>> Now please have a look on what should be the right value for your setup for <br>> G-space/CPU cores and K-points/machines. Hopefully your 'CPU usage' gonna get <br>> better.<br><br>It's very nice of you saying so. Do your have ever run calculation with the gamma point only via<br>1Gbit LAN? What about the efficiency? What about the CPU time ? Wall time? How large the system was?<br><br>> This depends on your system again (and that quantity what you are interested <br>> in). You have to keep in mind, that doing calculations with proper K-point <br>> sampling can be several times more expensive than a gamma point only <br>> calculations.<br>> If you get good enough results with Gamma point only with a larger supercell, <br>> than it is a happy case. If your system does not have a wide-band gap (like <br>> metallic systems and so on) the Gamma point only results are just far off, <br>> and you are forced to do K-point sampling.<br><br>My most interested system is the Titanium Oxide surface with 1~3 noble metal atom (too many is more interested,<br>but DFT can't handle this. I used to calculate from atomic simulation but the force field is not accurate enough)<br>I think K-point is enough for my ~80 atoms system. But Stefano Baroni tell me 'never trust a truth, unless you can <br>give by yourself a correct answer to a legitimate question'. I think this is quite right. So I'll try 2X2X1 and make a <br>energy and geometry comparison with the gamma point only.<br><br>> If you do Gamma point only, than you can exploit the parallelization over <br>> G-space only. If you wanna use the G-space parallelization over the whole <br>> cluster than as i mentioned you need a fast interconnect (infiniband, <br>> myrinet, SCI, quadrics and so on). Gigabit LAN is just too slow for that <br>> (except for insanely large supercells). <br><br>How could I exploit the parallelization over G-space? Only by use the the proper number of CPU? no pools can be used?<br>What's the exact meaning of Gigabit LAN is just too slow for that (except for insanely large supercells)? Do you mean I could<br>calculate with relative high efficiency by using large systems? How many atoms do you call it 'large' supercells?<br><br>It's very kind of you to share your experience of PWscf. Thank you so much for the detail discussion.<br><br>By the way, are you a German? I like German football team. Let's expect the final of Euro 2008 and the German's excellent<br>show, even though, Ballack maybe absence. Come on~ <br><br>best wishes.<br><br><font size="4"><span style="font-family: Lucida Handwriting,Cursive;">Vega Lew</span></font><br>PH.D Candidate in Chemical Engineering<br>
State Key Laboratory of Materials-oriented Chemical Engineering<br>College of Chemistry and Chemical Engineering<br>Nanjing University of Technology, 210009, Nanjing, Jiangsu, China<br><br><br /><hr />Get news, entertainment and everything you care about at Live.com. <a href='http://www.live.com/getstarted.aspx ' target='_new'>Check it out!</a></body>
</html>