<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html><head><title>Re[2]: [Beowulf] Nehalem memory configs</title>
<META http-equiv=Content-Type content="text/html; charset=utf-8">
<meta http-equiv="Content-Style-Type" content="text/css">
<style type="text/css"><!--
body {
margin: 5px 5px 5px 5px;
background-color: #ffffff;
}
/* ========== Text Styles ========== */
hr { color: #000000}
body, table /* Normal text */
{
font-size: 9pt;
font-family: 'Courier New';
font-style: normal;
font-weight: normal;
color: #000000;
text-decoration: none;
}
span.rvts1 /* Heading */
{
font-size: 10pt;
font-family: 'Arial';
font-weight: bold;
color: #0000ff;
}
span.rvts2 /* Subheading */
{
font-size: 10pt;
font-family: 'Arial';
font-weight: bold;
color: #000080;
}
span.rvts3 /* Keywords */
{
font-size: 10pt;
font-family: 'Arial';
font-style: italic;
color: #800000;
}
a.rvts4, span.rvts4 /* Jump 1 */
{
font-size: 10pt;
font-family: 'Arial';
color: #008000;
text-decoration: underline;
}
a.rvts5, span.rvts5 /* Jump 2 */
{
font-size: 10pt;
font-family: 'Arial';
color: #008000;
text-decoration: underline;
}
span.rvts6
{
font-size: 12pt;
font-family: 'times new roman';
}
/* ========== Para Styles ========== */
p,ul,ol /* Paragraph Style */
{
text-align: left;
text-indent: 0px;
padding: 0px 0px 0px 0px;
margin: 0px 0px 0px 0px;
}
.rvps1 /* Centered */
{
text-align: center;
}
--></style>
</head>
<body>
<p>Hallo richard,</p>
<p><br></p>
<p>Freitag, 10. April 2009, meintest Du:</p>
<p><br></p>
<div><table border=0 cellpadding=1 cellspacing=2 style="background-color: #ffffff;">
<tr valign=top>
<td width=2 style="background-color: #0000ff;"><br>
</td>
<td width=1683>
<p><br></p>
<p><span class=rvts6>> Kilian CAVALOTTI wrote:</span></p>
<p><span class=rvts6>>> On Wednesday 08 April 2009 19:48:30 Joe Landman wrote:</span></p>
<p><span class=rvts6>>>> As an FYI, Beowulf veteran Jeff Layton wrote up a nice article on</span></p>
<p><span class=rvts6>>>> memory</span></p>
<p><span class=rvts6>>>> configuration issues for Nehalem (I had seen some discussion on this</span></p>
<p><span class=rvts6>>>> previously).</span></p>
<p><span class=rvts6>>>></span></p>
<p><span class=rvts6>>>> Link is here:</span></p>
<p><span class=rvts6>>>> http://www.delltechcenter.com/page/04-08-2009+-+Nehalem+and+Memory+Configur</span></p>
<p><span class=rvts6>>>> ations</span></p>
<p><span class=rvts6>>></span></p>
<p><span class=rvts6>>> That's a great write-up, gathering all the useful information in one</span></p>
<p><span class=rvts6>>> place. A must read. Thanks Jeff!</span></p>
<p><span class=rvts6>It is a good write up, very good ... but there looks to be an error under "Memory Bandwidth</span></p>
<p><span class=rvts6>Performance" ... first, for socket-local-memory, I am not sure of the relevance of</span></p>
<p><span class=rvts6>the QPI GT rate. I would calculate the maximum rate to socket-local-memory</span></p>
<p><span class=rvts6>with 1 DIMM per channel at:</span></p>
<p><span class=rvts6> </span></p>
<p><span class=rvts6>1.333 GHz x 3 Channels x 8 bytes = 31.992 GBytes/sec </span></p>
<p><span class=rvts6> </span></p>
<p><span class=rvts6>not 35 GByters/sec. Maybe I am doing something wrong. ยด</span></p>
</td>
</tr>
</table>
</div>
<p><br></p>
<p><br></p>
<p>I understood the 35 GB/s to be the transfer rate that you reach when accessing Memory with 8 Cores. So it involves 6 Channels (2 CPUs, 3 channels each). ~64 GB/s (theory) - 35 GB/s (real life). Makes sense to me. </p>
<p><br></p>
<p>Jan </p>
</body></html>