SpecFP and cache

[Richard Rogers]

From: Gerben Roest <g.roest at linvision.com>


>My question is: Does anyone have experiences with old and new PIII's (that
>is, old and new cache) running big Fortran jobs? Is the difference there
>as much as the SpecFP bench seems to show?


Here are the results for some LSQR benchmarking I did. They do suggest
an advantage for the Coppermine's faster cache.

<table border><!-- alignment is |c|c|c|c|--><!-- 4 columns--><tr><td
align=center>Architecture</td><td align=center>Clock (MHz)</td><td
align=center>DRAM</td><td align=center>Time (ms)</td></tr>
<tr><td align=center>Alpha 21264</td><td align=center>500</td><td
align=center>PC200?</td><td align=center>40</td></tr>
<tr><td align=center>Dual PIII</td><td align=center>500</td><td
align=center>PC100</td><td align=center>70</td></tr>
<tr><td align=center>Athlon</td><td align=center>700</td><td
align=center>PC133</td><td align=center>75</td></tr>
<tr><td align=center>PIIIE</td><td align=center>550</td><td
align=center>PC100</td><td align=center>82</td></tr>
<tr><td align=center>Athlon</td><td align=center>650</td><td
align=center>PC100</td><td align=center>90</td></tr>
<tr><td align=center>PIII</td><td align=center>500</td><td
align=center>PC100</td><td align=center>120</td></tr>
<tr><td align=center>Celeron</td><td align=center>366</td><td
align=center>PC66</td><td align=center>125</td></tr>
<tr><td align=center>UltraSPARC II</td><td align=center>360</td><td
align=center>EDO 60ns</td><td align=center>130</td></tr>
</table>

The PC133 memory is CAS3. The PIIIE's memory is CAS2. I'm not sure what
the other PC100 memory is, but I'd guess CAS2. The PC66 is CAS2. The
Alpha is probably using custom "PC200" DDR-SDRAM.

The Athlon 700, PIIIE, and Celeron benchmarks were compiled with
more aggressive optimization than the other cases.