Not sure what version of Intel's Fortran compiler CPDN is using in BOINC, but the Intel 7.x Fortran compiler disables SSE and SSE2 paths if the processor is not genuine Intel. We've had these threads here before, but there is more information and admission from Intel here

<a href="http://www.theinquirer.net/?article=24574">http://www.theinquirer.net/?article=24574</a>

<a href="http://www.swallowtail.org/naughty-intel.html">http://www.swallowtail.org/naughty-intel.html</a>

They should upgrade to the 8.1 compiler, and apply the patch that is available on this site:

<a href="http://www.swallowtail.org/naughty-intel.html">http://www.swallowtail.org/naughty-intel.html</a>

Then recompile Boinc so that it will run faster on AMD chips. Maybe that's why it runs slower on AMD!

And it certainly does run slower than it should -- my Athlon64 3500+ averages 2.27 sec/TS whereas my wife's Dell Pentium 4 3.0 GHz HT machine gets 3.40 sec/TS PER CPU (of which 2 CPUS show up, because of hyperthreading)

Now do the math - her machine seems to be able to do more work per day than mine. It is a bit slower - but it should be twice as slow, since she's getting twice the Work Units done.

My machine has a faster FSB, etc. and should be getting things done faster than it is.

Matthew

<img src="http://boinc.mundayweb.com/one/stats.php?userID=2118&amp;prj=4">

&gt; They should upgrade to the 8.1 compiler, and apply the patch that is available
&gt; on this site:

It doesn't even really need to be patched to make a big difference. Looking at that matrix of benchmarks, 8.1 with the -xW switch is a huge performance gain over the default in any 7.x compiler, at least on Opterons/A64s. Not doubt it may help with AthlonXPs as well, although they didn't show any benchmarks for that.

Is it a big deal for them to recompile the code with a new version of the Intel compiler?

Are there any plans to do so? You'd think that a lot of people would love it, as it would mean getting more science done per day on EVERYONE's machine, without any hardware upgrades!

Matthew

<img src="http://boinc.mundayweb.com/one/stats.php?userID=2118&amp;prj=4">

Carl posted this in the phpBB forum...

&gt; hmmm, not surprising as AMD64 3800+ run .5 sec/ts slower than P4 3.4GHz boxes
&gt; from what I've seen on the sulphur cycle beta. I'm trying out Intel Fortran
&gt; version 9.0 which if the short tests go OK I'll try out on hadsm3, and if I get
&gt; adventurous (and Tolu lets me) may try out on a new version for the sulphur
&gt; cycle (but probably wait until the coupled model).