Scott,

I just had my first invalid result where a P4-Xeon (x86 Family 15 Model 2 Stepping 9) was teamed up with a core 2 based Xeon (x86 Family 6 Model 15 Stepping 7).
When looking at the source of the hr.C posted in the BOINC trac, I notice that all cpu's that have "Xeon" in their p_model record are put into one category (like in the example workunit I linked to), while the P2, P3, P4 and Core based Intels are all split into different categories.
Will this generalization of the different Xeon architectures be causing issues when the project starts using higher precision in the work later on?

Scott,

I just had my first invalid result where a P4-Xeon (x86 Family 15 Model 2 Stepping 9) was teamed up with a core 2 based Xeon (x86 Family 6 Model 15 Stepping 7).
When looking at the source of the hr.C posted in the BOINC trac, I notice that all cpu's that have "Xeon" in their p_model record are put into one category (like in the example workunit I linked to), while the P2, P3, P4 and Core based Intels are all split into different categories.
Will this generalization of the different Xeon architectures be causing issues when the project starts using higher precision in the work later on?

Well, actually, we're using higher precision right now. Anyways, adding some sub-classes to the Xeon processor would probably be a good idea. I'll get back to you when I take care of it.

---

Scott Kruger
Project Administrator, Cosmology@Home

well, actualy, what I mean is not subdividing the Xeons, but matching them to their desktop or mobile counterparts.

for example, the two Xeon system I have listed are based on the P4 Northwood desktop cpu, which places them in the

if (strcasestr(host.p_vendor, "Intel")) {
..
   if (strcasestr(host.p_model, "x86")) {
..
..
      if (strcasestr(host.p_model, "Family 15 Model 2")) return IntelPentium4;
..
..
   }
..
}

while host 509 for example is the SMP-capable version of host 446, which should place host 509 into the same group as host 446.

if (strcasestr(host.p_vendor, "Intel")) {
..
  if (strcasestr(host.p_model, "Core")) return IntelCore2;
..
}

but due to the setup of the filter, that will not happen


arf

ok

I think this might be a job for the BOINC-developers to sort out, as it may impact more (especially small) projects that use the cpu_fine section of the hr.C

How about this one then, Mr. Pernod? http://cosmos.astro.uiuc.edu/cosmohome/show_host_detail.php?hostid=78

That's the Xeon counterpart of the Q6600. So I guess it should also be put in the Core 2 category?

---

BOINC.BE: For Belgians who love the smell of glowing red cpu's in the morning
Tutta55's Lair

How about this one then, Mr. Pernod? http://cosmos.astro.uiuc.edu/cosmohome/show_host_detail.php?hostid=78

That's the Xeon counterpart of the Q6600. So I guess it should also be put in the Core 2 category?

In my opinion it should, as the Xeon X3220 is based on the same architecture as the Q6600.
Even though it is a Xeon, the basic architecture of this CPU is as different from my Xeons as the Core2 desktop cpu's are from the Pentium4's, which could potentially lead to different results on the same workunit.