I'll test it on one of my machines and than i give feedback in a few hours.

greetings

Franz

Ok, It works fine. Performance ist top!!

12437.64 sec instead of 24492.51 sec for the same WU-Type.

great Job

Franz

Wow, I can't believe you guys are still putting so much effort into optimizing the separation client for the CPU. Is there any point besides "but I want to run them, and so do others"?

@Martin Chartrand: reset MW if you used an app_info and just pull down the trigger on your HD4870 and let the bullets, ehm, Wus fly :)
My HD4870 takes between 137 and 274s/WU - still very nice. You CPU or nVidia can be much more useful elsewhere.

MrS

---

Scanning for our furry friends since Jan 2002

Wow, I can't believe you guys are still putting so much effort into optimizing the separation client for the CPU. Is there any point besides "but I want to run them, and so do others"?

I think there are many people which have no GPU for chrunching, and if somebody could increase the performance for them it's a good thing, and the people are happy. Every part is valuable even the smallest.

Every part is valuable even the smallest.

Yeah.. well. But what if that tiny contribution here could be big elsewhere?
The point is that ATIs slice through these WUs in a matter of minutes. End the searches are of sequential nature: the server sends out many WUs, but at some point it has to gather results and decide on the next steps. Everything returned so far can be taken into account. As far as I understand this happens at the time scale of tens of minutes. So if an optimized CPU is able to return a result within 3h now.. it might just be lucky and be the perfect fit, or it will very likely long be outdated.

Don't get me wrong, I like MW and there's a reason my previous and current cards are DP ATIs. But I really think the power of CPUs is much much more valuable elsewhere than for running separation tasks. For example I believe running a "unoptimized" (or at least uncustomized) n_body search on the CPU is more productive than running a highly optimized separation on the CPU.

MrS

---

Scanning for our furry friends since Jan 2002

As far as I understand this happens at the time scale of tens of minutes. So if an optimized CPU is able to return a result within 3h now.. it might just be lucky and be the perfect fit, or it will very likely long be outdated.

If you are right the the existence of a spearation CPU application is ridiculous.
There must be a statement of the staff (Matt, Travis)

So.. now that it works... we got a few new apps supporting Intel SSE4.1, Intel SSE3 and Intel/AMD SSE2 (PENTIUM4_SSE2 and AMD SSE2)

Using the SSE4.1 app v0.91 (I have a Xeon supporting up to SEE4.2), my units fail within a second with:

<core_client_version>6.10.58</core_client_version>
<![CDATA[
<message>
- exit code -1073741515 (0xc0000135)
</message>
]]>

All the same exit code.

EDIT: it works with the SSE3 application.

Any news on a possible CPU app that supports AVX extensions?