Another victim of AMD Cool'n'Quiet?

Didactylos, no. Cool 'n Quiet is not running on the quad. It is on one of the dual cores, and does not seem to be interfering with BOINC/WCG.

Thanks for the suggestion.

Sekerob, yes, that's the one. I see that I need to learn better search terms too.

At the time of that post I was only running Influenza work so that is the only project I noticed the slow down on. I reset my profile, as part of the trouble shooting the issue, to all projects. That is when I noticed that the issue seems to be across all projects, not just Influenza.

Thanks for the reply.

RMau

Didactylos, I don't think that you, or WCG, or BOINC have a problem. I think that I have a problem with my AMD quad core machine. I see a few other reports of problems with specific processors, but nothing so pervasive as to make me think that the problem is anywhere but in my machine.

The statistics I'm seeing are simply the elapsed times that the WUs take to complete. Something around six or seven hours on either of my AMD dual core machines. WUs for the same projects, downloaded in the same time frame, take 14 hours and up to 20 or more hours on the quad core. And it wasn't that way up until a month or so ago. The quad was the best cruncher I had running. And yes, good trouble shooting technique would begin with what changed a month or so ago. Try as I might, I cannot think of anything except a change in my BOINC client. And the ever offered Vista updates. Moving backward in BOINC clients has not changed the symptoms.

The WUs do finish successfully. So I suppose I could just limp along and let a WU take the better part of day to finish. But that seems almost counterproductive to the way that WCG measures performance and capability of attached machines. ie, the better your machine, the better your contributions and results.

I posted more as a shot in the dark that someone has experienced the same or similar issue. And more importantly, that some one had solved it and would share the fix. I did not post intending to point a finger at WCG or BOINC. Sorry if that is what you thought.

RMau

Hello, I have run into a similar problem. for some reason the work units are taking longer to do. However I suspected it was processor based. I suspect you are talking about phenoms. If this is true there may be a reason for this slowdown.
"AMD considers the quad core Phenoms to be the first "true" quad core design, as these processors are a monolithic multi-core design"
From my challenges from server motherboards that truly have 2 separate processers I can say this: The more processers you have the more memory each has to fight for. This is one of the reasons of lesser returns from more processers. Is this from the monolithic design? I may never know. Also I run Linux because of a bug in the TLB that is the look-aside buffer. Linux implements a kernel patch to avoid the bug that does not impact performance. However on windows machines, the "patch" makes a 20-50% performance hit. If you are stuck with windows then it is up to you on to disable it or not. From what I have seen it is about as common as if you suddenly won the lottery 10 times. But with these science apps I have no idea. Anyway Phenom gen 1s had some bumps and gen 2 is supposed to be so much better. Just my 2 cents i really have no idea what your problem is. I will say that on my core2 it takes a cretin project about half as long. However this is not true about HCC WUs (maybe due to 64 bit?) Hope this helps a bit.

Lightningrunner,

I'll look in to your comments. The machine in question has 5 GBs of RAM, and nothing I've seen, or been able to look at, says that I'm using more than half of that memory.

I was thinking along that line though, that it was possible that the BOINC CPU allocation and RAM memory were out of sync or something, causing the CPU to wait for data.

To explore that, I tried different BOINC settings for CPU use. Allowing BOINC to use CPU from 50% to 100% in 5% increments made no difference in my observed results. The WUs still ram 2x to 3x longer than my two dual core machines.