Thanks Lawrence for the taking the time to explain things a little more fully, and also in informing us that the 8.3 credit issue has been brought to the attention of the WCG staff

I have taken my 8-core cruncher off HCC indefinitely. To me, the issue is not just points, but principle. Users should not be penalized for running high-powered machines that crunch more WU's per day than others. It's easy for some people to dismiss this as much ado about nothing. But you are not the ones paying my power bills. 8-core machines consume a lot of electricity. I'm more than willing to donate the time and money, as long as fairness is involved.

Here are a couple of my personal favorites - as long as we're quoting stats.

All are from a Q6600 at 3.6ghz with 4 gig ddr2-1066 ram under Win XP

X0000041091526200410271439_ 0-- q01 Valid 11/26/2007 11:53:10 11/28/2007 23:58:40 4.66 103.0 / 70.6

X0000039141010200409241126_ 1-- q01 Valid 11/25/2007 06:53:51 11/28/2007 12:45:14 4.62 103.3 / 61.1

X0000039140400200409241137_ 0-- q01 Valid 11/25/2007 06:28:10 11/28/2007 12:45:14 5.04 112.7 / 62.8

Differences of 30, 40 and 50 points respectively.

From the 50 or 60 HCC wu's done on this machine, it looks like the run time is between 3 and 6 hours with the bulk clustering somewhere in the middle so I'm not sure we're comparing apples with apples.

I know the cpu is stable, but I can run some more intensive memory tests - although I'm running the chips slightly below their rated speed (because of limited memory multiplier options).

I'm not sure I've understood what you're trying to say, but it seems the gist of it is that testing memory function on my machine will not resolve the issue. That may or may not be true.

If we assume that we were, in each of our examples, dealing with comparable work units (which I think is at least subject to debate), then the question is how can a rig running at 2.4ghz is comparable to one running at 3.6ghz with twice as much memory? What I meant to say, although I wasn't explicit about it, was that if there are problems with memory access on my machine (a problem which i have in fact had on the other q6600's motherboard), that could explain why we have similar runtimes.

The other machine had problems with it's memory when I went from 3.2 to 3.4 ghz - even though the memory was running at a speed lower than it's rating. Since I've only gotten back into overclocking in a serious way in the past few months, I'm still trying to figure out what I need to do to resolve the issue. Do I play with FSB voltage, DDR2, MCH, all of the above, etc. It doesn't affect the stability of the cpu so I haven't generated many errors (forgetting about the computational errors discussed in another thread which are not computational in nature and may not even be system errors).

So that's why I suggested running a more intensive memory test. However we can probably get some idea of whether overclocking the cpu (aside from the effects this has on the rest of the motherboard) is an issue by just looking at our boinc benchmarks. When I set up the first q6600, i was getting about 2400 mega flops. Now I get a little over 3400. Not a linear progression, but more or less what you'd expect.

So if in fact our machines are running comparable wu's in comparable amounts of time, then issue has to be elsewhere. The first logical choice would be to look at the subsystems between the memory chips and the cpu.

As it turns out, on the first quad, memory access is stable when running a Prime95 test geared to testing memory but not when I try to run Prime95 with boinc. So it would seem I have a bit more tweaking to do.

If anyone wants to email me with suggestions, I'll be happy to hear them.