Page 2 of 2

Re: WU 4513, 4514

Posted: Fri Apr 25, 2008 1:37 am
by vvoelz
A bit more about the points for the 45xx series: each RUN consists of fragments that have slightly different numbers of atoms. (If we *didn't* do this, we'd quickly run out of project numbers). While points for any one WU should vary with the slightly different number of atoms depending on the sequence, the average should match the benchmarked values.

It's good to hear about any point discrepancies on the forum -- they are duly noted. Before I release any of the 'bonused' projects (4516 onward), I will doing even more benchmarks to double-check that we're assigning points correctly.

Vince

Re: WU 4513, 4514

Posted: Fri Apr 25, 2008 5:19 am
by alancabler
Yes, that behavior has been apparent to varying degrees.
The more we know the better we like it.
Thanks Vince.

Re: WU 4513, 4514

Posted: Fri Apr 25, 2008 3:03 pm
by alancabler
Just had a run of 5 p4513s on a Northwood (similar to benchmark) machine, and the behavior is exactly as Vince described.
There was a wide variance in ppd/GHz between Runs, (and even within each run), but the overall avg. was 41.68 ppd/GHz, which is in line with standard benchmark scores.

Re: WU 4513, 4514

Posted: Fri Apr 25, 2008 3:06 pm
by bruce
alancabler wrote:There was a wide variance in ppd/GHz between Runs, (and even within each run), but the overall avg. was 41.68 ppd/GHz, which is in line with standard benchmark scores.
How wide?

Re: WU 4513, 4514

Posted: Fri Apr 25, 2008 3:55 pm
by alancabler
Scores posted at fahinfo.org.
Between runs, ppd varied between 31 ppd/GHz - 52 ppd/GHz.
Frame times varied within each WU by 9%-19%, depending on WU. 10 or so completion points (frames) were calculated within each WU, and posted scores for each WU were averages of those frame times.

None of my machines running a GUI client have finished previous work and received one of these p451x units yet, but I'm looking forward to visualizations which might show something interesting to reflect these variances. Never can tell... 8-)

Re: WU 4513, 4514

Posted: Fri Apr 25, 2008 10:33 pm
by vvoelz
Hi all --
The psummary*.html should now be reporting the number of atoms and points in the current 45xx projects (4510-4515).
The problem was not only because of some pathname issues, but also because of high server load (the psummary script timeouts when ssh'ing into the server). The timeout issues will persist for the very near future, but should clear up by tomorrow.
--Vince