gromacs

[seeker010]

does gromancs increase ppd for the a-xp over tinker compared to the a64/p4?

[seeker010]

ok confirmed. on my xp-2000+ mobile laptop at least, I get a better ppd turning off advanced asm and running tinker instead of gromacs. doh!

[UDHA]

really? I thought gromacs simply got more work done in the same time,but the PPD should remain close or the same to tinker. but in practice I guess not.

[seeker010]

I think it's because the reference amd machine is an a64? xp's suffer from some sse latencies that was alleviated in the a64 core.
someone knowledgeable please say something.

[Lupine]

It would help if you could specify the WUs you've been processing along w/ actual PPD per WU.

Generally speaking, a Gromac WU holds a huge processing advantage over Tinker WUs with, among other optimizations, its support of SSE. The Gromac FAQ claims up to a 10x or 20x performance increase. But that leaves out a key factor: point bonuses.

Point bonuses are what make possible the 200, 300, and even higher PPDs we hear talked about (HT, dual core, and multi proc systems take those totals even higher). Point bonuses are generally awarded for beta WUs, including QMD WUs (Intel only WUs at this time). This is because these WUs may be less stable or more stressful on components than full release WUs. Without these bonuses, all WUs would be somewhat comparable in their PPD.

Back to Tinker vs Gromac. WUs w/ point bonuses are available in both flavors, so there is a wide range in PPD within each group. That is why it is difficult to say which is better since they both vary wildly from within.

Pande Group recently published a Points FAQ to help explain the discrepancies seen between different WUs and different cores that helps to explain things somewhat.

For what its worth, I run the -advmethods/bigWU combo on both my home rigs (one AXP & one A64 - see siggy) and, depending on the WU, the A64 is about 10% higher in PPD than the AXP, although there have been times when its been closer to 50% higher. But then there are other times when, due to luck of the draw IMHO, the A64 gets a WU w/o point bonus and is down to the 120PPD or so level instead of 250PPD (600 pointers) or some of the new 350PPD+ WUs I was getting last week.

[KingFish]

Also something noteworthy of mention is that there is no AMD reference machine for FAH. The work units are benchmarked on an intel machine. The reason for this is that intel processors outnumber amd processors by about three to one for folders. Here's the link for that breakdown on stanford's site:

http://vspx27.stanford.edu/cgi-bin/main.py?qtype=cpustats

Stanford also recently released a good FAQ on how points are calculated that may clear up some of your questions:

http://folding.stanford.edu/FAQ-points.html

Also, here's a link to a Stanford Gromacs FAQ:

http://folding.stanford.edu/gromacs.html

Hope these help!

[seeker010]

gromacs p888_p53peptide888 ppd ~60

tinker p1152_L939_K12M_ext_from638 ppd ~100