[Captain Newbie]

Hello. I'm doing a bit of an informal study on how well F@H works (simply driven by curiosity) and I'm interested in a few things:

1) The name of your core executables (e.g., Fah_Core78, 7A, ...), and
2) How long it takes you to complete each type of core.

Information on the core you are running can be found in Windows' Task Manager, OS X's Activity Monitor, or the ubiquitous Linux top. Time to complete can be found in your FAH logfiles.

Also, please include information on the systems you are running.

I'm sure that this has been done elsewhere, but OCF here seems to have a wide spectrum of users running various types of CPUs under various conditions. Thus—posting here. I will generate a statistical plot of this information every so often and post it.

[dicecca112]

Specs in my sig, except the memory is running at default timings and voltage

Running QMD core (ie FahCore_96) time per frames running two instances are 22:48 and 24:48, with times to complete each work unit is 37:53:20 and 41:20:00. These are all taken from EMIII

I'll hopefully have my 820 tomorrow and I'll post times as well for you. If you need anything else feel free to just ask. Also these times are high for me, seeing I'm running Firefox, WMP, and Outlook Express.

Also I think this is great idea Capt. Newbie and if you need any help don't hesitate to ask.

[Captain Newbie]

Thanks for the prompt reply.

My main interest here is that Stanford asserts that it takes about 1 CPU-day to simulate 1nsec of folding. I'm still reading several papers on the topic (papers here) and the F@H FAQs.

My aim is to calculate the relative efficiency and effective thouroughput in terms of nsecs folded per CPU-hour, to be honest. To this end, I'm doing a crapload of reading on the cores, algorithms in the core, and the science of folding. Can't put these papers down, to be honest. It's fascinating.

Man, I'm a _geek_.

[ChasR]

The number of atoms in a work unit are going to affect what you're attempting to do. 1-CPU day is rather nebulous as well. A lot less is accomplished in a 400 MHz P2 CPU day than in P4 840 CPU day. If you look at psummary, p1155 through p1161 have what appears to be the length of the simulation in the protien name. These WUs all have the same number of atoms and fold in the same amount of time and yet the DT in the name varies from 250ps to 20 ns. How will you determine the length of simulation of each protien?

[Captain Newbie]

Stanford benchmarks each core on a P4C 2.8 (from FAQ) before releasing...so that's my definition of a CPU-day, and a letter to the Physical Review in 2001 by Pande and Shirts says 'about 1nsec per CPU day' (This fact should be taken in isolation from the P4!). I still have to read a LOT on this topic before I can do this, but it would be great to collect and start tabulating the statistics anyway.

The issue of how much folding actually goes on is indeed quite nebulous. I need to drop back and do some more thinking on it.

[steveeb]

820 running 2 QMD's core96
Process 1 - 29hr 40mins
Process 2 - 29hr 36mins

EDIT:
Also, just finished a core78 that took 29mins/frame or about 48hours to complete on the same system.

[00:50:53] Project: 1149 (Run 1, Clone 24, Gen 6)
[00:50:53]
[00:50:54] Assembly optimizations on if available.
[00:50:54] Entering M.D.
[00:51:16] (Starting from checkpoint)
[00:51:16] Protein: p1149_RIBO_semihelixfrom1140

[Soichiro]

11:24 for a p2075_BBA5_ext (DGromacs, w00t! ) on the comp in my sig (the one with all the stuff written out). But remember that my processor is underclocked. Stupid stock heatsink. I can't wait until I get my Scythe Katana from newegg

[TollhouseFrank]

if you want, I can post a bunch of projects and average ppd for each project depending on if my compy or wife's (i got an a64, she has a socket A sempron)

[Captain Newbie]

Please do.

[Who]

Have you checked out walaka7's Core Performance Guide? Lots of info there.
http://www.ocforums.com/showthread.php?t=387980

[Captain Newbie]

That's a good start, but I'm actually trying to determine nanoseconds folded/reference CPU day.

[ChasR]

I can't see how you're going to do that unless you get Pande Group to tell you how many nanoseconds each simulation covers. Dr Pande's statement 1 nanosecond per cpu day is a broad generalization, an average and has to be different on each of the six different cores and on each of the WUs on each core. It may well have been made before the P4c 2.8 was even the benchmark machine, when the Tinker core was predominant and the Gromacs core was experimental. Surely there will be fewer nanoseconds per cpu day simulated on a protien with 100,000 atoms than will be simulated on a protien with 300. IMHO, you have too many unknowns to arrive at anything meaningful.