Posts by John

1) Message boards : Number crunching : Rosetta CPU optimization - how ? (Message 38444)
Posted 26 Mar 2007 by John
Post:
without saying too much about it, Intel and Rosetta are closing the license agreement, I ll be proud of accelerating it as soon as we protect properly Rosetta intellectual property.

In the mean time, we have to make sure we are fair to the entiere Rosetta users, so, I ll be proposing a code path for each architecture, and see if we can do a QA for it, a regression tool will have to be created.

I hope Matt from AMD, who has the code already will give me an hand and incluse this own best code in it.

Let's keep the race fair.


who?

Great! So how much should various combos of SSE/SSE2/SSE3,etc speed Rosetta up?
2) Message boards : Number crunching : PS3 preliminary crunching numbers (Message 38365)
Posted 25 Mar 2007 by John
Post:
Basically, are they saying that PS3 is crunching 73% worth of workloads or is it just measuring "perfomance"?
3) Message boards : Number crunching : PS3 preliminary crunching numbers (Message 38356)
Posted 25 Mar 2007 by John
Post:
Ok.. I think TFLOPS is baloney marketing hype term.
http://fah-web.stanford.edu/psummary.html
So what do you make of this link showing few PS3 units and few atoms per PS3 unit?
4) Message boards : Number crunching : PS3 preliminary crunching numbers (Message 38326)
Posted 25 Mar 2007 by John
Post:
http://fah-web.stanford.edu/psummary.html
Also, on that link, 264 333 264 333 576 576 576 576 582 582 582 582 are the number of atoms per ps3 work unit?

A few of the seemingly non PS3 work units.. 19399 6115 1135 14917 14836 91787 91787 87920 25541 18894 126006 77373 126006 308047
5) Message boards : Number crunching : PS3 preliminary crunching numbers (Message 38320)
Posted 25 Mar 2007 by John
Post:
How exactly and what is "TFLOPS" supposed to measure?
20x faster seems to be the most conservative number I've seen.
http://fah-web.stanford.edu/cgi-bin/main.py?qtype=osstats
I've calculated it to be ~26x more than PCs which isn't too far off (702TFLOPS/28672)/(153TFLOPS/160524).

That figure is probably inflated, considering most of the PS3 Folders are probably sitting there collecting dust, and the PC Folders are either ancient Celerons or busy doing something.
===
And that's just "TFLOPS".
http://fah-web.stanford.edu/psummary.html
There I see PS3 constituting 12/100+ WU. The PS3 WU's don't award that much (compared to Gromacs)... So how in the world is PS3 getting such a high TFLOPS number?
6) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38246)
Posted 24 Mar 2007 by John
Post:
http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=11332
"Help Cure Muscular Dystrophy: When the PDB code / Protein Symbols in the 'I' screen left hand bottom change and the 2 proteins in the main graph assume the same colours. Colour changes are pure random, thus could on outside chance assume same colour even if checkpoint was reached. Watch the PDB code change for absolute indication! (See Sample Image and FAQ for description)

Genome Comparison: Approximately every 20 minutes (See Sample Image and FAQ for description)

Help Defeat Cancer: at 25% intervals - writes large files (See Sample Image and FAQ for description)

Human Proteome Folding 2: Occurs after each structure attempt. Look at the graphics, one can see how far along an attempt is. When the 3 line graphs reach the end of the X axis and restart at the left, the structure attempt is complete and a checkpoint occurs (See Sample Image for UD Agent, BOINC Agent and FAQ for description)

FightAIDS@Home: When the Best Energy C graph green line has reached the end and returns to the beginning, whilst rescaling the graph and adding a red line indicating the path of the previous attempt. (See Sample Image and FAQ for description)"

I would imagine that each of these workloads are a good amount different, yet they are each able to save progress in a mindful matter...
7) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38245)
Posted 24 Mar 2007 by John
Post:
Also would like to note that WCG Boinc (not sure about United Devices) also has the "longer time to completion" problem. Yes, I realize if I come back to PC after say 2 hours it'll obviously be lower, but it still makes no sense to tick in basically +1 increments.
8) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38236)
Posted 24 Mar 2007 by John
Post:
Of course, not always ticking up 1 seconds, sometimes it goes down a few, generally up though in 1+.. I've never seen another time predictor (DVD burn, download, XP reformat) that is constantly adds time to completion.. And even if it did, did it dynamically, rather than ticking up to what it thinks is the time..


"Ok... why does WCG's seem to know "how much" is total/needed/done per "work unit"? Can someone explain the differences in the workloads"
And any thoughts on this?
9) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38233)
Posted 24 Mar 2007 by John
Post:
Thanks for the direct reply, quoted from a year ago... So I'm guessing no progress there?


"Ok... why does WCG's seem to know "how much" is total/needed/done per "work unit"? Can someone explain the differences in the workloads"
Hello

"Answer is not normal.. Come on, 1 second increments? How about recalculating it every ~10 minutes or something so that you won't have the randomness of download managers but still... a guesser that makes sense."

If any of the download managers behaved like this they probably wouldn't be downloaded any more. I'm not understanding why, if Rosetta thinks that it takes 6 hours, it needs to count via 1 second increments from 4 hours instead of just dynamically adjusting...
10) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38187)
Posted 23 Mar 2007 by John
Post:
So I'd like to hear from a Rosetta dev why they can't resume work (save often) in the middle of a crunching.. I mean, we can suspend BOINC then resume in a few minutes, so why can't we suspend over a shutdown?
11) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38185)
Posted 23 Mar 2007 by John
Post:
I realize that..
I don't think my question of why Rosetta doesn't checkpoint more often, why Rosetta resets timer (maybe everything also) has been answered.. Someone said it's because of "dumping memory" (for timer reset) but WCG is also set to "dump memory" but it doesn't reset time.
12) Message boards : Number crunching : Why can't Rosetta checkpoint more often (compared to WCG)? +feedback (Message 38182)
Posted 23 Mar 2007 by John
Post:
Rosetta, 1st credit, invalidated after many hours. WCG, "finished", not updated on website, I quit distributed computing for a while (couple days). Then I come back to WCG to see if it updated.. and it did! Whoo, after many hours, it didn't reset, the timer or the checkpoint (at least not significantly)...

I'm wondering. Can anyone explain the process of Rosetta processing vs WCG FightAidsAtHome/Genome Comparison (the only two I've done/doing)? Why those two can checkpoint at good intervals, while Rosetta goes for hours at 1%, I exit, then I have no idea if I start it again, timer resets to 0, I don't know if "actual" % is 1 or 4 hours worth.

For example, Rosetta processing is like a house of cards in the face of the wind, it must always need your "shielding".. Or with a PC, when your "shield" or RAM goes away, house of cards goes away. That would be my example of why Rosetta is quirky?
===
"Q: Progress Percent not advancing?
A: Rosetta recomputes the progress percent at the end of each model."
Ok... why does WCG's seem to know "how much" is total/needed/done? Can someone explain the differences in the workloads..

Q: "To completion" time is going UP!
Answer is not normal.. Come on, 1 second increments? How about recalculating it every ~10 minutes or something so that you won't have the randomness of download managers but still... a guesser that makes sense.
13) Questions and Answers : Windows : CPU Time reset after relaunching the program? (Message 38017)
Posted 19 Mar 2007 by John
Post:
So does Aids at home have a check point every .001% or something? I quickly exited it at 58.3%, and then it came back to 57.9% and chugged along. On the other hand, Rosetta seems to be indefintely on 1%.

I see why it's hard to calculate %, but can't there at least be a "raw" counter of how much has been crunched or something?

And my memory settings are the same for WCG as they are for Rosetta but I'm not having to deal with all the quirks of Rosetta.
14) Message boards : Number crunching : 1 percent status and Time to Completion (Message 38003)
Posted 19 Mar 2007 by John
Post:
How come WCG can do this nicely but Rosetta can't?
15) Questions and Answers : Windows : CPU Time reset after relaunching the program? (Message 37993)
Posted 19 Mar 2007 by John
Post:
Is the CPU timer on the tasks tab of the advanced view supposed to start at 0 everytime? Happens with Rosetta but not with WCG (aids at home)
16) Message boards : Number crunching : Screensaver CPU usage separate from regular CPU usage? (Message 37989)
Posted 19 Mar 2007 by John
Post:
And it completed and didn't validate.
17) Message boards : Number crunching : Screensaver CPU usage separate from regular CPU usage? (Message 37988)
Posted 19 Mar 2007 by John
Post:
More Rosetta weirdnesss. Maybe my "1 hour run time" is kicking in.. But taking a bath, my % was 3.3% and time to completion was ~7 hours. I come back less than an hour later to 67%.. Rosetta needs some fixing. And feel free to move this to the QA section that I just found.
18) Message boards : Number crunching : Screensaver CPU usage separate from regular CPU usage? (Message 37979)
Posted 18 Mar 2007 by John
Post:
Question.. Rosetta is frustrating me.. I did WCG and it autoed to Aidsathome.. and it started at 0.000% and I could see a +0.002% about each second or so.. Rosetta starts at 1%.. It'd be good if Rosetta could start at 0% and have smaller increments (I'm not sure if I've ever seen any)

And I chose 1 hour run time but keep getting ~7 hour tasks..
19) Message boards : Number crunching : Screensaver CPU usage separate from regular CPU usage? (Message 37974)
Posted 18 Mar 2007 by John
Post:
Will my 1% go in vain or will they pick up from where I left off?
20) Message boards : Number crunching : Screensaver CPU usage separate from regular CPU usage? (Message 37971)
Posted 18 Mar 2007 by John
Post:
PS, I kind of went nuts since I was stuck at 1% it seemed indefinitely and yeah.. I have 2 "in progress" results, cancelled both, waiting for another to do.. What happens to the ones I didn't finish?


Next 20



©2023 University of Washington
https://www.bakerlab.org