Contest: Who's got the highest Performance Fraction?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Contest: Who's got the highest Performance Fraction?
A little "contest" here. Who can boast the highest Performance Fraction? This "contest" is open to SPCR members only. Posts by non-SPCR members will be deleted at my discretion. The only prize is "bragging rights".
The Performance Fraction (PF) for a single WU is the percentage of time remaining to the deadline. The quicker a WU is completed and returned, the higher the PF. The PF FAH keeps track of internally is some kind of average or weighted average, I'm not sure, over the most recent N WU's. 7 comes to mind, but again I'm not sure.
It's based not just on how powerful your system is, but how much of the time it spends folding. Obviously then, the highest number would be expected to come from the most powerful dedicated folder.
At the end of each WU, FAH computes an "updated Performance Fraction". If you have -verbosity 9 specified, this will be printed in the log file, otherwise you will have to use EMIII, QD, or other utility to dig this value out of your queue.dat file. I don't know if it's displayed anywhere by LogStats or not. I haven't been able to find it.
My highest is .983601, on one of my 3200+ Barton blades (running Linux).
Who can top that?
David
The Performance Fraction (PF) for a single WU is the percentage of time remaining to the deadline. The quicker a WU is completed and returned, the higher the PF. The PF FAH keeps track of internally is some kind of average or weighted average, I'm not sure, over the most recent N WU's. 7 comes to mind, but again I'm not sure.
It's based not just on how powerful your system is, but how much of the time it spends folding. Obviously then, the highest number would be expected to come from the most powerful dedicated folder.
At the end of each WU, FAH computes an "updated Performance Fraction". If you have -verbosity 9 specified, this will be printed in the log file, otherwise you will have to use EMIII, QD, or other utility to dig this value out of your queue.dat file. I don't know if it's displayed anywhere by LogStats or not. I haven't been able to find it.
My highest is .983601, on one of my 3200+ Barton blades (running Linux).
Who can top that?
David
-
- Posts: 35
- Joined: Mon Apr 05, 2004 10:38 pm
LogStats doesn't seem to report that anywhere, but of course if you've got linux you can just go in the directory and grep for "perf" from *.txt as such:
$ grep perf *.txt
FAHlog.txt:[02:40:56] Updated performance fraction: 0.945658
FAHlog.txt:[11:13:50] Updated performance fraction: 0.943358
FAHlog.txt:[07:40:53] Updated performance fraction: 0.950429
FAHlog.txt:[16:46:02] Updated performance fraction: 0.945513
FAHlog.txt:[01:51:36] Updated performance fraction: 0.942398
That will find any such lines in your current FAHlog.txt or FAHlog-Prev.txt. Do I get some special prize for "most challenged" by the fact the Barton core is not supported on my mobo, and thus running only 995Mhz? Ah well, my 2Ghz Thoroughbred should show up in another day or two. Then I expect it would be roughly twice as fast, putting my fraction around 0.97 or so. Respectable I guess, but not quite in the league with you guys!
$ grep perf *.txt
FAHlog.txt:[02:40:56] Updated performance fraction: 0.945658
FAHlog.txt:[11:13:50] Updated performance fraction: 0.943358
FAHlog.txt:[07:40:53] Updated performance fraction: 0.950429
FAHlog.txt:[16:46:02] Updated performance fraction: 0.945513
FAHlog.txt:[01:51:36] Updated performance fraction: 0.942398
That will find any such lines in your current FAHlog.txt or FAHlog-Prev.txt. Do I get some special prize for "most challenged" by the fact the Barton core is not supported on my mobo, and thus running only 995Mhz? Ah well, my 2Ghz Thoroughbred should show up in another day or two. Then I expect it would be roughly twice as fast, putting my fraction around 0.97 or so. Respectable I guess, but not quite in the league with you guys!
Yes, absolutely. Two processors, one that runs at 2X speed, but folds only 12 hours per day, will have the same PF as one that runs at speed X but folds 24 hours per day.dukla2000 wrote:TBred @ 13*166 = XP2700+FAHlog-Prev.txt:[19:57:17] Updated performance fraction: 0.986504
Seems to me this varies according to uptime as well?
This works IF you have specified -verbosity N, where N is a number high enough to elicit the printing of the "Updated performance fraction" message. -verbosity 9 will definitely show it.AZBrandon wrote:LogStats doesn't seem to report that anywhere, but of course if you've got linux you can just go in the directory and grep for "perf" from *.txt as such:
$ grep perf *.txt
FAHlog.txt:[02:40:56] Updated performance fraction: 0.945658
FAHlog.txt:[11:13:50] Updated performance fraction: 0.943358
FAHlog.txt:[07:40:53] Updated performance fraction: 0.950429
FAHlog.txt:[16:46:02] Updated performance fraction: 0.945513
FAHlog.txt:[01:51:36] Updated performance fraction: 0.942398
That will find any such lines in your current FAHlog.txt or FAHlog-Prev.txt.
David
Barton 2500+ oc'd to 3200[03:34:31] Updated performance fraction: 0.986992
[03:34:31] Sending work to server
I have a P4 2.4 (not oc'd) thats at 0.985157 and it's used 8hrs a day -- I was kinda shocked -- the other ones are down in the 0.96's.
FWIW, wouldn't any HT enabled machine basically be eliminated since you're running two F@H's on the same CPU? My HT P43.06 is a better folder, but I'm pretty sure it doesn't do as well with performace fraction as the barton...
Dave
EDIT Fix messed up quote
-
- Posts: 35
- Joined: Mon Apr 05, 2004 10:38 pm
Yes, that's absolutely true. ANY cpu running two instances of FAH will have a lower Performance Fraction than if it were running just one. The best of my HT clients are in the .97 range.dasman wrote:FWIW, wouldn't any HT enabled machine basically be eliminated since you're running two F@H's on the same CPU? My HT P43.06 is a better folder, but I'm pretty sure it doesn't do as well with performace fraction as the barton...
It's also "luck of the draw" since PF is based on the deadline assigned to each WU. WU's with tigher deadlines will play hell with the PF.
As an aside, Pande Group isn't really very happy about people running more than one instance of FAH on any one processor, HT or otherwise, since it does slow down the completion of each WU. It's hard to convince people to do that though, since it results in fewer points. They would have to offer some incentive (more points) for the quickest possible return of work units, but when they have floated this idea, the response has been overwhelmingly negative.
David
-
- Posts: 35
- Joined: Mon Apr 05, 2004 10:38 pm
Woohoo! A new high...
Code: Select all
[09:11:30] Updated performance fraction: 0.987389
[09:11:30] Sending work to server
One of the "rules" I failed to mention is, you have to tell us the cpu and clock speed. Come on, fess up.Hortalonus wrote:Woohoo! A new high...Code: Select all
[09:11:30] Updated performance fraction: 0.987389 [09:11:30] Sending work to server
David
That makes absolutely no sense to me whatsoever.haysdb wrote:As an aside, Pande Group isn't really very happy about people running more than one instance of FAH on any one processor, HT or otherwise, since it does slow down the completion of each WU.
For example, they can get a 2 WUs back in, say, 16 hrs (no HT) or can get 2 WUs back in ~12hrs (assuming a 25% PPW increase for the HT machine -- I think I've seen that figure quoted around here somewhere)
They're getting more results returned in less time with HT machines running multiple instances. What's to complain about?
Dave
There are threads at folding-community.org where this has been discussed. The Machine ID's were introduced to accomodate multiple cpu systems, NOT explicitly to allow the running of multiple instances of FAH on one CPU. In fact, there was some discussion on how to prevent this. When it was agreed the practice could not be stopped, there was talk of giving "bonus points" based on how quickly a WU was returned. Nobody seemed to like this idea either. As recently as two months ago Vijay Pande made a comment that some changes were in the works that would negate any advantage to running more than one instance of FAH on one cpu. At firt I thought this was hinting at a new core that would take full advantage of HyperThreading, but I have since read this is not the case. Therefore it has to have something to do with the points system, somehow.dasman wrote:That makes absolutely no sense to me whatsoever.haysdb wrote:As an aside, Pande Group isn't really very happy about people running more than one instance of FAH on any one processor, HT or otherwise, since it does slow down the completion of each WU.
For example, they can get a 2 WUs back in, say, 16 hrs (no HT) or can get 2 WUs back in ~12hrs (assuming a 25% PPW increase for the HT machine -- I think I've seen that figure quoted around here somewhere)
They're getting more results returned in less time with HT machines running multiple instances. What's to complain about?
Dave
From a different perspective, I have read over and over that the next WU in a sequence cannot be determined until the previous one is completed and returned. This would place a high priority on returning one WU as fast as possible, vs. returning TWO in even a slightly longer time.
David
I thought that everyone was working on the same WU's -- how many current projects are there? 90? 100? I've got (6) of the p724 proteins going right now -- aren't they all the same thing?haysdb wrote: From a different perspective, I have read over and over that the next WU in a sequence cannot be determined until the previous one is completed and returned. This would place a high priority on returning one WU as fast as possible, vs. returning TWO in even a slightly longer time.
I thought they needed the entire project to finish before they could move on (I really don't know anything about this tho...) If so, they'll get the project done faster using multiple instances on the HT machines.
Well, I guess I'll just continue to not understand -- no big deal.
Dave
-
- Posts: 35
- Joined: Mon Apr 05, 2004 10:38 pm
Same CPU as my previous high... Mobile Athlon 2600+ @ 2500MHz.haysdb wrote:One of the "rules" I failed to mention is, you have to tell us the cpu and clock speed. Come on, fess up.Hortalonus wrote:Woohoo! A new high...Code: Select all
[09:11:30] Updated performance fraction: 0.987389 [09:11:30] Sending work to server
David
All the PF's in my log over the last 48 hrs, in chronological order:haysdb wrote:It's also "luck of the draw" since PF is based on the deadline assigned to each WU. WU's with tigher deadlines will play hell with the PF.
0.982948 99% p520 - 40 days
0.984392 99% p520 - 40 days
0.985817 99% - p563 - 25 days
0.981821 97% p730 - 5.1 days
0.978506 97% p732 - 5.7 days
The 2nd column is the percentage of time to deadline remaining, the next is the projectnr followed by the deadline.
Barton 2800+ running at 11x195=2145, folding 24/7
-
- Posts: 542
- Joined: Mon Aug 11, 2003 5:54 pm
It comes up at the completion of the WU.How long does it take get PF scores?
It will take forever if you don't fold.
I don't know, I'm going over and set my borg for them in the next few days. From that point I'm not sure but I will post when I find out. It's got to be more that the Xpired tinkers that have been turned in.What kind of scores do you get for timeless tinkers??
Hijack over
Our #1 folder has weighed in with a new high PF of 0.987909TRC-13 wrote:[09:04:55] CoreStatus = 64 (100)
[09:04:55] Unit 9 finished with 99 percent of time to deadline remaining.
[09:04:55] Updated performance fraction: 0.987909
[09:04:55] Sending work to server
Barton 2500+ @ 3200+ w/400MHz FSB no Vcore change.
David
That is low. My Linux server, with its 1800+ Palomino, boasts a .97 PF.Kemokim wrote:AND THE WINNER IS........
[11:00:48] Updated performance fraction: 0.747165
hehe my AOpen notebook Pentium mobile 1700 Mhz... damn this is bad!!
why is this so slow?? it uses 100% cpu, and even though its a mobile chip it doesnt speed down cause of constant 100% cpu usage..
David