Tinkers are a Folders New Best Friend!
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
-
- Posts: 3
- Joined: Sun Apr 18, 2004 10:08 pm
Tinkers are a Folders New Best Friend!
Hey everyone, I was reading all the posts recently and decided to take off -advmethods in the command line to get those new tinkers and I've been consistently getting ones ranging from 100 to 200 points. So I suggest to those that have more powerful machines (at least a 2 GHz P4 or a fast Athlon) to do the same and start working on those tinkers. I'm currently running a 2.4 GHz P4 and leaving it on all day and I'm estimating it's taking me approximately under 2.5 days (give or take) to complete Tinkers with 400 frames. But most likely those who have slower machines but are still able to take advantage of the -advmethods should not take that out of the command line as you'll get more points production out of folding Gromac projects. Oh and those of you who don't like to leave the folding program on when your using your computer because of energy usage issues, I found out my system draws about 0.3 more amps when the CPU is at 100% load which is about 35 watts (give or take if your voltage out of the wall is not exactly 120). SO KEEP ON FOLDING!
~AndreW
~AndreW
I now have a good assortment of work units, to try to draw some tentative conclusions from.
Tinker, Gromacs, P4, Athlon XP, it doesn't seem to matter a whole lot, which is really about what I would expect given that BY DESIGN, all Gromacs projects and all Tinker projects should be worth exactly the same, at least when run on a 2.8MHz P4.
Look at the first two machines, both P4's, one with a Gromacs, the other with a Tinker, each worth about the same.
Look also at the 2.4C (HyperThreaded), running one Tinker and one Gromacs, each worth almost exactly the same number of points.
I am surprised to see the Athlons not doing better than they are with the Tinkers, and the P4's not doing worse. With the Athlon's supposedly superior FPU, I would have expected the Tinkers to outperform the Gromacs, but I am not seeing that with this particular group of WU's.
David
Code: Select all
CPU Clients Core PPW
P4 2.4B 1 Gromacs 508
P4 2.66B 1 Tinker 510
P4 2.4C 2 Gromacs 293
Tinker 294
P4 3.1C 2 Tinker 361
Tinker 369
3200+ 1 Gromacs 771
3200+ 1 Gromacs 607
2400+ 1 Gromacs 557
1800+ 1 Gromacs 495
2600+ 1 Tinker 562
3200+ 1 Tinker 542
2000+ 1 Tinker 419
Look at the first two machines, both P4's, one with a Gromacs, the other with a Tinker, each worth about the same.
Look also at the 2.4C (HyperThreaded), running one Tinker and one Gromacs, each worth almost exactly the same number of points.
I am surprised to see the Athlons not doing better than they are with the Tinkers, and the P4's not doing worse. With the Athlon's supposedly superior FPU, I would have expected the Tinkers to outperform the Gromacs, but I am not seeing that with this particular group of WU's.
David
I too prefer smaller work units, that can be completed in no more than 24 to 30 hours on a fast cpu. Something about needing the constant flow of gratification. Still, I plan to leave the -advmethods option off all my Athlon blades, and therefore let Stanford assign me what they wish. I'm leaving -advmethods on my P4's, especially the ones with HyperThreading. With two clients running, these big Tinkers take a looong time.
David
David
-
- Posts: 154
- Joined: Thu Oct 17, 2002 5:53 pm
- Location: Sydney
-
- -- Vendor --
- Posts: 162
- Joined: Thu Dec 04, 2003 5:35 pm
- Location: Rohnert Park, California, USA
- Contact:
A shout out to bill_amer
If you catch this thread maybe you can help us get a handle on the new points system regarding Tinkers.
Didn't you say you had a new AMD Athlon64 system? If that is true perhaps you could remove -advmethods from your flags and try to attract a Tinker. Then post up some frame times and project info so we can see how an AMD64 based unit handles tinkers on a PPD and PPW basis vs other CPUs.
On the other issues of AMD vs Intel P4 PPW and PPD, the AMD on a raw MHz to MHz CPU comparison, the AthlonXP does indeed pound the P4 silly on Tinkers relative to actual clock speed. However the gap appears to have narrowed some with the addition of HT and DDR memory and faster FSB's on the P4's.
Hey David, are you using -forceasm on your AMDs working Tinkers? If not, I wonder if your PPD improves with that flag set.
Stevo
If you catch this thread maybe you can help us get a handle on the new points system regarding Tinkers.
Didn't you say you had a new AMD Athlon64 system? If that is true perhaps you could remove -advmethods from your flags and try to attract a Tinker. Then post up some frame times and project info so we can see how an AMD64 based unit handles tinkers on a PPD and PPW basis vs other CPUs.
On the other issues of AMD vs Intel P4 PPW and PPD, the AMD on a raw MHz to MHz CPU comparison, the AthlonXP does indeed pound the P4 silly on Tinkers relative to actual clock speed. However the gap appears to have narrowed some with the addition of HT and DDR memory and faster FSB's on the P4's.
Hey David, are you using -forceasm on your AMDs working Tinkers? If not, I wonder if your PPD improves with that flag set.
Stevo
I agree, I had one of those and it was really nice ppd/ppw...almost too niceMichael_qrt wrote:I don't know if it's a freak protien or what but my 2500+ is folding a p1110_L939_K12M_nat_min1 tinker and it's giving me almost double the ppw I see on many gromacs units.
If these 249 point tinkers remain at that value I think they will now be the disproportionally "good" units to work.
I swear I didn't get any credit for it though when I sent it in today...
I always use -forcesse on my Athlon clients. I have read in numerous places, by people I would expect to know, that -forceasm is redundant with -forcesse on Athlon processors.Stevo@ARM wrote:Hey David, are you using -forceasm on your AMDs working Tinkers? If not, I wonder if your PPD improves with that flag set.
No, I have not tried -forceasm with a Tinker. I am currently running 3 Tinkers, so I could try it on one. I don't think anything (different) will happen, but I won't know for certain until I try, now will I?
Edit: No difference, which is what I expected. -forceasm doesn't actually enable any optimizations, but makes sure the standard optimizations are always used, even when folding instability has been encountered. Without this flag, FAH can sometimes use "standard loops" rather than the optimized assembly language loops.
David
This is true for P4 processors because they only have SSE or SSE2 instruction code. So SSE will automatically be on by default if you use -forceasm. But with the Athlon processor you have 3DNow! or SSE. If you only use -forceasm here, then 3DNow! will be default and not SSE. Thus you must add -forcesse to make SSE default and then -forceasm will kick in if you have a lock up or restart of the client. At least that is my understanding of the situation. With Macs I use -forceasm because it will then default to Altivec (read Mac version of SSE) if the client locks up or restarts. Yes, it all makes sense in MY mind. Fold on amigos!that -forceasm is redundant with -forcesse on Athlon processors.
I am currently working 3 of the bigass Tinkers (projects p694, p696, and p697, worth 236, 235, and 238 points respectively according to the Currently Running Projects page) on 3 of my most potent Athlon blades, and FAHLogStats is showing over 1K PPW for each one. A fourth blade with the same specs is getting 792 PPW on a Gromacs.
In other words, no sooner do I say "it doesn't make any difference" than I see three examples where Tinkers really ARE a folders best friend, or at least an Athlon XP's best friend. I don't have any of these proteins on any of my P4's at the moment to compare. I do have a couple of 128 point Tinkers on my P4's, which aren't performing particularly well, but I will wait to see the same proteins on both platforms before I make any (more) generalizations about the performance of Athlons vs. P4's on Tinkers.
David
In other words, no sooner do I say "it doesn't make any difference" than I see three examples where Tinkers really ARE a folders best friend, or at least an Athlon XP's best friend. I don't have any of these proteins on any of my P4's at the moment to compare. I do have a couple of 128 point Tinkers on my P4's, which aren't performing particularly well, but I will wait to see the same proteins on both platforms before I make any (more) generalizations about the performance of Athlons vs. P4's on Tinkers.
David
I'll start loving those 238 point Tinkers if the damn Tinkers doesn't mess up the files!!! I'm working on my 5th WU and I've got 3-4 times file corrupt message and had to start from scratch.
One issue I noticed yesterday is that my protein averages a frame every 6 min (AthlonXP 2500+ @ stock speed) Yesterday it started doing one frame every 12 mins?? Any ideas ???? Shut down the PC, fired it up this morning and lo, I get 1frame per 6 min again (but it started from scratch, dumping the already completed 200+ frames, grrrr)
As for flags, I've tried -advmethods, got only one gromacs but four tinkers. Tried -forcesse and/or -forceasm, but mins/frame didn't change one bit. Anything I'm missing?
One issue I noticed yesterday is that my protein averages a frame every 6 min (AthlonXP 2500+ @ stock speed) Yesterday it started doing one frame every 12 mins?? Any ideas ???? Shut down the PC, fired it up this morning and lo, I get 1frame per 6 min again (but it started from scratch, dumping the already completed 200+ frames, grrrr)
As for flags, I've tried -advmethods, got only one gromacs but four tinkers. Tried -forcesse and/or -forceasm, but mins/frame didn't change one bit. Anything I'm missing?
This is not a Tinker problem. This is a system instability problem. In order to prove this to yourself, run Prime95 in Torture Test mode for 24 hours. I can virtually guarantee it will fail long before 24 hours. 24 hours is just to verify stability. INstability takes far less time to prove. If it doesn't fail, we can go from there to investigate other possible causes, but the probabilty of it being a Tinker core problem is infinitismally small.burcakb wrote:I'll start loving those 238 point Tinkers if the damn Tinkers doesn't mess up the files!!! I'm working on my 5th WU and I've got 3-4 times file corrupt message and had to start from scratch.
My guess? FAH is encountering errors which it is attempting to deal with, but when you shut down and restart, it just says "to hell with it" and starts over from the beginning. The finger still points to hardware instability.One issue I noticed yesterday is that my protein averages a frame every 6 min (AthlonXP 2500+ @ stock speed) Yesterday it started doing one frame every 12 mins?? Any ideas ???? Shut down the PC, fired it up this morning and lo, I get 1frame per 6 min again (but it started from scratch, dumping the already completed 200+ frames, grrrr)
Right now there is NOTHING you can do to get Gromacs. This may change in the future, but at the moment at least, -advmethods does not seem to accomplish anything at all.As for flags, I've tried -advmethods, got only one gromacs but four tinkers. Tried -forcesse and/or -forceasm, but mins/frame didn't change one bit. Anything I'm missing?
If you have an Athlon processor, use -forcesse. For an Intel processor, use -forceasm. These flags do not influence which work units you get, but they will optimize your cpu for the ones you do get.
David
Removed the -advmethods flag from my Athlons and now p1110 is rolling in like mad. I updated EMIII protein file for 4/20/04 (newest version) on Tuesday, only most of them are saying only worth 70.9 points. Only one instance reads the correct 249.0 points as listed on the psummary page. I assume that the EMIII protein file data needs to be updated again. Also, the one instance reading 249.0 points was downloaded in the midst of earlier and later '70.9 ers' so it is not like Stanford was in the process of updating the Tinker scores. It's a weird, wild, wonderful ride! Fold on!
I saw a post somewhere (probably the main EMIII thread in the Third Party forum at f-c.org) where TheWeatherman indicated he wasn't updating any more protein files "until things settle down".
FAHLogStats also went from having what appeared to be a current protein file (extra-nfo.csv), to again having the OLD values, including 70.9 for a long list of Tinkers.
I don't believe ANY Tinkers are worth 70.9 under the new points system.
David
FAHLogStats also went from having what appeared to be a current protein file (extra-nfo.csv), to again having the OLD values, including 70.9 for a long list of Tinkers.
Code: Select all
"p638_L939_K12M_ext","10.00","70.90","400","TINKER"
"p639_L939_K12M_nat","50.00","70.90","400","TINKER"
"p678_L939_K12M_355K","50.00","70.90","400","TINKER"
"p680_L939_K12M_414K","50.00","70.90","400","TINKER"
"p685_L939_K12M_ext_355K","30.00","70.90","400","TINKER"
"p686_L939_K12M_ext_novisc","30.00","70.90","400","TINKER"
"p687_L939_K12M_473K","50.00","70.90","400","TINKER"
"p688_L939_K12M_444K","50.00","70.90","400","TINKER"
"p689_L939_K12M_503K","50.00","70.90","400","TINKER"
"p690_L939_K12M_532K","50.00","70.90","400","TINKER"
"p692_L939_K12M_int","30.00","70.90","400","TINKER"
"p698_L939_K12M_600K","50.00","70.90","400","TINKER"
"p699_L939_K12M_700K","50.00","70.90","400","TINKER"
David
-
- -- Vendor --
- Posts: 162
- Joined: Thu Dec 04, 2003 5:35 pm
- Location: Rohnert Park, California, USA
- Contact:
Newsflash! This just in... Important Folding Breakthrough by the Pande Group!peteamer wrote:... About the Chromies, are you sure your right? Only Dad keeps telling me, he's the way he is... cause of me! ...
Pete.
The Pande group's folding efforts have discovered new human genetic material identified as 'Petesomes' which are a mutated human chromesome that posess a 'genetic temporal distortion field'. Apparently, individuals possessing the rare but potent 'Petesomes' have an inherited trait that actually travels back in time to genetically alter their ancestors This results in an infinate loop process, called 'Petetosis' by the Pande Group, that results in each new generation of male Amers to be even wilder and wackier than the last, while at the same time causing each existing male parent generation to suddenly have random but powerful spells of 'Petetalysis' which can best described as spastic outbursts of wackyness when people least expect it.
This newly identified syndrome begins showing its effects a few years after spawning the next generation of male Amers. Apparently the temporal looping effect of, 'Petetosis' , results in more frequent episodes of 'Petetalysis' in the parental generation with each passing year, and so far it appears to be an irreversable condition!!!
Hey Pete, watch out, your Dad is gaining on you man! You need to put some more folding power online quick. It's time to boost to your patented 'sheep-propulsion-system' by adding some new ingredients to your special fuel mixture. Here I'll help you get started!
Stevo
stevo@arm wrote . . . lots, actually
Stevo, now I understand.
No wonder I've been feeling bad all these years since Pete was born.
You chaps (Brit for 'guys') don't know how lucky you've been.
You've 'known' him a short while,
I've known him . . . ages
Daddy Bill
Pete's away for the weekend - computer-less - and I'll probably delete this post before he gets back
Stevo, now I understand.
No wonder I've been feeling bad all these years since Pete was born.
You chaps (Brit for 'guys') don't know how lucky you've been.
You've 'known' him a short while,
I've known him . . . ages
Daddy Bill
Pete's away for the weekend - computer-less - and I'll probably delete this post before he gets back
Regarding Tinkers on an AMD system... (eeek, not back ON topic!)
My 1.46ghz TBred has one now, a p1101. Its currently at 815ppw. That's just nutso. Before I was happy if all four of my home systems managed that much (1.4 Palomino, 1.4 TBird, 1.46 TBred, 566 P3).
The other AMD systems are cranking too. 590ppw on a p1109 for the 1.4 Palomino, and even the non-SSE 1.4 TBird (who's been without -advmethods for a long time now) is pulling out 365ppw on a p1108.
My 1.46ghz TBred has one now, a p1101. Its currently at 815ppw. That's just nutso. Before I was happy if all four of my home systems managed that much (1.4 Palomino, 1.4 TBird, 1.46 TBred, 566 P3).
The other AMD systems are cranking too. 590ppw on a p1109 for the 1.4 Palomino, and even the non-SSE 1.4 TBird (who's been without -advmethods for a long time now) is pulling out 365ppw on a p1108.
stevo@ARM wrote :
" If you catch this thread maybe you can help us get a handle on the new points system regarding Tinkers . . . .
. . . remove -advmethods from your flags and try to attract a Tinker. Then post up some frame times and project info "
Stevo, first info so far :
Arguments: -forceSSE
p638_L939_K12M_ext 393 25.00 37.00 128.00 400 TINKER
32.37 hrs
Hope this is what you're looking for.
Will post more as I finish the WUs
Daddy Bill
" If you catch this thread maybe you can help us get a handle on the new points system regarding Tinkers . . . .
. . . remove -advmethods from your flags and try to attract a Tinker. Then post up some frame times and project info "
Stevo, first info so far :
Arguments: -forceSSE
p638_L939_K12M_ext 393 25.00 37.00 128.00 400 TINKER
32.37 hrs
Hope this is what you're looking for.
Will post more as I finish the WUs
Daddy Bill
dukla2000 wrote : - which A-64 do you have?
dukla2000, I'm using
Asus K8V
A-64 3200+ 2.00 GHz
1.0 GB RAM - can't remember what sort, off hand
All running at standard clock speeds.
I intend to generate enough data to satisfy everyone, then wind the set up a bit.
Never had the inclination before, but should be an interesting exercise.
BTW the .37 was 37 minutes, not 0.37 of an hour.
I'll decimalise properly next time
Daddy Bill
Good morning, Zyzzyx
dukla2000, I'm using
Asus K8V
A-64 3200+ 2.00 GHz
1.0 GB RAM - can't remember what sort, off hand
All running at standard clock speeds.
I intend to generate enough data to satisfy everyone, then wind the set up a bit.
Never had the inclination before, but should be an interesting exercise.
BTW the .37 was 37 minutes, not 0.37 of an hour.
I'll decimalise properly next time
Daddy Bill
Good morning, Zyzzyx
-
- Posts: 154
- Joined: Thu Oct 17, 2002 5:53 pm
- Location: Sydney
Hmmm, I guess the enhancements to the A64 over the old K7 core don't help it much. At least in tinker units. My current and previous work units are both these. The times I've seen on these on my 2.2GHz Barton are
31:40 - previous work unit, actual time to complete from logs.
33:20 - prediction from EM3 based on current frame times.
I know there will be some variability between work units with the same protien so it would be good to see some more times from your system Bill.
31:40 - previous work unit, actual time to complete from logs.
33:20 - prediction from EM3 based on current frame times.
I know there will be some variability between work units with the same protien so it would be good to see some more times from your system Bill.
-
- Posts: 154
- Joined: Thu Oct 17, 2002 5:53 pm
- Location: Sydney