Folding is Hot!
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
-
- *Lifetime Patron*
- Posts: 125
- Joined: Sun Jan 05, 2003 12:59 am
- Location: Vancouver BC Canada
Folding is Hot!
This turns out to be a good testbench for running cpu's at full load. I almost had a meltdown after Folding all night with the closet door shut, my cpu was 82c in the morning!
There are probably other ways to test your cpu at 100% for a while, but this seems to be a pretty good way. Keep on eye on those temperatures.
There are probably other ways to test your cpu at 100% for a while, but this seems to be a pretty good way. Keep on eye on those temperatures.
Folding is a great stress testing thing, since I always run it to give my contribution to the wonderful world of biology (I love it, I got a 5 on the AP bio test last year ), so it gives not only 100% load stress testing, but a stress that represents real world stress completely, not impossible worst cases.
Prime95 is still a great test for stability though.
Prime95 is still a great test for stability though.
I started folding on Sunday (4 days). 6.5 WU later, my CPU seems pegged at 55C, MB at 32C+/-. A little warm, but reasonable for a 3.06 P4 with stock fan. I really need to change the CPU cooler, but have a few other fish to fry for a bit.
I'm pleased to do my small part for science and the SPCR team. Actually, I do my bit for science everyday at the lab, but that's another story....
I'm pleased to do my small part for science and the SPCR team. Actually, I do my bit for science everyday at the lab, but that's another story....
-
- Moderator
- Posts: 7681
- Joined: Mon Dec 09, 2002 7:11 pm
- Location: Maynard, MA, Eaarth
- Contact:
A guesstimate:
Hello:
It seems to me that the AMD model numbers make a pretty accurate comparison with Intel's mHz numbers. I also think that RAM latency makes a difference more so than bandwidth, but I could be wrong. Cache size doesn't seem to matter as much as it does with SETI@Home.
For the record, I have a 700mHz Athlon "Classic" running with 512MB low latency RAM (PC133 running at 100mHz) and two Athlon XP2100+ overclocked from 1.73gHz to 2gHz (so that puts them into 2400+ territory) with 512MB PC2100 CL2 RAM running at 308DDR 2-3-3CL. I run one of these 24/7 on Folding and the other runs Folding on the weekends and SETI during the week.
I joined the SPRC Folds team as #15 or 16, and I have been hovering in the 7th, 6th, or 5th place, depending on how my units score compared to Fredster's and PerArne's. Right now I'm in the #5 spot and we may be gaining a smidge on Powergyoza in #4!
It seems to me that the AMD model numbers make a pretty accurate comparison with Intel's mHz numbers. I also think that RAM latency makes a difference more so than bandwidth, but I could be wrong. Cache size doesn't seem to matter as much as it does with SETI@Home.
For the record, I have a 700mHz Athlon "Classic" running with 512MB low latency RAM (PC133 running at 100mHz) and two Athlon XP2100+ overclocked from 1.73gHz to 2gHz (so that puts them into 2400+ territory) with 512MB PC2100 CL2 RAM running at 308DDR 2-3-3CL. I run one of these 24/7 on Folding and the other runs Folding on the weekends and SETI during the week.
I joined the SPRC Folds team as #15 or 16, and I have been hovering in the 7th, 6th, or 5th place, depending on how my units score compared to Fredster's and PerArne's. Right now I'm in the #5 spot and we may be gaining a smidge on Powergyoza in #4!
Last edited by NeilBlanchard on Sun Apr 20, 2003 5:49 am, edited 1 time in total.
actually its your FPU, IPC and cache.
this is what i've found out so far:
the max operations per clock cycle for athlons & durons is 9, while for pentium 4s and celerons is 6. the older celeron's were 5. so clock per clock, amd cpu's are better for folding. higher the max IPC of the processor, the better. IPC=operations per clock cycle X speed of processor.
the FPU for athon xp's is also faster than the celeron's and pentium 4's. again, higher the FPU, the better.
more the cache, the better.
um.. and that's about all i know.
EDIT: had to check my numbers
EDIT: linkie http://www.cyberbreakcafe.com/jpj_compu ... uvscpu.htm
this is what i've found out so far:
the max operations per clock cycle for athlons & durons is 9, while for pentium 4s and celerons is 6. the older celeron's were 5. so clock per clock, amd cpu's are better for folding. higher the max IPC of the processor, the better. IPC=operations per clock cycle X speed of processor.
the FPU for athon xp's is also faster than the celeron's and pentium 4's. again, higher the FPU, the better.
more the cache, the better.
um.. and that's about all i know.
EDIT: had to check my numbers
EDIT: linkie http://www.cyberbreakcafe.com/jpj_compu ... uvscpu.htm
For a good page on how different processors compare in terms of F@H check out this: CPU Speed and Points Information It's from the same guy that produced the excellent Electron Microscopemonitoring software.
An example from his list: An XP1900 completes one of the frames in his example protein in 197 seconds, while a P4 2.7 takes 258. That's an improvement of 25% ( ! ) Not bad for a processor that costs one fourth as much.
An example from his list: An XP1900 completes one of the frames in his example protein in 197 seconds, while a P4 2.7 takes 258. That's an improvement of 25% ( ! ) Not bad for a processor that costs one fourth as much.
At first I was dismayed at how slow my 2.4GHz P4 was folding, especially when reading about the various AMD vs Intel folding speed comparisons. But using the "-advmethods -forceasm" switches made a HUGE difference for me. It used to take me ~30 hours to do a ~40 point Tinker core WU; my current Gromacs WU ("p540_BBA5_N in water", a 33-point WU) is scheduled to take 12 hours 20 minutes. So for my P4 the WU/hour rate is at least twice as fast using the Gromacs core.
I don't know if there's any improvement with Gromacs for you AMD people, but for me, now my P4 can keep up with you guys! I'm currently leading the pack among us lonely single-processor people, despite freqently taking the folding off-line for gaming.
From the projects page:
I don't know if there's any improvement with Gromacs for you AMD people, but for me, now my P4 can keep up with you guys! I'm currently leading the pack among us lonely single-processor people, despite freqently taking the folding off-line for gaming.
From the projects page:
So I'm glad I'm still doing being useful for the Folding@Home project people despite switching from Tinker to Gromacs. Happy folding!Gromacs is especially important to us since it will allow us to simulate explicit solvation models (much more accurate models) as fast or faster than the simpler models we've used in the past. We are interested in seeing how these more accurate models will make an impact, especially for the Alzheimer peptide work which appears to need the greater accuracy.
-
- Posts: 231
- Joined: Wed Apr 16, 2003 11:40 am
- Location: seattle, wa
-
- Moderator
- Posts: 7681
- Joined: Mon Dec 09, 2002 7:11 pm
- Location: Maynard, MA, Eaarth
- Contact:
Electron Microscope is great!
Hello:
Thanks Rusty, for reminding me again about this program -- it adds a great "front end" for the CLI F@H clients. Reminds me a lot of SETI Spy in it's usefulness. I see where you put the "-advmethods -forceasm" switches -- but how does one get them to take effect? Can/should the clients be restarted, and if so, is this done with EM?
Also, on the GROMACS bit, I have left my clients on "no preference" and it has mostly used the Tinker core, but a few GROMACS and few of another one that I can't remember. What setting did you use to get it to always use the GROMACS core?
Thanks Rusty, for reminding me again about this program -- it adds a great "front end" for the CLI F@H clients. Reminds me a lot of SETI Spy in it's usefulness. I see where you put the "-advmethods -forceasm" switches -- but how does one get them to take effect? Can/should the clients be restarted, and if so, is this done with EM?
Also, on the GROMACS bit, I have left my clients on "no preference" and it has mostly used the Tinker core, but a few GROMACS and few of another one that I can't remember. What setting did you use to get it to always use the GROMACS core?
-
- Patron of SPCR
- Posts: 543
- Joined: Mon Oct 21, 2002 1:01 am
- Location: Vancouver, BC, Canada
- Contact:
Neil,
-advmethods will force the use of the gromacs core.
herosformula,
How you include those switches depends on which client you're using and how you start the client.
If you're using the console client and you're starting as a scheduled task, you would add the switches to the end of the "run:" field of the properties dialog box. I think this can apply to other clients / startup methods too...
-advmethods will force the use of the gromacs core.
herosformula,
How you include those switches depends on which client you're using and how you start the client.
If you're using the console client and you're starting as a scheduled task, you would add the switches to the end of the "run:" field of the properties dialog box. I think this can apply to other clients / startup methods too...
-
- Moderator
- Posts: 7681
- Joined: Mon Dec 09, 2002 7:11 pm
- Location: Maynard, MA, Eaarth
- Contact:
what does the -forceasm switch do?
Hello:
What does the -forceasm switch do?
What does the -forceasm switch do?
It forces the "Force core assembly optimizations to be used if available" on. You can set it in the config as well, but apparently sometimes the core will turn the optimizations off if it senses a problem, even if the optimizations aren't really what's causing the problem. Having -forceasm on keeps it from turning it off.
There it is! I knew I had a link to a FAQ here somewhere: F@H Switches FAQ
-
- Posts: 231
- Joined: Wed Apr 16, 2003 11:40 am
- Location: seattle, wa
-
- Patron of SPCR
- Posts: 543
- Joined: Mon Oct 21, 2002 1:01 am
- Location: Vancouver, BC, Canada
- Contact:
herosformula, another method of checking which core you're using is if you use Electron Microscope to monitor your folding progress (or the F@H graphical client), it'll show you the name of the protein you're currently working on. All of the Gromacs proteins I've seen have "in water" as part of their WU name. I believe in the latest version of EM, mousing over the little computer icon will show you which core you're using as well.
I've gotten one Tinker core WU thus far despite having those switches set, but if that's what the project people have for me at that moment, that's what they need. Happy folding!
I've gotten one Tinker core WU thus far despite having those switches set, but if that's what the project people have for me at that moment, that's what they need. Happy folding!
-
- Moderator
- Posts: 7681
- Joined: Mon Dec 09, 2002 7:11 pm
- Location: Maynard, MA, Eaarth
- Contact:
"-forceasm" doesn't work for me
Hello:
I have tried both the -advmethods and the -forceasm switches. The -advmethods is fine -- I'm working on a Gromacs core now, but the -forceasm causes the client to finish a frame -- and then freeze! It flies through a frame in 2.5 minutes (rather than just over 7 minutes without it) BUT it does me -- or the team, no good if it crashes! In fact, it seems to confuse EM a bit too -- on one attempt, it ran through the frame in just 40 seconds! And I had to restart EM in order to straighten it out, since after I removed the -forceasm switch, it said that it was taking over 9 minutes and the client had already moved on to the next frame.
So, I'll finish a 33 point, 100 frame Gromacs core WU in 11.5 hours, whereas my other Athlon 2100+ (OC'd to 2gHz) is running a 23.6 point, 400 frame Tinker core WU in roughly 16.5 hours. That's 2.86 points/hour vs 1.43 -- exactly double! But, if only it ran it in 5.5 hours...but it failed to write anything to disk, so that's NG.
I have tried both the -advmethods and the -forceasm switches. The -advmethods is fine -- I'm working on a Gromacs core now, but the -forceasm causes the client to finish a frame -- and then freeze! It flies through a frame in 2.5 minutes (rather than just over 7 minutes without it) BUT it does me -- or the team, no good if it crashes! In fact, it seems to confuse EM a bit too -- on one attempt, it ran through the frame in just 40 seconds! And I had to restart EM in order to straighten it out, since after I removed the -forceasm switch, it said that it was taking over 9 minutes and the client had already moved on to the next frame.
So, I'll finish a 33 point, 100 frame Gromacs core WU in 11.5 hours, whereas my other Athlon 2100+ (OC'd to 2gHz) is running a 23.6 point, 400 frame Tinker core WU in roughly 16.5 hours. That's 2.86 points/hour vs 1.43 -- exactly double! But, if only it ran it in 5.5 hours...but it failed to write anything to disk, so that's NG.
i dont know what's wrong.. but using the -forceasm and -advmethods on the console version, i did 100/400 frames in eleven and a half hours. sucked.
i returned to the graphic version, and now i'm at 215/400 (from 100/400) in just under 8 hours. go figure.
i have p4 2G. all i do on my computer just browse, i dont listen to mp3's much. that 11.5 hours was while i was away from the keyboard. the computer hardly doing anything besides folding. the 8 hours was while i was browsing.
i returned to the graphic version, and now i'm at 215/400 (from 100/400) in just under 8 hours. go figure.
i have p4 2G. all i do on my computer just browse, i dont listen to mp3's much. that 11.5 hours was while i was away from the keyboard. the computer hardly doing anything besides folding. the 8 hours was while i was browsing.
-
- Moderator
- Posts: 7681
- Joined: Mon Dec 09, 2002 7:11 pm
- Location: Maynard, MA, Eaarth
- Contact:
I'm NOT complaining!
Hello:
The strange thing is, the other 2100+ finished it WU and moved onto a Gromacs core and the -forceasm is working fine there! It is doing an 18 point 100 frame WU in ~6.5 hours.
BTW, just because the CLI is slower on some WU doesn't mean it is a slower client -- these WU's vary all over the place, even with the same number of frames! The one that I finished yesterday was a 55.5 point 100 frame and it took about 12.5 minutes per frame. So, mostly the time it takes is rewarded with higher points. And look at my totals: I have two fast computers (one Folds part time) and one much slower computer (Athlon 700) -- I have only used the CLI client and I think all the folks above me on the team only use the CLI client! It's faster!
I'm not complaining one little bit -- I love my Athlons! Are you kidding? They only cost me ~$600 each and that include a retail purchase of the 120GB 'Cudas at the exorbitant price of $200, with sales tax.wussboy wrote:Neil, how can you be having so much trouble and still have 10x as many WU completed as I have? Please send me your troublesome computers! I will give them a good home.
The strange thing is, the other 2100+ finished it WU and moved onto a Gromacs core and the -forceasm is working fine there! It is doing an 18 point 100 frame WU in ~6.5 hours.
BTW, just because the CLI is slower on some WU doesn't mean it is a slower client -- these WU's vary all over the place, even with the same number of frames! The one that I finished yesterday was a 55.5 point 100 frame and it took about 12.5 minutes per frame. So, mostly the time it takes is rewarded with higher points. And look at my totals: I have two fast computers (one Folds part time) and one much slower computer (Athlon 700) -- I have only used the CLI client and I think all the folks above me on the team only use the CLI client! It's faster!
Re: I'm NOT complaining!
NeilBlanchard wrote:I think all the folks above me on the team only use the CLI client! It's faster!
okay, okay. i think my computer's just weird. but after this current WU (my 12th). i'll switch back to the CLI client for a few days and give it another chance.
-
- Posts: 231
- Joined: Wed Apr 16, 2003 11:40 am
- Location: seattle, wa