Project 2684 run time

The most demanding Projects are only available to a small percentage of very high-end servers.

Moderators: Site Moderators, PandeGroup

Re: Project 2684 run time

Postby johnerz » Sat Apr 09, 2011 9:39 pm

Grandpa_01 wrote:In most cases it is not worth the $$$ spent to fold GPU and bigadv at the same time. Due to the increased point value of shortened return time. If you reduce or increase your frame times by 1 min. it can be a increase or decrease of 1000's of PPD. Stanford wants the bigadv WU's returned as fast as possible and the faster you return it the more points recieved. And the increase and decrease in PPD vs time per fram is non linear, The more you decrease the time per frame the faster the PPD increase 1 min may = 5000 ppd and 1 min 30 sec may = 10,000 points.



I've read around on this and so far your reply is what I was looking for.

I have just gone from a 4 core cpu to a 4 core +4 cpu (Q9650 to 2600k :).

I was already aware of the differences in ppd based on what the config of the base pc against my pc was :(

I have 2 good to very good GPU (1 X 285 and 1 X 570) from your post I will run 1 X bigadv only, rather that 1 X bigadv and 2 x GPU, as if I understand correctly the watt vs ppd against cost makes this the optimum choice per £ and for the science. :)

Now if I'm wrong................................:)
johnerz

Intel 2600K @ stock
EVGA 670 FTW @ stock
12GB 1600
Asus P67 Sabertooth bios version 3209
Corsair hx 1000 psu
WD Black 500 GB

Win 7 64, updated
Microsoft Security Essentials - updated daily

Updated 4 Dec 2012
johnerz
 
Posts: 78
Joined: Thu Jun 19, 2008 3:21 pm

Re: Project 2684 run time

Postby codysluder » Sun Apr 10, 2011 4:09 am

johnerz:
A good GPU works well in a slow computer. Go buy something other than your 2600K. It can be really, really slow and cheap as long as it has a pcie slot and enough power to run the GPU. Check your local thrift store. People are discarding fully functional computers when they upgrade. The computer won't cost as much as the GPU but together they'll fold.
codysluder
 
Posts: 2222
Joined: Sun Dec 02, 2007 12:43 pm

Re: Project 2684 run time

Postby tear » Wed Apr 13, 2011 4:36 am

Bob8421 wrote:I'd still like an explanation of just why Stanford thinks that project 2684 is exactly like all the other 8955-point work units.

It seems it was assumed it would perform on par with other WUs with similar number of atoms but it hasn't been explicitly benchmarked*.
Apparently simulation parameters differ so much (GROMACS file format difference is another possibility) that the assumption is no longer valid.

P2684 should be given a spin on benchark machine and its base points should be re-evaluated.

*) we are yet to see a report of a machine simulating P2684 as fast as, say, P690x or P2686
One man's ceiling is another man's floor.
Image
tear
 
Posts: 924
Joined: Sun Dec 02, 2007 4:08 am
Location: Rocky Mountains

Re: Project 2684 run time

Postby tear » Wed Apr 13, 2011 4:41 am

Update: now that I think of it, it could've been other way around. P2686/P690x may be newer, therefore faster, so it may actually be the new units that need re-evaluation.

Analysis would require close examination of GROMACS data files and performing a run or two on hardware-identical-to-the-benchmark-machine.

FYI -- I'm not involved in the project, just a random reverse engineer.
tear
 
Posts: 924
Joined: Sun Dec 02, 2007 4:08 am
Location: Rocky Mountains

Re: Project 2684 run time

Postby Tobit » Fri May 06, 2011 2:16 pm

P2684 was just given some much needed love this morning. P2684 now has a base points value of 12,790 which will hopefully make people much happier now. Thanks to Vijay for listening and working with several donors to understand the problem with this and other low producing units (see the P7200 announcement as well).
User avatar
Tobit
 
Posts: 743
Joined: Thu Apr 17, 2008 2:35 pm
Location: Manchester, NH USA

Re: Project 2684 run time

Postby Grandpa_01 » Sat May 07, 2011 1:45 pm

Tobit wrote:P2684 was just given some much needed love this morning. P2684 now has a base points value of 12,790 which will hopefully make people much happier now. Thanks to Vijay for listening and working with several donors to understand the problem with this and other low producing units (see the P7200 announcement as well).


That is good to hear. It is nice to see that Stanford adjusted the point value of these to be more in line with the rest of the bigadv WU's. Perhaps in the future they should not use an i5 to benchmark a project that by their rules can not be run on an i5. May be they should use a CPU that matches the requirements set forth by them. As tear mentioned earlier I never once saw a post from anyone that said there performance on a 2684 even came close to any of the other bigadv WU's. Thanks Vijay for adjusting these now maybe people will quit dumping them. I am willing to bet the % of 2684's I get will drop now. :wink:
Image
4 - SuperMicro H8QGi-F AMD 6xxx=192 cores @ 2.8Ghz
3 - SM X9QRI-f+ Intel 4650 = 192 cores/threads @ 3Ghz
2 - I7 980X 4.4Ghz 2 - GTX680
1 - 2700k 4.4Ghz GTX680
Total = 419 cores folding
User avatar
Grandpa_01
 
Posts: 1747
Joined: Wed Mar 04, 2009 7:36 am

Re: Project 2684 run time

Postby Tobit » Sat May 07, 2011 1:55 pm

Grandpa_01 wrote:That is good to hear. It is nice to see that Stanford adjusted the point value of these to be more in line with the rest of the bigadv WU's. Perhaps in the future they should not us an i5 to benchmark a project that by their rules can not be run on an i5. May be they should use a CPU that matches the requirements set forth by them.

They don't use an i5 for benchmarking. They use an i5-class Xeon, the exact model being an X3450 which is a Quad Core with HT for 8-threads total so it meets the requirements albeit just barely.

Source: http://folding.stanford.edu/English/FAQ-Points#ntoc10
User avatar
Tobit
 
Posts: 743
Joined: Thu Apr 17, 2008 2:35 pm
Location: Manchester, NH USA

Re: Project 2684 run time

Postby jimerickson » Sat May 07, 2011 1:58 pm

thank you dr. vijay pande and team for fixing this. my frustration levels have dropped considerably.
jimerickson
 
Posts: 673
Joined: Tue May 27, 2008 11:56 pm
Location: ames, iowa

Re: Project 2684 run time

Postby Nathan_P » Sat May 07, 2011 6:36 pm

Grandpa_01 wrote:In most cases it is not worth the $$$ spent to fold GPU and bigadv at the same time. Due to the increased point value of shortened return time. If you reduce or increase your frame times by 1 min. it can be a increase or decrease of 1000's of PPD. Stanford wants the bigadv WU's returned as fast as possible and the faster you return it the more points recieved. And the increase and decrease in PPD vs time per fram is non linear, The more you decrease the time per frame the faster the PPD increase 1 min may = 5000 ppd and 1 min 30 sec may = 10,000 points.


The only way to test if it is beneficial is to test it on your machines, I have a dual x5670 machine that i also run GTX 460 on. On my set up i lose ~3k from running the gpu but gain 10.8k from the Gpu resulting in a net gain of nearly 8k. For me that make it worth it as its nearly the same ppd as my GTX 275 for less power and heat but at always YMMV.

I would say that you should only try this with a high end GPU that can pump out 10k or more to make it worthwhile.
Censorship leads to dictatorship

Image
Nathan_P
 
Posts: 1321
Joined: Wed Apr 01, 2009 9:22 pm
Location: Jersey, Channel islands

Re: Project 2684 run time

Postby bruce » Sat May 07, 2011 6:46 pm

A HD5450 can produce a reasonable PPD, but it doesn't match the loss to the SMP bonus on my C2Q. (This will likely change as Core_16 is optimized, too). As Nathan suggests, YMMV, but we're off-topic. We should be talking about p2684 running by itself, as that's the only thing that is considered when the Pande Group sets the points.
bruce
Site Admin
 
Posts: 16886
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Project 2684 run time

Postby Nathan_P » Sat May 07, 2011 7:16 pm

bruce wrote:A HD5450 can produce a reasonable PPD, but it doesn't match the loss to the SMP bonus on my C2Q. (This will likely change as Core_16 is optimized, too). As Nathan suggests, YMMV, but we're off-topic. We should be talking about p2684 running by itself, as that's the only thing that is considered when the Pande Group sets the points.

True we are off topic - but since the project has been re scored points is less of an issue. Frame time on the other hand is still out of whack. I would be intruged to know why it is slower than the others. I know Kasson has responded to the DAB aboout it being an older project and using a different version of the core but maybe there is more to it than that.
Nathan_P
 
Posts: 1321
Joined: Wed Apr 01, 2009 9:22 pm
Location: Jersey, Channel islands

Re: Project 2684 run time

Postby bruce » Sat May 07, 2011 8:07 pm

TPF depends on many factors including the number of steps, the number of atoms, the exact nature of the analysis, (and, of course, hardware). You cannot compare the TPF of one project with another.

GROMACS is very versatile. This project uses a different type of analysis inside of Gromacs.
bruce
Site Admin
 
Posts: 16886
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Project 2684 run time

Postby Grandpa_01 » Sat May 07, 2011 10:38 pm

bruce wrote:TPF depends on many factors including the number of steps, the number of atoms, the exact nature of the analysis, (and, of course, hardware). You cannot compare the TPF of one project with another.

GROMACS is very versatile. This project uses a different type of analysis inside of Gromacs.


But you can compare the reward for doing the work. And some WU's just do not return the same reward as others which creates a problem with donors. In all actuality when you have a machine that gets X amount of points per minute / hour /day for one type of work it should receive the same amount of points for all of the work done on that type of work. IE all smp wu's should return the same approx value, all bigadv should return the same approx value etc. Stanford created the points system to encourage more people to fold, (we all know if you want more people to do something just make it completive) but when you have a WU like 2684 which is out of line with the rest of the active WU's you have a problem. Stanford has slowed the project down themselves by encouraging people to dump 2684's by giving them a value of 30% less than other active bigadv WU's. I have 2 of them ruining and even with the bump they received they are still 15% lower in value than their rest of the bigadv WU's. Why is it like that it still takes the same amount of $$$$ to build the rig the same amount of electricity per hour to operate so why is there less value.

I do appreciate the 15% increase now how about another 15% and get them on par with the rest of the bigadv WU's, after all if you went to work today and was told you were going to get 15% less how would you like it ?. I would most likely just take the day off and start looking for another job. :wink: Do not get me wrong here I will continue to fold whatever I receive from Stanford I will just get more satisfaction from folding some WU's more than others.
User avatar
Grandpa_01
 
Posts: 1747
Joined: Wed Mar 04, 2009 7:36 am

Re: Project 2684 run time

Postby Tobit » Sat May 07, 2011 10:47 pm

Grandpa, that is why it's being addressed finally. Don't worry, be happy. :)
User avatar
Tobit
 
Posts: 743
Joined: Thu Apr 17, 2008 2:35 pm
Location: Manchester, NH USA

Re: Project 2684 run time

Postby ei57 » Sun May 08, 2011 12:44 pm

I used to get a ppd of 36k on a specific PC. With the new base points, it has increased to 52k. That is only 4k less than the average big adv WU and is acceptable. Another 1k base points would make them equal, but I'm happy as is.

6701/6702 should be subject to a similar appraisal. They are lagging the average smp in the same way. 400 extra base points will do the trick.
ei57
 
Posts: 98
Joined: Thu Jun 12, 2008 10:23 am

PreviousNext

Return to SMP with bigadv

Who is online

Users browsing this forum: No registered users and 1 guest

cron