Folding time Q

J

jmackay

2[H]4U
Joined
Nov 23, 2006
Messages
3,928
Hey guys, I just checked the status on my secondary (old) box and it seems to be folding very slowly....

P4 2.8 / 1 gig ram

I am running 2 instances of the console on it, and checking the log I saw it was taking 1 hour, 6 minutes (on avg) for 1% for one instance; and 1 hour, 8 min (avg) on the second instance.

66 mins x 100 = 6600 minutes to finish or 4.58 DAYs
68 mins x 100 = 6800 minutes to finish or 4.72 DAYs
So 4.65 DAYs avg, 2.326 days each.

Is this correct or is it taking too long? My C2D burns through them quickly taking about 13 mins running the SMP client and I think about the same running 2 instances of the console.

Did I configure something wrong? Is the ram limited for two instances? Should I try the SMP client? It does have HT on the old box.
 
Correct me if I am wrong, but I think it would be more efficient to run only one instance of the console on a P4 since it is a single core.

 
Exactly.

The only time you should run dual instances of the Normal Client, or run the SMP Client is when you have More than one Core.

And well, its obvious that the CPU stated is singled cored.
 
Sgraffite said:
Correct me if I am wrong, but I think it would be more efficient to run only one instance of the console on a P4 since it is a single core.

Click to expand...

Well because it has HT it only shows as 50% cpu usage, so still only run 1?
 
Blueagle87 said:
Exactly.

The only time you should run dual instances of the Normal Client, or run the SMP Client is when you have More than one Core.

And well, its obvious that the CPU stated is singled cored.
Click to expand...

http://www.hardfolding.com/index.php?go=3&gtype=1&id=4&show=1

Multiple CPU/ Hyperthreading setup guide

Here is how you run Multiple FAH clients on a Hyperthreading or Multiple CPU system.

1) create seperate folders for each client. (I use " fold1" and " fold2" , etc) and download a console client into each one. Warning: The graphical client has many problems with running 2 or more clients at the same time , use the console only!

2) right click on each Icon and add the flags " -local " and " -config" (see picture)
Click to expand...

So guess the FAQ needs to be updated then if you should only run 1 instance?

How fast should the folding go on a P4 2.8 w/1 gig ram and HT?
 
Sorry, Didn't know that was stated in the FAQ.

Are you sure you followed the Instructions precisely?

I would uninstall and try again, following those same instructions.

If the same problem occurs then its time to only run one instance. Because those times are Insane.

I dunnue the normal times for one instance on your specific proc, but it would no doubt be around 23 hours or less per WU.
 
Ok, I thought the times were crazy too thus the question... I noticed the anti-virus had been doing something for the last 2 hours, so maybe it was acting up as well... rebooted the machine and just running 1 instance for now, I'll prob have to wait for a while for it to update but I'll post some new times soon :)

Thanks for the help so far... I was hoping my P4 wasn't that crappy :p
 
It's currently not worth running more instance on a hypterthreaded P4. When they had the timeless tinkers, it actually helped production a bit. Since the Tinkers did not use SSE and the gromacs did, you could run one instance with the regular WUs running full bore. Then you could run the second instance with the timeless tinker at about 15-20% and it would not affect the first instance much. This way the regular instance stayed at about the same speed to finish the unit which made Stanford happy and the second instance with the timeless tinker would use what was left over for the extra points. Stanford did not care how long it took to return the timeless tinkers so it ended up being the best of both worlds.

However, since the discontinuation of the timeless tinkers, there is no reason to run two clients for a hyperthreaded CPU.

I'd still love to see how the C2D's would chew up the QMDs. The efficiency and optimizations of the C2D would make me think that it would tear through QMDs like there was no tomorrow. My P4 2.8 at work would get about 400PPD out of the QMDs which was unheard of production at that time due to the steep memory requirements of the WUs. Not to mention they liked to saturated the FSB or the memory bandwidth. I don't remember which right now.

 
jmackay said:
Ok, I thought the times were crazy too thus the question... I noticed the anti-virus had been doing something for the last 2 hours, so maybe it was acting up as well... rebooted the machine and just running 1 instance for now, I'll prob have to wait for a while for it to update but I'll post some new times soon :)

Thanks for the help so far... I was hoping my P4 wasn't that crappy :p
Click to expand...


Not that we can compare setups but I am clocked @ 3.357 Ghz and My SMP WU's take 12.75 Minutes a Percent. (When Its the only thing running)
 
jmackay said:
Thanks for the help so far... I was hoping my P4 wasn't that crappy :p
Click to expand...

Right now, P4s are crappy. With the current WUs I'm lucky to get 100 PPD out of a P4 2.8. It would be better if it would pick up Amber, gbgromacs, Dgromacs or the former 600 and 364 pointers but it never does. It's always stuck on the 186-206 point Gromacs which have a tendency to run rather slow as far as PPD is concerned. It annoys me since I have the "benchmark system" and the benchmark is supposed to be 110 PPD without running SSE boost. I can't even get near that with SSE boost.

 
Blueagle87 said:
Not that we can compare setups but I am clocked @ 3.357 Ghz and My SMP WU's take 12.75 Minutes a Percent. (When Its the only thing running)
Click to expand...

I'm running @ 3.2ghz and getting 1% @ 12min 25 sec with some other programs running in the background.

SmokeRngs said:
Right now, P4s are crappy. With the current WUs I'm lucky to get 100 PPD out of a P4 2.8. It would be better if it would pick up Amber, gbgromacs, Dgromacs or the former 600 and 364 pointers but it never does. It's always stuck on the 186-206 point Gromacs which have a tendency to run rather slow as far as PPD is concerned. It annoys me since I have the "benchmark system" and the benchmark is supposed to be 110 PPD without running SSE boost. I can't even get near that with SSE boost.

Click to expand...

That was all waaay over my head (new to the folding thing), but knowing that P4s are slow now is good (?) to hear.

 
SmokeRngs said:
Right now, P4s are crappy. With the current WUs I'm lucky to get 100 PPD out of a P4 2.8. It would be better if it would pick up Amber, gbgromacs, Dgromacs or the former 600 and 364 pointers but it never does. It's always stuck on the 186-206 point Gromacs which have a tendency to run rather slow as far as PPD is concerned. It annoys me since I have the "benchmark system" and the benchmark is supposed to be 110 PPD without running SSE boost. I can't even get near that with SSE boost.
Click to expand...

They are benched on at 2.8 Ghz P4 with SSE on but SSE2 off running Linux.
You should be running within 10% of the benchmark speed, so look for 100-120 PpD.
If your running alot under 100 PpP then theres something wrong.
A lot over 120 PpD and you've a bonus workunit.

Luck ............ :D
 
Tigerbiten said:
They are benched on at 2.8 Ghz P4 with SSE on but SSE2 off running Linux.
You should be running within 10% of the benchmark speed, so look for 100-120 PpD.
If your running alot under 100 PpP then theres something wrong.
A lot over 120 PpD and you've a bonus workunit.

Luck ............ :D
Click to expand...

My work maching is a P4 3.2 and from fahmon, it's averaging 100-105 ppd so I guess it's normal. I cannot wait to get a new C2D maching during summer :)

 
ok I looked at all these posts, and what protein are you folding, that can make a difference in the times , there are a couple of really chewy units that i have had recently
 
Ok after restarting it and running only 1 instance, times went down to 45min avg.

It finished off Protein: p2418_Ribo_Mg 53-99-4 and did 250000 steps.

Now its working on p2418_Ribo_Mg 22-38-5 and also doing 250000 steps @ about the same time.

Isn't this still pretty slow? I mean if I was doing two @ 66mins and now 1 @ 45ish, seems like I was getting more done running 2 instances although they were both going slower.

45 x 100 = 4500 mins to finish
4500 / 60 / 24 = 3.125 days for a single, while I was getting 4.65 DAYs avg (for 2), 2.326 days each before.

Should I reinstall or is this about correct for a P4 2.8?
 
If you want points, run 2 instances, according to Stanford if you want to help them produce results more quickly, run 1.
 
ND40oz said:
If you want points, run 2 instances, according to Stanford if you want to help them produce results more quickly, run 1.
Click to expand...

Well its not about points either, as I would be giving them TWO completed units, it would just take an extra day and a half to deliver two instead of one.

Assuming this time stays the same:

30 days time:
9.6 completed if running single
12.9 completed if running two

Thats about 9.5 vs 13 finished.

But again.. is this taking too long to finish and is there something wrong with my install / computer? Or do P4s just take this long?
 
p2418_Ribo_Mg is worth 500 points.
500/3.125 = 160 Points per Day.
So your doing slightly better than Stanford folding this protien.

The reason Stanford would like the work back quick is the next generation of protiens is made from the results of this generation.
So the faster they get the results back, the faster they can get the next set ready.
But as long as you can get both results back before their first deadlines then I dont think it make that much of a difference if you run one or two copies of folding.
Its if running a second copy makes you miss the first deadlines then thats bad for the science as the protien get sent out again for someone else to fold.

Luck ........... :D
 
Where can I see the deadline info for these? :)

I'd rather finish 2 than 1 since it would help more in the long run, but if I start missing deadlines here and there then its not worth it hehe

Edit:

Also, is there a way to close the SMP version besides just closing the cmd window? I close it whenever I game or need to do some heavy work on the computer and it says that I improperly terminate it... never had this problem with console, but it doesn't appear to hurt anything.
 
jmackay said:
Where can I see the deadline info for these? :)

I'd rather finish 2 than 1 since it would help more in the long run, but if I start missing deadlines here and there then its not worth it hehe

Edit:

Also, is there a way to close the SMP version besides just closing the cmd window? I close it whenever I game or need to do some heavy work on the computer and it says that I improperly terminate it... never had this problem with console, but it doesn't appear to hurt anything.
Click to expand...

For the SMP window, just do Ctrl-C before closing and it will get closed properly. However, no matter how you close, it will say it is closed improperly, which is not always true. I believe there are many useless messages in the SMP client so just ignore them as long as it doesn't have errors.

 
Xilikon said:
For the SMP window, just do Ctrl-C before closing and it will get closed properly. However, no matter how you close, it will say it is closed improperly, which is not always true. I believe there are many useless messages in the SMP client so just ignore them as long as it doesn't have errors.

Click to expand...

Rgr thanks, it seems to close what it was doing and saves the progress, but I just noticed
Examination of work files indicates 8 consecutive improper termination.
Click to expand...

What is the deadline time for the console stuff for my p4? Or where can I find that info?
 
Tigerbiten said:
Points, deadlines, etc are found at http://fah-web.stanford.edu/psummary.html
The other page you'll need is http://fah-web.stanford.edu/serverstat.html as it shows which of the servers is overloaded or down when you cannot send work in or get fresh work out.

Luck ......... :D
Click to expand...

Ah nice TY :)

So according to that, I have 30 days (preferred) to finish my piece, or is that when they want all pieces in? Or I guess it doesn't really matter since multiple pieces will be given to multiple people right?

p2418_Ribo_Mg

So even though it might take a few days, I have weeks to get it done right? And that box is just running 24/7 folding atm.

Started up a second instance on it :). Thanks everyone for the help and helping me understand more about how it all works :)
 
jmackay said:
Ah nice TY :)

So according to that, I have 30 days (preferred) to finish my piece, or is that when they want all pieces in? Or I guess it doesn't really matter since multiple pieces will be given to multiple people right?

p2418_Ribo_Mg

So even though it might take a few days, I have weeks to get it done right? And that box is just running 24/7 folding atm.

Started up a second instance on it :). Thanks everyone for the help and helping me understand more about how it all works :)
Click to expand...

The deadlines is always in relation to your own work unit (what pieces is called exactly), not for the whole lot. So, from the time you downloaded it, you have 30 days to return it. However, with the SMP client, the deadlines is much tighter with onlu 3 or 4 days to return, depending on which WU. This is the price to pay for the huge points bonus of the SMP client.

 
You must log in or register to reply here.
Back
Top