• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

WTH??? 1QMD w/ht faster than 1 w/oht??

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

the garynator

Chief folding_monkey
Joined
Nov 16, 2002
Location
Neenah, WI
i have 1 qmd running right now and nothing else...it's getting 18minutes per frame on a qmd... here's the kicker...Hyper Threading is on!!! what the hell, i know with HT off at a higher clock speed, it does each frame in about 20-22 minutes... and it reads out as about 320ish ppd on EMIII... right now it's reading out as 350ppd on EMIII... it has also been going for about 3.5% so it has been running long enough to even out. Anyone herd of this? maybe it's just something stupid, but i've never seen it and figured i'd post about it.
 

Attachments

  • emIII.jpg
    emIII.jpg
    60.7 KB · Views: 76
it looks like you have two qmds running? or is that not the case? basically hyperthreading would not make a difference if folding was the only thing running. it is still just one cpu. But when you have other windows services and processes, hyperthreading greatly increases how fast things get done. that's the only reason i can think of for HT helping so much. also there is a decent range on the QMD bigpackets in terms of time per frame. check out chasr's cool xls sheet for that (http://www.tysa.com/15jg52_uploads/fah_production_calc_v2.xls).

btw, much better avatar! i'm glad you cleaned up the corners, it looks great.
 
veryhumid said:
it looks like you have two qmds running? or is that not the case? basically hyperthreading would not make a difference if folding was the only thing running. it is still just one cpu. But when you have other windows services and processes, hyperthreading greatly increases how fast things get done. that's the only reason i can think of for HT helping so much. also there is a decent range on the QMD bigpackets in terms of time per frame. check out chasr's cool xls sheet for that (http://www.tysa.com/15jg52_uploads/fah_production_calc_v2.xls).

btw, much better avatar! i'm glad you cleaned up the corners, it looks great.

lol, i actually never noticed that my avatar was like that... guess i never really looked at it...

Also, only one instance of FAH is running right now, HT is enabled, so it's only using 50% of my cpu... i know HT adds power, but its not very much... Anyway, basically, at the moment it's folding a QMD AT the same speed/faster than it did with HT Disabled. it just doesn't make sense to me lol.
 
Here's an update, still truckin' along(i left the other instance off just to see what happens)

Notice the TaskManager icon in the corner?
 

Attachments

  • emIII2.jpg
    emIII2.jpg
    43.8 KB · Views: 61
According to ChasR, HT will improve memory bandwidth. Your results seem to prove that.

I guess you just proved that ChasR was right (again!) :cool:
 
wow, at this rate i may just fold on one client until I get dual channel...if this keeps going the way it's going.
 
the garynator said:
wow, at this rate i may just fold on one client until I get dual channel...if this keeps going the way it's going.


Unless that is a dual core, you should only fold 1 QMD anyway. Better points & better science. I only fold 2 qmds instance on my 830D, and 920D. All the rest of my P4's are single QMD instance.
 
pscout said:
According to ChasR, HT will improve memory bandwidth. Your results seem to prove that.

I guess you just proved that ChasR was right (again!) :cool:

Does anyone ever doubt him? :p

Tested and approved :thup:
 
veryhumid said:
also, it is saying it is using 50% (because you have two logical cpus), but in fact it is using 100%

I thought it acted just like having dual cpus/dual core so it is only able to use 50% of the cpu...Maybe i've been thinking totally wrong...



wow i feel like a dumb a$$ :bang head :bang head :bang head :bang head :bang head
 
Yeah, HT only acts like 2 cores and windows can't tell the difference so it shows each as only using 50%. I would, and always have thought that a machine would fold faster with HT enabled. This is because folding would be getting 100% of the cpu more often because the other "core" could be used for the times windows services orother programs need some power. Even tho the other core is only simulated it stills helps more than a single core with no HT or else Intel would stop using it.
 
ht is not like dual cores/dual cpus. i think that is what is confusing you. you have one processor. the best thing to do with HT chips is to leave it on and totally forget about it. your chip has one core, but the feature HT allows very substantial increases in computing when you are multitasking. don't think of it as two things working on the same program, what it really does is help your single chip process multiple threads much more efficiently.

it's hard to comprehend just by the fact that when you run folding it says 50%, but it is using 100% of your cpu's resources. ahhhh i can't explain it. anyway leave it on, you get better PPD as you see :D


ahhh! here we go! everyone give this a once over!: http://www.intel.com/technology/hyperthread/index.htm
 
A single process will go faster with HT than without. HT compensates for an inherent P4 performance problem. Windows happens to utilize HT by making things as if there were two processors, shoving simultaneously running processes into little gaps in the pipeline (I forgot what causes these gaps to occur, but either way processor power that could have been used is instead wasted) where without HT those gaps would simply not be used and be wasted use of processor capability.

So, with Windows 2K/XP, let's say we got 1 QMD running on a P4 w/HT. It is using 100% minus the little pipeline gaps that occur here and there. The reason it will run faster however, is because other stuff that's running, even if it is using very little in terms of CPU, will be filling those gaps instead of interfering with the QMD, or the QMD will be filling some, so there's less "this program or that program" and a little more "both at the same time." However, those pipeline gaps only account for several percent performance loss in total (at most) I think, so it isn't anything close to being like 2 CPUs.

Running two QMDs at once on a HT P4 can yield a worthy extra PPD, in my case around 20. However, if other programs are running, like say download/upload stuff, virus scans, or anything that uses a few percent or more CPU, running two will more likely hurt production than help, or just not really help. This is because those other processes are already vying for the pipeline gaps that HT allows to be utilized, thus negating any possible production increase from the second QMD instance ;)

So only run two QMDs if you're running nothing else. AND you have ample dual channel RAM capacity ;)

Atleast that's my understanding of it (very crude understanding it is!)
 
Max0r said:
Running two QMDs at once on a HT P4 can yield a worthy extra PPD, in my case around 20. However, if other programs are running, like say download/upload stuff, virus scans, or anything that uses a few percent or more CPU, running two will more likely hurt production than help, or just not really help. This is because those other processes are already vying for the pipeline gaps that HT allows to be utilized, thus negating any possible production increase from the second QMD instance ;)

There is still a problem with running 2 qmds on a HT p4 ... the pretty marginal, if any increase in ppd does not justify the slowdown in the folding project that it causes. The qmd's are a set of sequenced simulations. The next one in the sequence cannot be assigned until the previous is completed and turned in. So by running 2 concurrently on a HT will nearly double the elapsed time ... which slows down how quickly each stream of the protein project can complete.

so any marginal increase in ppd you may see is counter productive to the primary reason for folding ... which is why stanford asks that we do no do it.
 
yes, i have decided that with one qmd running i'm only loosing about 20-30ppd but i'm keeping my subteam's average more constant because i'm putting one out about every 28-30 hours. and dual instances don't help very much cuz my mem bandwidth is teh suxx0rz right now anyway.
 
Back