I was just wondering if 2 clients with HT disabled was as fast as 3 client using HT
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited April 2004
OK, that's a very good question. I don't have an answer. Sorry! Also, I'm not yet sold on 3-client Folding.
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited April 2004
So, try three on one CPU, two on other. PLEASE stop there unless you have lots of backups, please, or want to see Windows XP pretend it is on a 133 MHz processor for a LONG while while you VERY slowly fix undocumented things.
Only three for me, all set at low priority. There is no hang or slowdown with other programs or apps at all. Multitasking is just as robust as it was before - one advantage of hyperthreading, IMO. CPU usage is evenly split between all three clients.
If you have the verbosity 9 switch after a WU finishes it will give you a performance fraction of hom much time is left before the WU will be reissued.
[23:13:12] Folding@home Core Shutdown: FINISHED_UNIT
[23:13:16] CoreStatus = 64 (100)
[23:13:16] Sending work to server
[23:13:16] Unit 2 finished with 97 percent of time to deadline remaining.
[23:13:16] Updated performance fraction: 0.978005
On my hyperthreaded Xeons I set the affinity each to their own CPU.
Seems to work better for me.
"g"
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited April 2004
After analyzing the production from three simultaneously running clients for three days, I've decided to revert back to only two clients for the HT P4 rig. Bottom line is that the three were not producing any more than the two.
I shut down my third client a couple days ago. My perf. frac. droped below 94 and I did not get any of the doulble gromacs, and with single gromacs it was slower with 3 ( less points ) So it seems to be a double edged sword. with the dbl. gromacs 3 clients bumped my daily point production from around 115-120 to 130-138. But dropped my performance fraction enough that I did not get anymore. hurumph !
Now ...of the 8 clients I have on 7 machines all but 1 are running the 520's
The P4's daily total is 115 with those.
I guess when my perf. frac. goes back up I will get more. I actually liked them. My typical full load temp was 43C , but running the doubles it had dropped to 38C.
Scott
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited April 2004
It was really difficult for me to gauge 3 vs. 2-client production. I think it is accurate to say though, that it was about dead even. On two of the clients, performance fraction did not drop below 96.5%. I don't know about the third, as I never got a fix on it. I didn't even know about the perf. frac. thing until about one day before I went back to 2 clients. The third client at that time was running a 400-unit (s)Tinker; I was not about to wait for it to finish to obtain the performance data. I cleaned out the third client folder's contents and put them in a temporary folder. When one of my other clients had just finished an assignment, I dumped the Tinker into that Folder and let it finish. I honestly believe this machine would have maintained at least 95% perf. frac. with all three clients running; I just didn't see a point in the complexity of three manual-config startups when the end result was at 1:1 ratio with my previous two-client setup.
If I do get any more of those Gromacs that fluctuate so greatly with CPU utilization, I will again invoke three-client Folding.
0
Straight_ManGeeky, in my own wayNaples, FLIcrontian
It was really difficult for me to gauge 3 vs. 2-client production. I think it is accurate to say though, that it was about dead even. On two of the clients, performance fraction did not drop below 96.5%. I don't know about the third, as I never got a fix on it. I didn't even know about the perf. frac. thing until about one day before I went back to 2 clients. The third client at that time was running a 400-unit (s)Tinker; I was not about to wait for it to finish to obtain the performance data. I cleaned out the third client folder's contents and put them in a temporary folder. When one of my other clients had just finished an assignment, I dumped the Tinker into that Folder and let it finish. I honestly believe this machine would have maintained at least 95% perf. frac. with all three clients running; I just didn't see a point in the complexity of three manual-config startups when the end result was at 1:1 ratio with my previous two-client setup.
If I do get any more of those Gromacs that fluctuate so greatly with PSU utilization, I will again invoke three-client Folding.
Well, that shows you have two very well configged clients-- they are taking all the box has to give, instead of having to share with a third client.
But, the exercise was interesting for me, nice to know how things work under the hood. Now, throw a 4 GHZ CPU in it, three will fold nicely... More extra capacity, upon which folding THRIVES...
Thank God, they have not come out with an O\S that has a minimum CPU requirement of 2 GHz yet, I would BAN such from any of my systems as bloated-hog ware and sell it immediately to a rendering house-- aka a hotdog or pork sausage processing plant....
John D.
0
LeonardoWake up and smell the glaciersEagle River, AlaskaIcrontian
edited April 2004
Hyperthreading turned on folds two clients at maximum efficiency; when turned off it's a real dog for Folding purposes. This P4C/I875 has always folded two clients at 100% utilization for both the hardware CPU and the second (virtual) CPU. I never would have considered a three-client setup had it not been for those strange Gromacs that fluctuated so greatly on CPU usage.
Now, throw a 4 GHZ CPU in it, three will fold nicely...
I'm only 500MHz away from that.
0
Straight_ManGeeky, in my own wayNaples, FLIcrontian
edited April 2004
Yeah, if each client fluxes at different times, fine.... THEN, you would get help out of three, as you thought. BUT, if any two flux at once, you get very close to unity result between two and three per pipe in terms of hourly\daily productivity-- I understand...
I'm only 500 MHz away from that....
BUT, BUT..., XP wants 1 GHz of it.... When you have 1 GHZ for XP, one GHZ each for three, then you can fold 6 per box.... (aka j/k and pushing gently). Yeah, my lowly Northwood (for now) box is running one at about 3.5 GHz in reality now..... Some Doofus at NewEgg thought a P4 would be a nice thing to put on sale for one day-- look at link below.... Talk about not-buying torture test....
If only I could spend that now without my wife knowing!
Since they are likely to do it again as the 3.4 GHz EE Prescotts come out in more than scarcer-than-hens-teeth quantitiies, I am not TOO upset, but still.... Did you notice the 90 nm spec and the L2 on-die-cache spec on that chip??? And that some DOOFUS forgot to raise the price at 16:00 PDT???? OOOPS! Just had to sneak that price in somewhere....
The advantage with more than 2 clients only comes when you have the DGromacs. Thats when 2 cant keep the CPU full so it needs another to keep it going.
I could recommend the 2 normal clients set to low and a Timeless Tinker on idle that you can start up if the CPU usage dips below 100% when you get a DGromacs.
Comments
[23:13:12] Folding@home Core Shutdown: FINISHED_UNIT
[23:13:16] CoreStatus = 64 (100)
[23:13:16] Sending work to server
[23:13:16] Unit 2 finished with 97 percent of time to deadline remaining.
[23:13:16] Updated performance fraction: 0.978005
Seems to work better for me.
"g"
Now ...of the 8 clients I have on 7 machines all but 1 are running the 520's
The P4's daily total is 115 with those.
I guess when my perf. frac. goes back up I will get more. I actually liked them. My typical full load temp was 43C , but running the doubles it had dropped to 38C.
Scott
If I do get any more of those Gromacs that fluctuate so greatly with CPU utilization, I will again invoke three-client Folding.
Well, that shows you have two very well configged clients-- they are taking all the box has to give, instead of having to share with a third client.
But, the exercise was interesting for me, nice to know how things work under the hood. Now, throw a 4 GHZ CPU in it, three will fold nicely... More extra capacity, upon which folding THRIVES...
Thank God, they have not come out with an O\S that has a minimum CPU requirement of 2 GHz yet, I would BAN such from any of my systems as bloated-hog ware and sell it immediately to a rendering house-- aka a hotdog or pork sausage processing plant....
John D.
BUT, BUT..., XP wants 1 GHz of it.... When you have 1 GHZ for XP, one GHZ each for three, then you can fold 6 per box.... (aka j/k and pushing gently). Yeah, my lowly Northwood (for now) box is running one at about 3.5 GHz in reality now..... Some Doofus at NewEgg thought a P4 would be a nice thing to put on sale for one day-- look at link below.... Talk about not-buying torture test....
http://www.newegg.com/app/viewproduct.asp?description=19-116-170&DEPA=0
Some SCARY-low pricing for retail and OEM....
John D.-- who is :bawling: (almost) because he cannot afford one until NEXT week.....
If only I could spend that now without my wife knowing!
Since they are likely to do it again as the 3.4 GHz EE Prescotts come out in more than scarcer-than-hens-teeth quantitiies, I am not TOO upset, but still.... Did you notice the 90 nm spec and the L2 on-die-cache spec on that chip??? And that some DOOFUS forgot to raise the price at 16:00 PDT???? OOOPS! Just had to sneak that price in somewhere....
John D.
I could recommend the 2 normal clients set to low and a Timeless Tinker on idle that you can start up if the CPU usage dips below 100% when you get a DGromacs.