Page 5 of 56

Posted: Tue Jun 19, 2007 1:37 pm
by aristide1
KansaKilla wrote:aristide1, when you mention that you are hitting a checkpoint every 6 minutes, do you mean that you are going through one percent every 6 minutes, or is it something else? Also, to finish a work unit in 10 hours is pretty fast, what hardware are you running it on? Sorry if you've mentioned it elsewhere--no time to search right now.
Yes 1% every 6 minutes. It's was on a lowly Pentium 4, obviously with a total of 44 points it wasn't much work to begin with. On average I would say it lowered the PPD that that PC usually gets.

No on my Core 2 Duo a couple of SMPs were about 23 minutes each and worth about 1385 points. I like those better.

Posted: Tue Jun 19, 2007 5:00 pm
by kittle
well ive added my processors to the mix here.
xeon 1.8 ghz (x2) runnning 2 instances 24/7 and my pair of opteron 246s on the desktop PC while im sleeping and away at work.

I cant run the GPU client because its win32 only and ATI only - both my video cards are nvidia & the desktop runs XP x64.

but the WUs are cranking away

Posted: Wed Jun 20, 2007 11:24 am
by KansaKilla
Every bit helps, we really appreciate it.
Thanks!

Posted: Wed Jun 20, 2007 3:32 pm
by disphenoidal
kittle wrote:well ive added my processors to the mix here.
xeon 1.8 ghz (x2) runnning 2 instances 24/7 and my pair of opteron 246s on the desktop PC while im sleeping and away at work.

I cant run the GPU client because its win32 only and ATI only - both my video cards are nvidia & the desktop runs XP x64.

but the WUs are cranking away
Is it possible to run the SMP client on those systems, rather than 2 instances? It sounds like the SMP client runs a lot faster than 2 instances.

Posted: Wed Jun 20, 2007 7:48 pm
by NeilBlanchard
Hello,

Yes, IME the SMP client is 2-4X faster than two "regular" F@H clients.

Posted: Thu Jun 21, 2007 3:50 am
by vg30et
Sadly I've had several SMP WU's take >5 days on a dual proc Xeon 2.4GHz machine. From what I have read, I would get no credit for these units as they are over the final deadline. I have been forced to run 4x individual instances on these machines.

Posted: Thu Jun 21, 2007 4:23 am
by KansaKilla
That just happened to me, too. I don't know if I just happened to get a particularly difficult core or something. I went back and re-installed FAH SMP. That seemed to help.

Posted: Thu Jun 21, 2007 4:29 am
by NeilBlanchard
Hello,

Remember, the SMP client is designed for quad-cores, but runs on dual cores -- so even with the Linux SMP client, you only have about 10-24 hours of "cushion". On my dual core X2, the Windows SMP was never able to complete a WU on time, even with (virtually) 100% CPU time.

Posted: Thu Jun 21, 2007 4:39 am
by jaganath
On my dual core X2, the Windows SMP was never able to complete a WU on time, even with (virtually) 100% CPU time.
we should really warn people that unless they have quite a powerful machine and can let it Fold at high utilisation (90%+?) the SMP client is not the way to go.

Posted: Thu Jun 21, 2007 5:35 am
by ryboto
jaganath wrote: we should really warn people that unless they have quite a powerful machine and can let it Fold at high utilisation (90%+?) the SMP client is not the way to go.
My Opteron 170 completed windows SMP WU's with 35% time to deadline remaining, and it was running at stock, 2.0ghz. To me that implies that even the slowest X2 dual core can handle it, I can't speak for Pentium D machines though.

Posted: Thu Jun 21, 2007 9:57 am
by peteamer
jaganath wrote:we should really warn people that unless they have quite a powerful machine and can let it Fold at high utilisation (90%+?) the SMP client is not the way to go.
I think we need to work out what quite powerful is.
i.e. my 4400+ Brisbane is running the SMP client on 64bit OpenSUSE and hasn't got to within 24HR's... yet, closeish on some WU's though(2608,2609)
It's on 24/7 (Apart from when I've killed the MOBO, CPU or No.1 HD... :roll: )

Currently:
Project: 2605
Downloaded: June 20, 03:53
Preffered Deadline: June 23,03:53
Final Deadline: June 24, 03:53

Estimated Deadline: June 21, 23:32,
PPD 994.53
Courtesy of FahMon.



Anybody else got any info to help build a guideline?


Pete

Posted: Thu Jun 21, 2007 11:54 am
by Myth!
Hi Folks and Folkesses,

am running XP SMP client on Asus Striker, 2gb ram and C2D6600 at stock. Have had no instability issues since it was installed (2 months?) and am folding 1% every 15mins on this current 2610 project :lol:

Also running normal XP client on 3.2 prescott, 1gb ram....was reading this thread and it suggested I might be able to run the SMP client on it? can you give details (unless its obvious..ie just install and run it :wink: )

cheers, Myth

Posted: Thu Jun 21, 2007 1:11 pm
by KansaKilla
My "new" Opteron 165 OC'd to 2.7 (wouldn't have tried if I hadn't been drunk, as posted in OT) running Linux SMP on SuSE 10.2 completes 1% in about 18:40 (circa 32 hours - 36 hours for a full WU, depending on the WU). It's using about 1.2-1.3 Gb of ram constantly. The core duo laptop at 1.83 (not core 2 duo) and 1 Gb ram takes about 50 hours to complete an SMP WU using the windows client.

Posted: Thu Jun 21, 2007 1:20 pm
by NeilBlanchard
Greetings,
ryboto wrote:
jaganath wrote: we should really warn people that unless they have quite a powerful machine and can let it Fold at high utilisation (90%+?) the SMP client is not the way to go.
My Opteron 170 completed windows SMP WU's with 35% time to deadline remaining, and it was running at stock, 2.0ghz. To me that implies that even the slowest X2 dual core can handle it, I can't speak for Pentium D machines though.
Does the Opteron have 1MB L2 caches? My X2 4200+ has only 512KB L2 caches...

Posted: Thu Jun 21, 2007 2:36 pm
by Myth!
I was gutted when Nathan breezed past me...am glad I'm catching up again! If I can just beat Kansakilla to that goalpost first...i'll give myself a prize 8)

btw TSC may stand for 'The Source Code' but I'm not sure.

And I think SPCR should get extra points for folding QUIETER than the rest :wink:

Posted: Thu Jun 21, 2007 3:04 pm
by qviri
Myth! wrote:btw TSC may stand for 'The Source Code' but I'm not sure.
http://tsc.overclockers.ru/: "The Successful Crunchers! Russia"

Posted: Thu Jun 21, 2007 4:04 pm
by KansaKilla
Neil,
Both the Opteron 165 and 170 have 2 x 1 MB level 2 cache.

Posted: Tue Jul 03, 2007 9:50 am
by aristide1
Well we keep slipping, we're down to 130 users, and we can't expect vg30et to continue towing the line.

About 6 weeks ago team 18, ABXZone was on our list of teams we could overtake. When I first looked it would take us 8.3 years to catch up to them. :shock:

With progress we dropped that to 2.1 years. Then it jumped up again. Now we either can't catch up to them at all or it will take more than 30 years. Currently we are in no position to overtake anyone. :(

Come on people :!:

Posted: Tue Jul 03, 2007 9:54 am
by aristide1
vg30et wrote:Sadly I've had several SMP WU's take >5 days on a dual proc Xeon 2.4GHz machine. From what I have read, I would get no credit for these units as they are over the final deadline. I have been forced to run 4x individual instances on these machines.
??? Why have I gotten points for work units that have taken 10-14 days to complete ???

Posted: Tue Jul 03, 2007 10:52 am
by kittle
im going to have to look into the SMP client then..

so if i understand correctly, the WINDOWS smp client is faster than 2 single clients (on a dual CPU box) ?

if so, i'll definately switch over, as I only run folding while sleeping and at work.

Posted: Wed Jul 04, 2007 8:46 pm
by ryboto
NeilBlanchard wrote:
Does the Opteron have 1MB L2 caches? My X2 4200+ has only 512KB L2 caches...
yea, I think that's the issue. My lab computer is an x2 4200, and it takes approximately 44min/frame, whereas my opteron at 2.0ghz gets 30min/frame. Almost 15 minutes less on a chip that's 200mhz slower. The cache makes a huge impact, if we don't assume it's memory related.

Posted: Thu Jul 05, 2007 12:22 am
by Plissken
kittle wrote:so if i understand correctly, the WINDOWS smp client is faster than 2 single clients (on a dual CPU box) ?
Yes, I have 2 dual core machines folding, and my PPD more than doubled :shock: when I switched from the standard console to SMP.

Posted: Thu Jul 05, 2007 8:01 am
by aristide1
Hey Plissken,

How the heat from your NB? I'm going to jack up my E6400 once I mod my "utlimate" case. Haha.

Posted: Thu Jul 05, 2007 8:57 am
by Plissken
It's pretty hot, but I'm not sure exactly how hot. Speedfan shows "system" temp gets up to around 52C.
My system has been folding for months now, with no blue screens or errors, so it seems stable enough. :)
I use the stock NB heatpipe heatsink, removed the dumb foil cover, didn't replace the TIM (yet), no fan blowing directly on it, but good airflow.

Posted: Fri Jul 06, 2007 3:27 pm
by kittle
Plissken wrote:
kittle wrote:so if i understand correctly, the WINDOWS smp client is faster than 2 single clients (on a dual CPU box) ?
Yes, I have 2 dual core machines folding, and my PPD more than doubled :shock: when I switched from the standard console to SMP.
Well I setup the SMP client on my main PC last nite. i'll have to see how much faster it goes.

other pc needs .net installed first, that should come this weekend.

Posted: Mon Jul 09, 2007 3:20 pm
by kittle
well after letting the SMP clients run on overnite friday, sat & sun on my 2 PCs, it seems much slower. Gauging by the number of percent points complete, 2 copies of the FAH client run 1.5 to 2x as fast as the smp version on the same PC.

anyone else see stuff like this?

Posted: Mon Jul 09, 2007 3:47 pm
by vg30et
kittle wrote:well after letting the SMP clients run on overnite friday, sat & sun on my 2 PCs, it seems much slower. Gauging by the number of percent points complete, 2 copies of the FAH client run 1.5 to 2x as fast as the smp version on the same PC.

anyone else see stuff like this?
SMP may run slower but it should give you a lot more points than a regular WU. To see your ppd total, try using fahmon (http://fahmon.silent-blade.org/)

Posted: Mon Jul 09, 2007 6:11 pm
by NeilBlanchard
Hello,

If you are running the Windows SMP client on a slower dual core (esp. one with smaller L2 caches), then it may or may not be worthwhile. However, if it has larger caches, and/or you are running the Linux SMP client, then the PPD will be 3-4X higher than two instances of the "regular" F@H client on the same machine.

Posted: Mon Jul 09, 2007 11:44 pm
by kittle
vg30et wrote:
SMP may run slower but it should give you a lot more points than a regular WU. To see your ppd total, try using fahmon (http://fahmon.silent-blade.org/)
ahh so it does.. had to update my old fahMon, and I see the single cpu WUs are 200-600 points, whereas the smp ones are 1500-1700
Definately worth the wait for them to complete

Posted: Tue Jul 10, 2007 5:55 am
by ryboto
I just got two of my friends to start SMP folding for me. I think they have X2's running at 2.2 and 2.4ghz. Small caches compared to my opteron, but my lab computer completes the WU's before the deadline, so I'm not worried. Also, a friend of mine is building a Quad core machine, and has said that he'll fold for me in return for my help assembling. And he's also going to fold with his PS3 under my username. Should be interesting to see how this all adds up.