http://fah-web.stanford.edu/psummary.html



SMP seems to be the way to go. If you have a dual core it is probably worth trying to get it set up. It looks like it will take a bit more management. So if you don't want to manage it, then the console or service might be better.

Page 35

6/22/2007 10:46:35 AM

I saw the headline @tww, started my machines back folding again. Though how long does it take for the sig to update (below)? Extremeoverclocking stats show my submitted WUs tho.

6/22/2007 10:57:11 AM

^^I've been doing it for a few weeks on my C2D smp. I just had to rebuild it, and build my mom a new computer. Once i get both fully running and stable after OCing, i'll run the smp on both.

6/22/2007 11:07:48 AM

Operator1 is back up and running... sadly not on the 4 computers i had it going on before

6/22/2007 3:44:21 PM

the smp client is good
the g4 is pathetic at this though

6/23/2007 12:17:21 AM

Quote :

"I saw the headline @tww, started my machines back folding again. Though how long does it take for the sig to update (below)? Extremeoverclocking stats show my submitted WUs tho."

IIRC the signatures average one months worth of data, so it will take a bit of time to become accurate. My average is about 1000 points too low. The signature stats had to be reset, so after 1 month they should be accurate again for the daily avg.

6/23/2007 7:00:34 AM

you can go here http://www.syprus.com/fah/ and see when the stats last updated

6/23/2007 12:00:24 PM

did we lose Prospero?

6/23/2007 1:59:52 PM

Yeah, it looks like Prospero's stats are down again. You can always use other signatures such as:

6/23/2007 2:18:32 PM

Nah, Prospero is on a trip right now and will be back in action soon, don't worry.

6/23/2007 4:28:19 PM

6/23/2007 5:18:46 PM

OK I'm back after a long hiatus...

Need some updated info - specifically how to install FAH as a hidden service on a couple of my other computers...one of which is a dual core intel PC. Also, are there still different builds, like gromacs and tinker? Which one is best and where do I get them? Also, is the GPU core worth it yet?

thanks, B

[Edited on June 23, 2007 at 5:55 PM. Reason : ]

6/23/2007 5:40:39 PM

hey guys, sorry bout the long stint away, i was moving from DC to denver, long trip. got everything setup here but my apartment runs a lot hotter than in DC, so until i get my ambient room temperatures down, i won't be able to fold much.

the stats page is up, but for some reason the cron job keeps failing, i'll try to fix it this week.

keep up the great work

6/24/2007 5:00:36 PM

^ I hear ya...the heat thing only helps in the winter We look forward to your return.

Great day today everyone, hopefully we're starting to come out of that funk we've been in lately.

6/25/2007 2:10:06 AM

We've climbed back to 101. Keep it coming so we can break the top 100 again and stay there which is the hardest part.

6/25/2007 9:04:32 AM

The cool part is I think we are one of 2 University based teams to be near our level (Mississippi State and Us. Except for University of Kentucky Systems Labs which seems like a sysadmin using lab resources for folding). The rest are other type of communities. If only we could get the word out to more NCSU students ( and their possible ps3s ).




[Edited on June 25, 2007 at 11:29 AM. Reason : .]

[Edited on June 25, 2007 at 11:35 AM. Reason : ps3]

6/25/2007 11:28:23 AM

NVM,
Got it working on two new c2d machines.

How do you redo the settings on the smp client? Like when you install it the first time.

[Edited on June 25, 2007 at 12:35 PM. Reason : .]

6/25/2007 12:18:24 PM

Quote :

"Need some updated info - specifically how to install FAH as a hidden service on a couple of my other computers...one of which is a dual core intel PC. Also, are there still different builds, like gromacs and tinker? Which one is best and where do I get them? Also, is the GPU core worth it yet?"

To install as a service follow the instructions on the first page of this thread. message_topic.aspx?topic=118820 You can install SMP as a service, but it might take a couple tries. Let us know if you have any problems.

It doesn't look like they have different builds any more...you can get a variety of Work Units assigned to each client. You can download clients from here: http://folding.stanford.edu/download.html

For the GPU client, someone said they average around 400PPD.

Quote :

"How do you redo the settings on the smp client? Like when you install it the first time. How do you redo the settings on the smp client? Like when you install it the first time."

To redo the settings, I make a shortcut to the fah executable, and in that shortcut I go into the properties, in the target box, and add the -config switch. Then I run that shortcut.

Quote :

"Got it working on two new c2d machines. "

Great! Have you checked your current PPD on those two machines? That will be a nice boost when those WUs start to come in.

6/25/2007 12:48:20 PM

^yeah i'm hoping to get 4k+ per day. I have these two c2d's and 2 other p4's.


Alright got both c2ds going. Lets see how fast i get these WUs.

Heh ambient temp in room is bout to go up.

[Edited on June 25, 2007 at 1:02 PM. Reason : .]

6/25/2007 12:51:40 PM

badass, hell even 3k/day would be a massive help.

6/25/2007 6:55:30 PM

I was folding before, and i got atleast 1800ppd on my c2d. Now i have that running, plus my mom's c2d. I also have 2 other pentium 4's going, so that'll add in a few extra 100 ppd. So hopefully i'll get 4k+ ppd like i said. We'll see.

6/25/2007 8:29:42 PM

now i just put FAH on my work computer (Windows) in addition to my home (Linux). im confused about the different clients.

my Linux PC is 1.8GHz, and completes a "250,000 step" WU in about 7-8 days.

my Windows PC is 3.0GHz and it looks like it will complete a "4000 frame" WU in about 90 days (is that right?).

so whats going on here? i want to stick FAH on some other work computers, but since i dont have control over them, im skeptical they will be able to complete a 90 day job without someone fucking with it.

suggestions?

6/25/2007 10:28:01 PM

nothing should take 90 days...that makes no sense. the linux client might take 7-8 days on a 1.8 (even that sounds a little long)...but a P4 3.0GHz should cut thru work units in a day or two. Post your log here and we'll get it figured out.

6/25/2007 10:38:51 PM

well, never mind. it cleared itself up.

after the "first" frame completed (25 mins or so), it jumped to 40/4000 frames completed, and the WU completion time changed from 89 days to just under 2 days.

that does sound a lot better.



[Edited on June 25, 2007 at 10:51 PM. Reason : ]

6/25/2007 10:50:48 PM

Got the new Dell C531 up and folding last night and it's amazing how this Athlon performs against my C2D - and not in a good way.
The Athlon X2 3600+ stock 2.0 Ghz does ~750PPD with the SMP client running on Vista.
The C2D E6320 OC@ 2.0Ghz does in the neighborhood of 1400+ PPD with the SMP on XP.

The big difference to me is that the X2 has a mere 256Kb per core of L2, where the 6320 has 4Mb shared. I never realized such a disparity in performance could be from cache size and possibly manufacturer, but it's something to think about.

Does the OS have anything to do with lacking performance or is this strictly due to being a much cheaper processor?

6/26/2007 9:23:01 AM

^I'm running smp on both xp and vista, and i don't think the PPD is OS dependent. It has much more to do with the clock speed. I'm running a 6320 @ 2.2, and a 6600 @ 3.2, and yesterday morning one was at 65% and the other was at 55%, started them both at the same time. I then oced the 6320 to 2.4, and dced the 6600 to 2.4, and they settled around the same ppd probably because they had the same clock speed, and same 4mg of L2 cache.

[Edited on June 26, 2007 at 9:31 AM. Reason : .]

6/26/2007 9:29:40 AM

Quote :

"The Athlon X2 3600+ stock 2.0 Ghz does ~750PPD with the SMP client running on Vista. The C2D E6320 OC@ 2.0Ghz does in the neighborhood of 1400+ PPD with the SMP on XP."

Yeah the SMP client is all about L2 cache size. The bigger the better...and I bet that trend will stay for future multi-core proccessors/clients too.

http://forum.folding-community.org/forum12,general-fah.html is a good place for info about stuff like that. I found this there, though there are probably better explanations if you look hard enough

Quote :

"Some workunits require the larger cache to run well. I have a suspicion that it might be because those workunits are much more multi-threaded and so you've got two threads on each core that are being very rigorously swapped on and off the core.. Which might be why you need the 4MB cache on a dual-core processor but probably wouldn't on a quad.. Just a guess though.. But for now, get the Intel chip with the 4MB cache for folding knowing full well that if workunits become even more quad-targeted.. What I mean by this is that as the Beta SMP client evolves, there may be even higher amounts of thread-swapping on each core of a dual-core chip. Or, put another way, you could look at it as the 4 threads having a higher dependency on the output from each other.. So that one can't execute unless it has the most recent results from thread two and two can't execute until it has the most recent results from thread three and thread three won't execute until it has the most recent results from thread four. At first this doesn't seem very synchronus, until you throw in that thread 4 may be one step ahead of thread 3 and thread 3 a step ahead of thread 2 and thread 2 a step ahead of thread 1.. So when all threads finish their current step, the results just flow to the next thread instead of having to swap the threads off the cores. The result on a dual-core would be two threads spending most of their time going in and out of the each core rather than actually being processed on some workunits and on other workunits that are less synchronous, the thread swapping will be less and they'll do more work because of less overhead.... Swapping threads on a single core quickly becomes a VERY cache intensive operation (especially as the number of thread swaps increases) and those with little cache would perform very badly (some say almost a 50% drop in performance from 4MB to 2MB on some workunits).. The 4MB cache might not be enough in the near future on a dual-core chip, I think it's risky.. Of course this is all speculation so far, I still await people to post more information about how FAH is performing on quad core processors with little cache (guess I'll have to wait until AMD's chips come out for that!).. I predict, when running on a quad-core the FAH client will do very well with very little cache at all because the threads aren't being swapped rigorously on and off the core and are running fine synchronously even though there is a large amount of dependency between the threads. There was somebody a couple months ago who noticed something interesting.. That on a dual-core processor, over a long period of time.. Measuring how much time each thread is on the CPU with respect to the total time the workunit takes to execute, two of the threads were spending around 35% and the other two at 15% which would imply very little thread-swapping happening.. But on the newer workunits that require 4MB of cache to run well, I'm betting to see 25%/25%/25%/25% with massive amounts of thread swapping on both cores and all threads spending the exact same amount of time being executed.. But the extra cache reduces the overhead of those thread swaps. Just a guess.. (for those a bit slow, the percent I'm using is NOT CPU usage of each thread, rather it is the amount of time each thread spends in execution over the course of making a complete workunit) "

6/26/2007 10:17:43 AM

Any idea why they have the SMP client optimized for 4 cores and not 2? Are most of these clients running on quad-core servers, or am I wrong in assuming that most of it's users are dual-core desktop owners?

It would seem that they would release a client optimized for dual core systems to eliminate the latency of thread-swapping, since so many users in the project are running c2D's and the like.

6/26/2007 12:19:47 PM

^ I think the folding crew is mainly testing on 4core boxes like the Mac Pro from their FAQ. I'm guessing they are trying to be well tested and by the time they are out of beta, 4core boxes might be the norm.

I'm running the smp client only on my quad core boxes right now. The machines of mine with 2 cores and hyperthreading only get the normal client =/ though they seem happier with it. And my dual core pentium d seems happier with 2 normal clients as well. Kinda sucks points wise tho.

[Edited on June 26, 2007 at 2:49 PM. Reason : .]

6/26/2007 2:49:00 PM

your 2 core machines do better running on the normal client?

[Edited on June 26, 2007 at 2:53 PM. Reason : .]

6/26/2007 2:50:10 PM

^ not point wise...

my dual core constantly crashes, or gets a fileio error, etc... but as far as points go, if it does one SMP a week, it is twice as good as the other (and it can do 2-3 smp's a week, just needs constant checking)

6/26/2007 2:53:39 PM

so like a month ago before i even got it on a machine worth a damn i was 167

then i was 200 like 2 weeks ago...now i'm 205

6/26/2007 2:58:34 PM

^ do you leave it on all the time or have it at a small percentage of the CPU?

6/26/2007 3:18:40 PM

^^^ What Lego said.

I'd rather not baby sit the smp client if it wants to punk out. For example, the smp work unit would not even start/initialize on my dual core pentium d. mpiexec.exe would just eat up the cpu rather than start doing any work. I think it had to do with regardless of physical cpu core count, f@h smp always started 4 fahcores. So I just went back to 2 clients at around 500 or so points each based on fahmon (large work units allowed).

[Edited on June 26, 2007 at 3:25 PM. Reason : ^]

[Edited on June 26, 2007 at 3:25 PM. Reason : .]

6/26/2007 3:24:44 PM

^ Yeah i was wondering about that. On both my c2d machines 4 fahcores processes always start up, and each utilize ~25% of the cpu. How does this relate to the cpu being dual/quad core?

6/26/2007 3:27:15 PM

^ there are four processes running, basically if you have a quad core each process will get a core. on the dual core two processes share a core... having four processes sharing two cores is better than using only one of the cores, it is just not optimized for two cores.

it is probably a little more complicated then that, but the big difference is that the points are a lot more. and here i thought ps3 were getting all the points for being faster...

6/26/2007 3:57:10 PM

For quad core machines, there is 1 fahcore per 1 cpu core. On dual core machines, there is 1 fahcore per 1/2 cpu core. And I think that is their reasoning for suggesting running the smp client on quad core boxes. It's just a suggestion though and the smp points are nice

[Edited on June 26, 2007 at 3:59 PM. Reason : ']

6/26/2007 3:59:27 PM

^How do the quad core ppd compare with the dual core ones? Also if the ps3 has 8 cores, why haven't they designed the fah software for the ps3 to get more than a few hundred ppd?

Sorry for all the questions.

[Edited on June 26, 2007 at 4:02 PM. Reason : .]

6/26/2007 4:02:27 PM

Quote :

"I'd rather not baby sit the smp client if it wants to punk out. For example, the smp work unit would not even start/initialize on my dual core pentium d. mpiexec.exe would just eat up the cpu rather than start doing any work. I think it had to do with regardless of physical cpu core count, f@h smp always started 4 fahcores. So I just went back to 2 clients at around 500 or so points each based on fahmon (large work units allowed)."

Thats strange, I run the SMP on the cheapest Pentium D, the 2.8GHz one, and I average 500+PPD with no problems. I complete WU's around every 2-3 days with it (and thats with me using the PC for other stuff from time to time like encoding DVDs). I would run the SMP on it for a day or so and check Fahmon to see how many PPD its generating. As long as you make the deadlines, I'm sure its way more than the standard client would get. Do you know how many PPD you're getting with the two clients?

Quote :

"and here i thought ps3 were getting all the points for being faster..."

Yeah a Core2Duo with a 4MB cache can generate WAY more PPD than a PS3. But the PS3 is a bit easier to get setup

6/26/2007 4:04:10 PM

^It's silly though because the ps3s have so many cores. Although they're older ppc 3.2 chips i believe. So maybe the newer FAH smpt software is only optimized to run on dual/quad core chips, whereas the older ps3 chips have to rely on the slower software (in terms of WUs completed.)

[Edited on June 26, 2007 at 4:10 PM. Reason : .]

6/26/2007 4:08:58 PM

^^ Wow yeah that is strange. I'll dig into it more. My pentium d is 2.8ghz as well. Might have to check the ram with memtest86 later too. I get the worst possible error code from folding though right as it starts up. It has happened on multiple work units:

[22:13:43] Client-core communications error: ERROR 0x1

6/26/2007 6:32:18 PM

Quote :

"^ do you leave it on all the time or have it at a small percentage of the CPU?"

i have the no nonsense version on my personal laptop, and put the regular version on my parents ~3-3.5 year old computer 24/7...i think so many new users have signed up lately that they all have computers that gain more than the 2 i got running can

6/26/2007 6:34:45 PM

^^ I had issues with that error once...I just removed the old installation, dleted all the remaining work files, rebooted, installed again and it worked like a charm.

6/27/2007 8:51:16 AM

so, is there any way to get the SMP to run as a service?

6/27/2007 7:37:54 PM

^ yes, that option is in the setup just like the console version.

6/27/2007 8:05:35 PM

I thought someone mentioned earlier that the SMP service does not terminate properly and can corrupt WUs.

6/27/2007 10:55:21 PM

think my parents turned it off

6/27/2007 10:58:23 PM

^^ I haven't heard that. I run it on a couple PCs and it works fine.

6/27/2007 10:59:39 PM

dammit dammit dammit.

i just found out my work has firewall set so i cant send my completed WUs. i can "get" them, but not send them back.

i found this out after i got it set up on my workstations 3.0GHz P4 c2d, and it completed its first WU.

and theres like a lot of other computers i could have tried to install it as a service on too.

i suspect we have port 8080 blocked. i think its set so only certain things are allowed, and the rest is blocked by default.

now how to tell the Vice Pres i need port 8080 freed up on a permanent basis?





[Edited on June 28, 2007 at 3:23 AM. Reason : ]

6/28/2007 3:21:03 AM

^It's not worth jepordizing your job. Just be careful.

6/28/2007 9:03:55 AM