SPCR Folds Team Blog

[aristide1]

I wonder how long before GPU folding has some kinda of bonus system?

GTX 470 - 448 shaders? A multiple of 112, I guess that's good for somebody, but certainly not not for me. The 9800 GT crowd?

Anyway no power consumption numbers no deal.

[aristide1]

But as far as ppd for cost goes, you also really want to consider the power used too. From what I've been able to read, if you run bigadv jobs on the more costly i7s, you can get around 20k ppd, and imagining a system with just the processor and maybe on-board video, that would burn maybe 200W worst case, giving you 100 ppd/W, and that's really good. My most efficient machines now are around high 50s ppd/W. With the new A3 stuff and the bigadv units, it might make more sense to run more SMP folding than gpu folding.

This is an interesting scenario. Before if you replaced a CPU with one that's 50% faster you could expect about a 50% increase in PPD. But with bonuses for early completion how does that scale? To 100%? More? Much more?

Is there a point where any additional speed will not yield any more bonus points? At that point do you bother to OC anymore?

Answer:

And the bonus is exponential. A few minutes faster, and the points really increase a lot.

Cordis you're telling me an i860 on a board with integrated video will use 200 watts? (Anandtech agrees with you.) And it will get 20K PPD? That kind of stinks. I have a motherboard and 2 9600GSOs on it, using about 200 watts and they don't get anywhere near 20K PPD. This is rather annoying, I have a substantial hardware investment.

If NVidia doesn't do better PPD-wise than Intel they are going to be hurt. And the i860s and i870s are still 45nm. Wait until they come out at 32nm, and 32 nm chip sets as well.

[cordis]

Well, I'm not sure that an i860 will get 20k, looking at this thread, there's a list at the top that lists how much people are getting per thread, and one guy posted getting 16.6k on a non-OCd i860. Here's the thread:

http://foldingforum.org/viewtopic.php?f=55&t=11314

It's important to note that this is bigadv folding, I'm currently doing A3 folding on my i860, and it gets around 9.2k ppd. I haven't tried doing bigadvs on it, the bigadv jobs seem to require a large memory space, and the i860 is still running 32 bit windows xp, so it can't address more than 4GB. I'm really thinking more about getting an i7-960 with at least 6GB, that should do 20k bigadv jobs. I suppose an i7-920 OC'd to the hilt would work too, but I try to stay away from OCing stuff. Especially processors, I wouldn't want to push the power envelope too much, ppd/watt is my new goal. In terms of silencing too, I'd rather keep stock speeds going. But if bigadv stuff does work out for me, I could see moving from the gpu heavy systems to something like a single gpu and an aftermarket cooler to keep it cool quietly. Currently, I have a system that does about 26k ppd, but it's the 275/295 system, and it's my loudest system by far. With a fast i7, I could make a 26k system with just the processor and a 260.

[NeilBlanchard]

[19:09:03] Completed 250000 out of 250000 steps (100%)
[19:09:05] DynamicWrapper: Finished Work Unit: sleep=10000
[19:09:15]
[19:09:15] Finished Work Unit:
[19:09:15] - Reading up to 21122496 from "work/wudata_07.trr": Read 21122496
[19:09:16] trr file hash check passed.
[19:09:16] - Reading up to 4541240 from "work/wudata_07.xtc": Read 4541240
[19:09:16] xtc file hash check passed.
[19:09:16] edr file hash check passed.
[19:09:16] logfile size: 199556
[19:09:16] Leaving Run
[19:09:18] - Writing 26013076 bytes of core data to disk...
[19:09:18] ... Done.
[19:09:20] - Shutting down core
[19:09:20]
[19:09:20] Folding@home Core Shutdown: FINISHED_UNIT
[19:12:38] CoreStatus = 64 (100)
[19:12:38] Sending work to server
[19:12:38] Project: 2669 (Run 10, Clone 6, Gen 121)


[19:12:38] + Attempting to send results [March 8 19:12:38 UTC]
[19:14:38] + Results successfully sent
[19:14:38] Thank you for your contribution to Folding@Home.
[19:14:38] + Number of Units Completed: 181

[19:14:39] - Preparing to get new work unit...
[19:14:39] Cleaning up work directory
[19:14:41] + Attempting to get work packet
[19:14:41] - Connecting to assignment server
[19:14:41] + No appropriate work server was available; will try again in a bit.
[19:14:41] + Couldn't get work instructions.
[19:14:41] - Attempt #1 to get work failed, and no other work to do.
Waiting before retry.
[19:14:58] + Attempting to get work packet
[19:14:58] - Connecting to assignment server
[19:14:59] + No appropriate work server was available; will try again in a bit.
[19:14:59] + Couldn't get work instructions.
[19:14:59] - Attempt #2 to get work failed, and no other work to do.
Waiting before retry.
[19:15:17] + Attempting to get work packet
[19:15:17] - Connecting to assignment server
[19:15:18] + No appropriate work server was available; will try again in a bit.
[19:15:18] + Couldn't get work instructions.
[19:15:18] - Attempt #3 to get work failed, and no other work to do.
Waiting before retry.
[19:15:45] + Attempting to get work packet
[19:15:45] - Connecting to assignment server
[19:15:45] + No appropriate work server was available; will try again in a bit.
[19:15:45] + Couldn't get work instructions.
[19:15:45] - Attempt #4 to get work failed, and no other work to do.
Waiting before retry.
[19:16:32] + Attempting to get work packet
[19:16:32] - Connecting to assignment server
[19:16:32] + No appropriate work server was available; will try again in a bit.
[19:16:32] + Couldn't get work instructions.
[19:16:32] - Attempt #5 to get work failed, and no other work to do.
Waiting before retry.
[19:18:07] + Attempting to get work packet
[19:18:07] - Connecting to assignment server
[19:18:07] + No appropriate work server was available; will try again in a bit.
[19:18:07] + Couldn't get work instructions.
[19:18:07] - Attempt #6 to get work failed, and no other work to do.
Waiting before retry.
[19:20:58] + Attempting to get work packet
[19:20:58] - Connecting to assignment server
[19:20:59] + No appropriate work server was available; will try again in a bit.
[19:20:59] + Couldn't get work instructions.
[19:20:59] - Attempt #7 to get work failed, and no other work to do.
Waiting before retry.
[19:26:28] + Attempting to get work packet
[19:26:28] - Connecting to assignment server
[19:26:28] + No appropriate work server was available; will try again in a bit.
[19:26:28] + Couldn't get work instructions.
[19:26:28] - Attempt #8 to get work failed, and no other work to do.
Waiting before retry.
[19:37:12] + Attempting to get work packet
[19:37:12] - Connecting to assignment server
[19:37:12] + No appropriate work server was available; will try again in a bit.
[19:37:12] + Couldn't get work instructions.
[19:37:12] - Attempt #9 to get work failed, and no other work to do.
Waiting before retry.
[19:58:47] + Attempting to get work packet
[19:58:47] - Connecting to assignment server
[19:58:47] + No appropriate work server was available; will try again in a bit.
[19:58:47] + Couldn't get work instructions.
[19:58:47] - Attempt #10 to get work failed, and no other work to do.
Waiting before retry.
[20:41:35] + Attempting to get work packet
[20:41:35] - Connecting to assignment server
[20:41:35] + No appropriate work server was available; will try again in a bit.
[20:41:35] + Couldn't get work instructions.
[20:41:35] - Attempt #11 to get work failed, and no other work to do.
Waiting before retry.
[21:29:46] + Attempting to get work packet
[21:29:46] - Connecting to assignment server
[21:29:46] + No appropriate work server was available; will try again in a bit.
[21:29:46] + Couldn't get work instructions.
[21:29:46] - Attempt #12 to get work failed, and no other work to do.
Waiting before retry.
[22:17:50] + Attempting to get work packet
[22:17:50] - Connecting to assignment server
[22:17:50] + No appropriate work server was available; will try again in a bit.
[22:17:50] + Couldn't get work instructions.
[22:17:50] - Attempt #13 to get work failed, and no other work to do.
Waiting before retry.
[23:06:04] + Attempting to get work packet
[23:06:04] - Connecting to assignment server
[23:06:05] + No appropriate work server was available; will try again in a bit.
[23:06:05] + Couldn't get work instructions.
[23:06:05] - Attempt #14 to get work failed, and no other work to do.
Waiting before retry.
[23:54:10] + Attempting to get work packet
[23:54:10] - Connecting to assignment server
[23:54:10] + No appropriate work server was available; will try again in a bit.
[23:54:10] + Couldn't get work instructions.
[23:54:10] - Attempt #15 to get work failed, and no other work to do.
Waiting before retry.
[00:42:19] + Attempting to get work packet
[00:42:19] - Connecting to assignment server
[00:42:19] + No appropriate work server was available; will try again in a bit.
[00:42:19] + Couldn't get work instructions.
[00:42:19] - Attempt #16 to get work failed, and no other work to do.
Waiting before retry.
[01:30:27] + Attempting to get work packet
[01:30:27] - Connecting to assignment server
[01:30:27] + No appropriate work server was available; will try again in a bit.
[01:30:27] + Couldn't get work instructions.
[01:30:27] - Attempt #17 to get work failed, and no other work to do.
Waiting before retry.
[02:18:38] + Attempting to get work packet
[02:18:38] - Connecting to assignment server
[02:18:38] + No appropriate work server was available; will try again in a bit.
[02:18:38] + Couldn't get work instructions.
[02:18:38] - Attempt #18 to get work failed, and no other work to do.
Waiting before retry.
[03:06:39] + Attempting to get work packet
[03:06:39] - Connecting to assignment server
[03:06:39] + No appropriate work server was available; will try again in a bit.
[03:06:39] + Couldn't get work instructions.
[03:06:39] - Attempt #19 to get work failed, and no other work to do.
Waiting before retry.

[cordis]

Is a server down again? That's what happened when the gpu server went down, what kind of job was that?

[NeilBlanchard]

It is a Mac SMP client -- my son's MacBook was doing the same thing as well, yesterday. But this morning, both are back Folding again. So, a temporary glitch in a server, apparently.

[dereksbelanger]

Anyway no power consumption numbers no deal.

Card watts = PCIe slot + PCIe accessory cable

comprehensive list here: http://www.hardware.fr/articles/781-5/d ... iques.html

Awesome resource IMO.

[dereksbelanger]

I have an a_1 core making empty threats....

Code: Select all

[05:46:01] Unit 2's deadline (March 9 04:58) has passed.
[05:46:01] Going to interrupt core and move on to next unit...
[05:46:02] Completed 455000 out of 500000 steps  (91 percent)
[05:46:02] Unit 2's deadline (March 9 04:58) has passed.
[05:46:02] Going to interrupt core and move on to next unit...
[06:50:37] Writing local files
[06:50:37] Unit 2's deadline (March 9 04:58) has passed.
[06:50:37] Going to interrupt core and move on to next unit...
[06:50:37] Completed 460000 out of 500000 steps  (92 percent)
[06:50:37] Unit 2's deadline (March 9 04:58) has passed.
[06:50:37] Going to interrupt core and move on to next unit...
[07:57:19] Writing local files
[07:57:19] Unit 2's deadline (March 9 04:58) has passed.
[07:57:19] Going to interrupt core and move on to next unit...
[07:57:19] Completed 465000 out of 500000 steps  (93 percent)
[07:57:19] Unit 2's deadline (March 9 04:58) has passed.
[07:57:19] Going to interrupt core and move on to next unit...
[09:02:46] Writing local files
[09:02:46] Unit 2's deadline (March 9 04:58) has passed.
[09:02:46] Going to interrupt core and move on to next unit...
[09:02:46] Completed 470000 out of 500000 steps  (94 percent)
[09:02:46] Unit 2's deadline (March 9 04:58) has passed.
[09:02:46] Going to interrupt core and move on to next unit...
[10:09:41] Writing local files
[10:09:41] Unit 2's deadline (March 9 04:58) has passed.
[10:09:41] Going to interrupt core and move on to next unit...
[10:09:42] Completed 475000 out of 500000 steps  (95 percent)
[10:09:42] Unit 2's deadline (March 9 04:58) has passed.
[10:09:42] Going to interrupt core and move on to next unit...
[11:14:54] Writing local files
[11:14:54] Unit 2's deadline (March 9 04:58) has passed.
[11:14:54] Going to interrupt core and move on to next unit...
[11:14:54] Completed 480000 out of 500000 steps  (96 percent)
[11:14:54] Unit 2's deadline (March 9 04:58) has passed.
[11:14:54] Going to interrupt core and move on to next unit...
[12:21:11] Writing local files
[12:21:11] Unit 2's deadline (March 9 04:58) has passed.
[12:21:11] Going to interrupt core and move on to next unit...
[12:21:11] Completed 485000 out of 500000 steps  (97 percent)
[12:21:11] Unit 2's deadline (March 9 04:58) has passed.
[12:21:11] Going to interrupt core and move on to next unit...
[13:26:52] Writing local files
[13:26:52] Unit 2's deadline (March 9 04:58) has passed.
[13:26:52] Going to interrupt core and move on to next unit...
[13:26:52] Completed 490000 out of 500000 steps  (98 percent)
[13:26:52] Unit 2's deadline (March 9 04:58) has passed.
[13:26:52] Going to interrupt core and move on to next unit...
[14:31:39] Writing local files
[14:31:39] Unit 2's deadline (March 9 04:58) has passed.
[14:31:39] Going to interrupt core and move on to next unit...
[14:31:39] Completed 495000 out of 500000 steps  (99 percent)
[14:31:39] Unit 2's deadline (March 9 04:58) has passed.
[14:31:39] Going to interrupt core and move on to next unit...

You will finish your assigned WU and like it - is that understood a_1?!

Sir, yes sir!

[dereksbelanger]

This morning I found that several of my standard clients have unknown projects with unknown points, deadline, client type and core.

A little digging revealed that these are project 1740-1779, WUs....which you wont find in the project list.

However, you can find a description here: http://fah-web.stanford.edu/cgi-bin/fah ... ned?p=1763

Finally some interesting projects for standrd old clients!

These projects use Gromacs Core 3.3 Version 1.93 (FahCore_a0.exe)

Anything is beter than Amber!

[PaleMelanesian]

Me too! This is MUCH better - 800 ppd each client instead of 200 ppd with Amber.

[dereksbelanger]

How are you calculating PPD for these WUs?

In unrelated news...

Power consumption of the new six core Intel processor: http://www.tomshardware.com/reviews/cor ... 73-12.html

[PaleMelanesian]

That's what FahMon tells me. Also, looking at the log files gives me ~50 hours total working time, 2137 points = about 1000 ppd. This is my work machine, though, so it has other tasks that slow it down.

[dereksbelanger]

Even after updating projects HFM and Fahmon both tell me: "Unknow" for credit, PPD and deadline.

[PaleMelanesian]

That's odd. FahMon updated fine for me yesterday. Still, you can look at the log files to see the time for each 1% step, and calculate it out for the 2137 points.

[cordis]

Here's the anandtech review for the 6 core i7:

http://www.anandtech.com/cpuchipsets/sh ... spx?i=3763

Looks pretty good to me, although it will cost some money. I just recently started doing bigadv jobs on an i7-930, and that's running at about 12.5kppd with a video card folding, the card (gtx 275) does about 8k ppd and without it the cpu jumps to 17.5kppd, so it still makes sense to fold on the gpu too. But with a 6(12) core system, you probably wouldn't even lose that much to the gpu. I will really have to make a point of getting my hands on one at some point.

As far as the projects go, I've found that FahMon is usually better at keeping up with new projects than hfm.net, but both can get weirdly out of sync with the servers at times. Not really sure why, I've seen it go for a couple of days and not be able to get new projects, but then they suddenly all show up. Try a shutdown and restart of the monitoring program, and see if that helps.

[aristide1]

Any day now!

http://www.behardware.com/news/10803/bo ... -470s.html

And where will you plug them into? Why here of course:

http://www.behardware.com/news/10808/as ... puter.html

[frenchie]

Any day now!

http://www.behardware.com/news/10803/bo ... -470s.html

And where will you plug them into? Why here of course:

http://www.behardware.com/news/10808/as ... puter.html

Me thinks you'll get an email from stanford like this one :
"too many ppds, please stop folding."

[cordis]

A 4 slot 1156, that's an interesting idea. I think the 1366 boards have more legs, but that's not a bad idea, depending on the price, of course. On the other end of the spectrum is this board:

http://www.anandtech.com/mb/showdoc.aspx?i=3753

A board like this could possibly do bigadv folding, and with boards this small you could but together a mini-itx cluster in a pretty small space. It does seem like this board is power limited, though, so you probably couldn't overclock an i7-860 or something on it, but it's an intriguing idea.

And those nvidia cards, nice! I want one, but I'll wait until some reviews come in. But they exist, awesome!

[cordis]

Hey, check this out:

http://www.engadget.com/2010/03/17/evga ... -exemplif/

Now with dual 6 core (12 thread) processors, and dual gtx 295s, should get at least 58kppd, hmm...
It's a shame that flat fan behind the slots would probably keep the system from being quiet. Although I have noticed that frozencpu.com is selling some noctua nh-12 variants specifically designed for xeons. Very interesting....

[dereksbelanger]

its a good thing they invented folding....gives us something to do with all this ridiculous hardware

[aristide1]

Yeah throw a pair of Intel i7-980X's on that mobo plus 4 GTX480s.

http://www.xbitlabs.com/news/cpu/displa ... ssors.html

EVGA also has a 1200 watt power supply for the setup.

Now how about a high volume discount for electricity?

[aristide1]

Looking at that 7 slot board I thought hmmm, let's fill those slots with GT240 cards (at this time). But despite having lower power requirements than the prior 96 shader model, the 9600GSO, they have these goofy fan setups and most of them are not single slot cards.

[aristide1]

More NVidia info

http://vr-zone.com/articles/nvidia-gefo ... /8635.html

Did I happen to say OUCH!

Bizzare, the power consumption figures are not really any lower than say a GTX295 with 480SP. Same price same power consumption same # of shaders. Big gains for them (only 1 smaller chip, less associated gear to power just one chip, smaller GPU cooler) while SOS for everyone else.

But I will bet the power consumption will be a little lower than max with folding. But we need real world results before we know for sure.

I have 20 bucks that says NVidia ships just 1 to NewEgg on the last day of March, looks at their deadline, and yells, "Mission Accomplished!"

[cordis]

Yeah, I heard about that. Not nearly as impressive as I thought it would be. There is some improvement, 700Mhz for the 480 vs. 576Mhz for the 295, 250W for the 480 vs. 289 for the 295, but that's not nearly enough to make me jump up and buy one. I'd be interested to see what the idle power is like, but that will have to wait for a real review, I suppose. Given that it's a single chip on a single board, it should be a lot easier to come up with a 3rd party heat sink for it, so it could potentially be a lot quieter. On the other hand, the lag on 3rd party vga heat sinks seems to be a couple years. Kind of sad, really. Hope we get an fahaddict review on it to see what the real folding numbers are like. Meh.

[aristide1]

So far the work/watt of 32nm processors doesn't seem all that great either.

[aristide1]

So what would the killer system come to?

$1000 - For that EVGA dual socket board & matching 1200 watt PS
$2000 - 2 i7-980s
$400 to $800 for memory, depending on 12 or 24GB (purchased 2 sticks at a time, the price on 3 stick packages is outrageous.)
Low voltage on these http://www.mushkin.com/Memory/Blackline/996825.aspx
$2000 - 4 GTX280s
$100 - Enterprise HD
$100 - Small SSD for the FAH folders, including the work folder (eliminates seek times)
$200 - heatsinks, fans, misc cooling HW.
$300 - OS with multi processor chip support (XP, Vista, support multiprocessors but only on 1 chip).

Expensive box

[aristide1]

Nice power consumption numbers on the 980

http://www.anandtech.com/cpuchipsets/sh ... =3763&p=14

[cordis]

Well, for the perfect system you can cut the cost of the enterprise hd and the os, I have my fastest system running on 30GB ssd and linux.

Speaking of which, funny thing happened today. I didn't get a chance to read the paper yesterday, so I opened it up today and checked out the friday Fry's ad. To my astonishment, they were having a sale on the i7 980x! Now I figured they had probably run out by now, but I had to get some batteries anyway, so I went by. I asked if they had any left, and they had one left! So I got it! I have two 1366 motherboards, so I took the slowest one, the i7-930, out and stuck the 980x in, did a little bios update, and now it's running smoothly. I've started running it without the gpu folding on, and currently it's doing 28,900 ppd! The kill-a-watt is saying it's using 252W total, and that's 114.7 ppd/W! I'll try turning on the gpus later, and see what the combined ppd and power usage looks like, but for now, I'm loving this thing! I'll put up more data when I have it.

[frenchie]

Grrrrrrrr !!!
3000 points lost... My PC froze and after I rebooted, the GPU client had lost my ID and the team ID... I noticed it at the end of the day...
Oh well

Aristide : Expensive box that's for sure ! But the amount of ppd is so huge !! This thing could fold only a couple of hours a day and still make it to the top 10 producers of SPCR...

[cordis]

Sorry to hear about your lost time there, that sucks.

But in other news, I got more data from my new system! It turns out that when it got started on a fresh WU (check my stats to see what happens when two bigadv WUs finish the same day), the ppd went up even more, even with the two 295 gtx jobs on. So the data is kind of strange, when the next one gets checked in, I may try to shut down the 295 again to see if I get even more throughput, but in the meantime it looks like this:

bigadv ppd: 31134.1
gtx 295 pdd: 7353.4 + 7434.2 = 14787.6
total ppd: 45921.7
power consumption: 430W
ppd/watt = 106.8 !!!!!

Now if I shut down the 295, the bigadv ppd would go up, but even without that, if it was still 252W, it would get at least 123.5 ppd/W!! I'm already in the process of shutting down one machine, and I may be able to shut down another one. Or maybe shut down a bunch of the gpus, haven't decided yet. Either way, I should be able to drop my power consumption quite a bit!!!