WU's 1010x

Moderators: slegrand, Site Moderators, PandeGroup

WU's 1010x

Postby Jester » Tue Feb 09, 2010 11:41 am

Been getting a few of these lately, and they obviously stress the hardware a little more, which isn't an issue with watercooled cards,
but power consumption is,
running 3 of these Wu's on a rig with 3 x GTX275's sees a slight drop in ppd, again, not a real issue, until I checked the little "power monitor" gadget that still happened to
be connected to the rig, and I'm seeing close to 100W more than I was used to seeing with previous Wu's (over the 3 cards) drawn from the mains,
being only one of 3 rigs with similar spec's, getting these Wu's across all rigs would mean an extra 300W per hour, or a little over 7Kw/h per day extra, and with current costing
that means close to $50 a month extra... for what ?
If it's not possible/desirable to add a little bonus for the power usage, similar to those added for the upload of large Wu's etc, then at least give a choice by making them available
only when running a switch like "advmethods",
With power cost escalating, and the whole "carbon" thing of power production, plus an economic downturn, I feel it hard to justify these new Wu's,
just "rolling them out" might be the last straw for some otherwise dedicated Folders come power bill time....
Not for me, at least for the moment, but few have "bottomless" wallets... :roll:
Jester
 
Posts: 150
Joined: Sun Mar 30, 2008 1:03 pm

Re: WU's 1010x

Postby toTOW » Tue Feb 09, 2010 1:42 pm

You just discovered that previous WUs weren't using you GPU at full potential ... :roll:

The official TDP for the GTX 275 is around 225W depending on the clocks ... with that said, you know what to expect from your card when used at full potential.
Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.

FAH-Addict : latest news, tests and reviews about Folding@Home project.

Image
User avatar
toTOW
Site Moderator
 
Posts: 8792
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France

Re: WU's 1010x

Postby Wrish » Tue Feb 09, 2010 4:39 pm

See if you can't undervolt those cards. Stock voltages are usually set high for stability against manufacturing variations, and power is proportional to voltage squared. I know Rivatuner works on Volterra voltage regulators, see this post on what happened when I tried it on an original GTX280: viewtopic.php?f=52&t=7965&start=255#p130357. My wall socket meter is down 65-70W on the same WU, same clocks/PPD off just one GPU.

Found a much more detailed post for multi-GPU installations:
http://www.ocxtreme.org/forumenus/showthread.php?t=4427
Wrish
 
Posts: 319
Joined: Thu Jan 28, 2010 5:09 am

Re: WU's 1010x

Postby 7im » Tue Feb 09, 2010 4:50 pm

Pande Group is continually looking for ways to improve the code to speed up the calculations, and/or to simulate larger proteins. It is well known that the CPU client was able to push CPUs beyond their typically stable overclock settings. People even starting testing system stability using F@h instead of programs like Prime95 because Prime95 just didn't stress test the CPU as well as F@h could.

And it looks like PG is continuing to improve the ability of the GPU clients to more fully utilize the folding power of video hardware. Just as folding enthusiasts adjusted to CPU client improvements, predicted GPU improvements that have been a long time in the making. We shouldn't be too surprised, and we will keep adjusting as clients keep improving. ;)
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: WU's 1010x

Postby JimF » Tue Feb 09, 2010 5:26 pm

7im wrote:Pande Group is continually looking for ways to improve the code to speed up the calculations, and/or to simulate larger proteins.

I would like to see an option to receive work units that use more graphics memory, as I will be devoting several cards to Folding that will have 1 GB each. It would maybe be like the "work assignments greater than 10 MB" option now. Especially with the Fermi cards coming up, they will have to think of some way (manual or automatic) to do it.
GTX 970 (i5-3550), GTX 980 (i7-3770); Win10 64-bit; FAH 7.4.4
JimF
 
Posts: 499
Joined: Thu Jan 21, 2010 2:03 pm

Re: WU's 1010x

Postby 7im » Tue Feb 09, 2010 5:31 pm

Yes, a Small/Normal/Big option for the GPU client sounds reasonable. I have seen this suggested before, but always good to bring it up again. Let's see what GPU3 brings us. ;)
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: WU's 1010x

Postby bruce » Tue Feb 09, 2010 6:50 pm

I don't believe that the Small/Normal/Big settings are directly related to GPU RAM use. The size of the WU is based on the sizes of the download and upload files as well as CPU RAM. Vram has never entered into that setting. Surely a protein with a large atom-count will use more Vram and will probably have a larger download size, but the size of the upload depends also on what type of data is collected from the run and probably on how many steps are simulated, as well, neither of which is in any way related to Vram demands.

Allowing Big WUs certainly might increase the chances of getting a WU that makes use of more vram, but you really don't have much control over it.
bruce
 
Posts: 22853
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: WU's 1010x

Postby Jester » Wed Feb 10, 2010 10:59 am

toTOW wrote:You just discovered that previous WUs weren't using you GPU at full potential ... :roll:

The official TDP for the GTX 275 is around 225W depending on the clocks ... with that said, you know what to expect from your card when used at full potential.


I know that, that's why there's a 1Kw Corsair Psu running the 3 GTX275's :roll:
Just checked the rig tonight, and it's currently running 3 smaller 353 point Wu's and down a little over 100W in power consumption, but up over 1,000ppd ?
So the newer Wu's are "more fully using my Gpu's", where's the evidence in the only way the average Folder can see.... ppd ??
I know it's a different scenario, but if I donated xx$ to a charity on a monthly basis, then saw the same amount doing less, I think at least asking "why" is a fair enough question ?
Re voltage regulating, the 280, the 295 and some 285's have the Volterra regs (even my old GX2's), generally the 275's don't, mine at least do not...
Jester
 
Posts: 150
Joined: Sun Mar 30, 2008 1:03 pm

Re: WU's 1010x

Postby toTOW » Wed Feb 10, 2010 11:11 am

Jester wrote:So the newer Wu's are "more fully using my Gpu's", where's the evidence in the only way the average Folder can see.... ppd ??


Search on the forum about this, it has been extensively discussed multiple times : this has only to do with benchmark and GPU architecture.

The benchmark is an ATI, which is doing very well on big WUs (ie atoms), but definitely not on small ones. The nVidia architecture usually react in an opposite way : it's doing very well on small WUs, and not as well when things get bigger.

The same applies to every client, because of the nature of the points ... unless you have the exact same machine as the benchmark, your hardware will react differently, and you have to deal with that.

And it's not likely to change (no they won't change the benchmark hardware, as the same issue would arose whatever benchmark hardware you use) ... these are the rules of the FAH point system.
User avatar
toTOW
Site Moderator
 
Posts: 8792
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France

Re: WU's 1010x

Postby Jester » Wed Feb 10, 2010 12:02 pm

toTOW wrote:
Jester wrote:So the newer Wu's are "more fully using my Gpu's", where's the evidence in the only way the average Folder can see.... ppd ??


Search on the forum about this, it has been extensively discussed multiple times : this has only to do with benchmark and GPU architecture.

The benchmark is an ATI, which is doing very well on big WUs (ie atoms), but definitely not on small ones. The nVidia architecture usually react in an opposite way : it's doing very well on small WUs, and not as well when things get bigger.

The same applies to every client, because of the nature of the points ... unless you have the exact same machine as the benchmark, your hardware will react differently, and you have to deal with that.

And it's not likely to change (no they won't change the benchmark hardware, as the same issue would arose whatever benchmark hardware you use) ... these are the rules of the FAH point system.



Thanks, I understand what you're saying, but with a limited budget for power costs I have little alternative, if these "bigger" Wu's become more prevalent, but to shut down one rig to keep things in check,
maybe with "Gpu3" there may be a way to delegate Wu's to the hardware that can run them more efficiently, "fingers crossed"...
Sadly, if this can't be easily achieved, then there's little option left but to cut production to an affordable level,
That's on a purely personal note, but if my situation is mirrored by others how much science will actually be gained by these "bigger" Wu's ?
only time will tell I suppose.
Jester
 
Posts: 150
Joined: Sun Mar 30, 2008 1:03 pm

Re: WU's 1010x

Postby toTOW » Wed Feb 10, 2010 2:06 pm

Also, don't forget that the original purpose of the project is to use idle power of the machines ... if you start building dedicated rigs, it's definitely not prohibited, but no one will blame you if you have to stop a machine because you can sustain the costs. You might also want to look at the efficiency (PPD/W) for you future hardware.

The future is definitely bigger and bigger WUs ... look at the development of the CPU clients : SMP showed a progression from 100k atoms WUs to 1M atoms ones ("BigAdv"). The same progression will probably be seen on GPU clients too.
User avatar
toTOW
Site Moderator
 
Posts: 8792
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France

Re: WU's 1010x

Postby Jester » Thu Feb 11, 2010 1:34 pm

toTOW wrote:Also, don't forget that the original purpose of the project is to use idle power of the machines ... if you start building dedicated rigs, it's definitely not prohibited, but no one will blame you if you have to stop a machine because you can sustain the costs. You might also want to look at the efficiency (PPD/W) for you future hardware.

The future is definitely bigger and bigger WUs ... look at the development of the CPU clients : SMP showed a progression from 100k atoms WUs to 1M atoms ones ("BigAdv"). The same progression will probably be seen on GPU clients too.



I've kept a close eye on ppd/Kw, that was one of the driving forces behind retiring 2 9800GX2's in favour of 2 x GTX295's, at the time, projecting forward 12months of running costs vs ppd production, I could sell off
the GX2's and buy the 295's and still be ahead at the end of the year,
Maybe for now a "wait and see" approach until Gpu3 is released and it's effect on Ati hardware is known, and possibly the release of the "300" series from Nvidia, as you say, Wu's get bigger and bigger, and there's a point where
"last generation" cards cease to be cost effective.....
Seem to be getting a mix of Wu's for now, so the "big Wu" impact on power may be still within budget if I don't get too many (fingers crossed again) :wink:
Jester
 
Posts: 150
Joined: Sun Mar 30, 2008 1:03 pm


Return to NVIDIA specific issues

Who is online

Users browsing this forum: Majestic-12 [Bot] and 3 guests

cron