It Lives, It Lives !!

Message boards : Number crunching : It Lives, It Lives !!
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · Next

AuthorMessage
doug
Volunteer tester

Send message
Joined: 10 Jul 09
Posts: 202
Credit: 10,828,067
RAC: 0
United States
Message 1228988 - Posted: 8 May 2012, 3:00:12 UTC

The Monster's cooling system intrigued me. I have to do something to my P4/gt430 rig. When betreger starting making fun of of it I thought I'd better take a look. From what I've seen it should be capable of 3000-3500 RAC. The first thing GPU-Z showed me was that the GPU wasn't being pushed. I installed Process Lasso and set it to change the cuda apps to high priority. That pushed the GPU usage to 85-90%. Perfect. Then I started overclocking. Didn't take long before my GPU temps were topping 90c, so I backed down to the default again. The temp in my house hit maybe 25c and when I checked on it after work the GPU was running 90-95c. It's not going to last long like that. I've got a CPU fan, the power supply fan and a 80mm case fan in there. The 80mm seems pretty useless, but I guess you get what you pay for. I have space for a couple of good fans and I'm willing to drill. Any suggestions? I'm new to this. The house will never hit higher than 30c.

By the way betreger has a gt430 attached to a dual core and he's getting less RAC than me. Just fess up and ask for help!

Doug
ID: 1228988 · Report as offensive
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1817
Credit: 262,693,308
RAC: 44
Australia
Message 1229010 - Posted: 8 May 2012, 5:01:43 UTC - in response to Message 1228988.  
Last modified: 8 May 2012, 5:08:10 UTC

Hi Doug
I'm not familiar with the GT430 but I do know that a P4 can drive any GPU up to a GTX470 at full blat and will run a GTX580 with 5% speed loss.

I'm not familiar with your card, does it have a fan or is it passively cooled?
Are you running the GPU fans (if any) on Auto or manually set to 100% ?
"Side on" or "side off" ? :-)

Can you post the make and model of your card so I can get a look at it ?

Some observations on P4 CUDA crunching.
On my regular and experimental systems, running a single task per GPU gives between 90 and 95% GPU usage. Even running through the PCI bus a GTS250 (470GFlops) ran at "normal" speed (approx 18 minutes for "standard" 0.4AR unit).

The Win7 video system is less efficient at SAH crunching than XP so going back to XP on that rig could get you up to an additional 5% GPU usage.

Memory is KING-- Recommended is a minimum of 750MB of RAM. I found you can get away with 500MB but you have to strip the OS absolutely bare and have SAH as the only running program. On "The Monster" there are only 26 services running with SAH shut down.

Run GPU tasks only. In this type of crunching the CPU is used only to feed the video card. Keeping the GPU fed and watered keeps the CPU quite busy enough, (I found one GPU task required a constant 25% CPU on a 3GHz P4). Trying to run CPU tasks as well just binds the whole thing up.

This could be the reason why yours and betreger's GPU tasks are taking so long. They are the equivalent of times I used to get with an 8600GT, and I presume the 430 should have more grunt than that.

BOINC Manager and explorer.exe (the desktop) can get very memory and processor hungry at times, between them they can use around 150MB of memory and >50% of processor time on a low end system. Shutting them down saves heaps of resources. (You can restart them at any time by using the "File->Run" tab in Task Manager.)

HTH
T.A.
ID: 1229010 · Report as offensive
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1817
Credit: 262,693,308
RAC: 44
Australia
Message 1229020 - Posted: 8 May 2012, 5:52:46 UTC
Last modified: 8 May 2012, 6:04:14 UTC

A Question.
What's the expected RAC from a 560Ti ?

Just fantasizing what this beast could do with 7 of them in it....

Bwahahahaha

T.A.
ID: 1229020 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13715
Credit: 208,696,464
RAC: 304
Australia
Message 1229036 - Posted: 8 May 2012, 8:34:19 UTC - in response to Message 1229020.  

A Question.
What's the expected RAC from a 560Ti ?

Around 21,000. That's how much higher my E6600 is with a GTX560Ti in it (Factory overclock model).

I think Jason has a GTX680 these days. I remember him mentioing it putting the GTX560Ti to shame for it's output, and while using very considerably less power.
If i had the money, and it were actually possible to buy one, a GTX690 or 2 would be a very good investment.

Grant
Darwin NT
ID: 1229036 · Report as offensive
tbret
Volunteer tester
Avatar

Send message
Joined: 28 May 99
Posts: 3380
Credit: 296,162,071
RAC: 40
United States
Message 1229038 - Posted: 8 May 2012, 8:47:38 UTC - in response to Message 1229020.  

A Question.
What's the expected RAC from a 560Ti ?

Just fantasizing what this beast could do with 7 of them in it....

Bwahahahaha

T.A.


It isn't that nobody wants to answer you. It's just that nobody believes you could get 7 560ti's to run long enough without errors for it to matter. ;-)

Something over 20k and less than 30k each, probably less than 25k each.

ID: 1229038 · Report as offensive
Profile red-ray
Avatar

Send message
Joined: 24 Jun 99
Posts: 308
Credit: 9,029,848
RAC: 0
United Kingdom
Message 1229042 - Posted: 8 May 2012, 8:55:24 UTC - in response to Message 1229036.  

I think Jason has a GTX 680 these days. I remember him mentioing it putting the GTX560Ti to shame for it's output, and while using very considerably less power.

Yes Jason has a 680. With 7 x 680 the RAC would be over 200K but the current CPU would not fast enough to feed them.

How hot do all the GPUs get on this system and how do you monitor them? I suspect SIV would do OK, but have never tested it with more than 4 GPUs.
ID: 1229042 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1229046 - Posted: 8 May 2012, 9:17:09 UTC - in response to Message 1229038.  

It isn't that nobody wants to answer you. It's just that nobody believes you could get 7 560ti's to run long enough without errors for it to matter. ;-)

Something over 20k and less than 30k each, probably less than 25k each.


FWIW, to fill in the picture with those: The 'main' problem turned out to be the horrible drivers that came with these (266.66 IIRC), that corrupted a Cuda compute cache in Windows... I never had that issue, probably as Windows automatically put in a newer WHQL driver, so it was a toughie to find, and I found a way to force the drivers to use 'my' kernel binaries instead of driver runtime produced ones... overall reliability has improved significantly since that time on many cards.

This was followed by noticing a strange coincidence where at least 3 users buying
560ti's, also purchased the same model of SSD... all of which had bung firmware. The symtom was bizarre file access problems encountered only with Cuda apps, and periodic bluescreens &/or crashes. likely paging / virtual / VRAM related. In those cases I understand a SSD firmware flash fixed that.

The last 'main' issue was that a few models/brands were selling factory overclocked models with insufficient core voltage.

I think, IIRC, our poor buddy Brad (Slavac) copped all three of these issues. He's now an appointed guinea pig, because anyone that can find all the problems that quickly is an excellent tester ;)

560ti is an excellent card, and has been from the start. The technology from that refinement clearly went into the 680/690 GPU design... so funnily enough some of my test code is scaling really well. Just goes to show experience down in the trenches can teach you a lot.. .the hard way of course ;)

Jason

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1229046 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1229091 - Posted: 8 May 2012, 12:50:40 UTC - in response to Message 1228724.  

"His name is Robert Paulson!"


ROFL!

If you don't name your machine that, I'm going to use that name on a future build. :-)

I'm pretty sure I don't own any sort of rights to that name. If I do then some movie company owes me quite a lot of money.
Generally I use names that come from the same movie as my user name.If I were to make a big crunchy thing like this then I suppose I might consider giving it some other kind of name. Maybe "Nibbler"
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1229091 · Report as offensive
doug
Volunteer tester

Send message
Joined: 10 Jul 09
Posts: 202
Credit: 10,828,067
RAC: 0
United States
Message 1229108 - Posted: 8 May 2012, 13:54:39 UTC - in response to Message 1229010.  

Hi Doug
I'm not familiar with the GT430 but I do know that a P4 can drive any GPU up to a GTX470 at full blat and will run a GTX580 with 5% speed loss.

I'm not familiar with your card, does it have a fan or is it passively cooled?
Are you running the GPU fans (if any) on Auto or manually set to 100% ?
"Side on" or "side off" ? :-)

Can you post the make and model of your card so I can get a look at it ?

Some observations on P4 CUDA crunching.
On my regular and experimental systems, running a single task per GPU gives between 90 and 95% GPU usage. Even running through the PCI bus a GTS250 (470GFlops) ran at "normal" speed (approx 18 minutes for "standard" 0.4AR unit).

The Win7 video system is less efficient at SAH crunching than XP so going back to XP on that rig could get you up to an additional 5% GPU usage.

Memory is KING-- Recommended is a minimum of 750MB of RAM. I found you can get away with 500MB but you have to strip the OS absolutely bare and have SAH as the only running program. On "The Monster" there are only 26 services running with SAH shut down.

Run GPU tasks only. In this type of crunching the CPU is used only to feed the video card. Keeping the GPU fed and watered keeps the CPU quite busy enough, (I found one GPU task required a constant 25% CPU on a 3GHz P4). Trying to run CPU tasks as well just binds the whole thing up.

This could be the reason why yours and betreger's GPU tasks are taking so long. They are the equivalent of times I used to get with an 8600GT, and I presume the 430 should have more grunt than that.

BOINC Manager and explorer.exe (the desktop) can get very memory and processor hungry at times, between them they can use around 150MB of memory and >50% of processor time on a low end system. Shutting them down saves heaps of resources. (You can restart them at any time by using the "File->Run" tab in Task Manager.)

HTH
T.A.

Passively cooled and powered through the PCI bus. I should run XP. I have an unused copy. Currently I have 1Gb in the box. I have another 1Gb, but things didn't run as well for some weird reason when I went to 2Gb. It's the correct memory, clock speed etc. That's another project. I seem to be getting ~90% GPU usage now that I upped the priority of the Lunatics thread. I still use only ~6% of the CPU on that thread. I know Explorer chews a lot of CPU and memory as does Chrome, Firefox et.al. Surprised at Boinc client though. I usually leave it running. That's an easy enough change. Here's the URL for exactly where I got the card and my exact model. The site has front/left/right/back pictures of it.
http://www.circuitcity.com/applications/SearchTools/item-details.asp?EdpNo=1416091&CatId=3669

I got it because it's a Fermi card and I'm taking an online course in parallel programming using none other than Cuda and Fermi cards. I also have an ATI OpenCL card in my i7 laptop, so from the computer science end of things my bases are covered.

I pulled off the side cover and used your "external" fan approach for a few minutes (without the fags). GPU temps dropped to low 60s before I had to terminate because the cats got curious. I think I can clock this thing pretty high if I can get air to it.

I am running CPU tasks as well, but I'm getting good GPU usage now. I'm not sure whether I should stop those or not given my low CPU usage for cuda apps. Doesn't appear to be a bottleneck.

Thank you T.A. and thank "The Monster." You're an inspiration.

Doug[/url]
ID: 1229108 · Report as offensive
Profile Peter M. Ferrie
Volunteer tester

Send message
Joined: 28 Mar 03
Posts: 86
Credit: 9,967,062
RAC: 0
United States
Message 1229152 - Posted: 8 May 2012, 15:36:03 UTC

seing as your able to run 7 ..

assuming you had enough power to run all the cards what is the max amount of gpus the project would support per 1 computer


ID: 1229152 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1229157 - Posted: 8 May 2012, 15:41:58 UTC - in response to Message 1229152.  

seing as your able to run 7 ..

assuming you had enough power to run all the cards what is the max amount of gpus the project would support per 1 computer


IIRC the server will not believe you have more than 8. Which only effects the amount of work you can request.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1229157 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11354
Credit: 29,581,041
RAC: 66
United States
Message 1229167 - Posted: 8 May 2012, 19:16:43 UTC - in response to Message 1228988.  
Last modified: 8 May 2012, 19:19:02 UTC

The Monster's cooling system intrigued me. I have to do something to my P4/gt430 rig. When betreger starting making fun of of it I thought I'd better take a look. From what I've seen it should be capable of 3000-3500 RAC. The first thing GPU-Z showed me was that the GPU wasn't being pushed. I installed Process Lasso and set it to change the cuda apps to high priority. That pushed the GPU usage to 85-90%. Perfect. Then I started overclocking. Didn't take long before my GPU temps were topping 90c, so I backed down to the default again. The temp in my house hit maybe 25c and when I checked on it after work the GPU was running 90-95c. It's not going to last long like that. I've got a CPU fan, the power supply fan and a 80mm case fan in there. The 80mm seems pretty useless, but I guess you get what you pay for. I have space for a couple of good fans and I'm willing to drill. Any suggestions? I'm new to this. The house will never hit higher than 30c.

By the way betreger has a gt430 attached to a dual core and he's getting less RAC than me. Just fess up and ask for help!

Doug

Doug I did not intend to make fun of your GT430, I was merely making an observation. FYI I crunch Einstein on the gpu and am getting apx 8000 RAC there, Now I know E@H pays apx 30% more than seti so the expectation was conservative. BTW congrats on getting the card producing more.
ID: 1229167 · Report as offensive
doug
Volunteer tester

Send message
Joined: 10 Jul 09
Posts: 202
Credit: 10,828,067
RAC: 0
United States
Message 1229355 - Posted: 9 May 2012, 2:20:30 UTC - in response to Message 1229167.  


Doug I did not intend to make fun of your GT430, I was merely making an observation. FYI I crunch Einstein on the gpu and am getting apx 8000 RAC there, Now I know E@H pays apx 30% more than seti so the expectation was conservative. BTW congrats on getting the card producing more.


Actually it was good for me to look at it. I was sort of letting things settle out before I paid much attention to it. That's when I discovered I was barely using the GPU. There are others out there with equipment similar to mine that are doing far better RAC-wise. I want to know why.

Doug
ID: 1229355 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11354
Credit: 29,581,041
RAC: 66
United States
Message 1229373 - Posted: 9 May 2012, 3:13:34 UTC - in response to Message 1229355.  
Last modified: 9 May 2012, 3:15:01 UTC

Doug, I suspect, expect about 4k from that gpu, pls keep posted. It is very cool to keep old stuff productive. I would like to see your RAC after a couple of weeks of 24/7.
ID: 1229373 · Report as offensive
Profile Peter M. Ferrie
Volunteer tester

Send message
Joined: 28 Mar 03
Posts: 86
Credit: 9,967,062
RAC: 0
United States
Message 1229552 - Posted: 9 May 2012, 15:33:46 UTC - in response to Message 1229157.  

well what i was getting at was if you split the pcie lanes and were able to put for example 12 gtx 690's in a custom built enclosure with a few 1200 watt power supplys and a AMD's Magny-cours (12 core) Opteron just to feed the gpus

1. what kind of rac
2. would the project even give you enough work to fill it up
3. what problems does anyone think a system like this would have
ID: 1229552 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65690
Credit: 55,293,173
RAC: 49
United States
Message 1229618 - Posted: 9 May 2012, 17:46:26 UTC

Well the Monster might not have happened If I had not mentioned that I'd bought one of those pci-e x16 extenders over in another thread.

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 1229618 · Report as offensive
Profile red-ray
Avatar

Send message
Joined: 24 Jun 99
Posts: 308
Credit: 9,029,848
RAC: 0
United Kingdom
Message 1229621 - Posted: 9 May 2012, 17:53:52 UTC - in response to Message 1229552.  
Last modified: 9 May 2012, 18:01:36 UTC

10 x GTX 680 will give at least a 300K RAC with the next Lunatics_x41 release. More if you power mod and overclock the GTX 680s.

Most days I can get enough WUs for a 680+460 50K RAC system in 2 to 3 hours, so 10 x 680 in 24 hours should be possible.

The system will run out of PCI memory space which is limited to 4GB. It's possible to have 64-bit PCI memory space but I have never seen GPU BARs above 4GB.

You will need to have 1.0GB for RAM + 0.5 GB for PCIEXBAR and other BARs below 4GB and looking at my 980X with 4 GPUs it's got 2.5GB of memory so each GPU takes 0.25GB. 4.0 - 1.0 - 0.5 = 2.5 GB so 10 GPUs is the likely limit.
ID: 1229621 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13715
Credit: 208,696,464
RAC: 304
Australia
Message 1229623 - Posted: 9 May 2012, 18:00:17 UTC - in response to Message 1229621.  

10 x GTX 680 will give at least a 300K RAC with the next Lunatics_x41 release.

So 10 GTX690s would give over 600,000 for a single machine. You'd also be chewing through 2.5-3kW of power.

Grant
Darwin NT
ID: 1229623 · Report as offensive
Profile red-ray
Avatar

Send message
Joined: 24 Jun 99
Posts: 308
Credit: 9,029,848
RAC: 0
United Kingdom
Message 1229627 - Posted: 9 May 2012, 18:02:59 UTC - in response to Message 1229623.  

10 x GTX 680 will give at least a 300K RAC with the next Lunatics_x41 release.

So 10 GTX690s would give over 600,000 for a single machine. You'd also be chewing through 2.5-3kW of power.

No, 10 x 690 is 20 GPUs so the PCI memory limit applies.
ID: 1229627 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13715
Credit: 208,696,464
RAC: 304
Australia
Message 1229631 - Posted: 9 May 2012, 18:14:17 UTC - in response to Message 1229627.  


In that case i'd go with 5 GTX690s. One 690 uses 100W less than 2 680s.
Grant
Darwin NT
ID: 1229631 · Report as offensive
Previous · 1 · 2 · 3 · 4 · Next

Message boards : Number crunching : It Lives, It Lives !!


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.