Bigger and better GPU sieving drive: Discussion - Page 5

Flatlander · 2010-10-08, 09:38

Quote:

Originally Posted by vaughan

...The factors file is growing nicely, its at 2331kb so far and we're up to 120233 (range is 120000 - 130000) so that's nearly 25 percent completed in a 2.5 hours.

2.3%?

vaughan · 2010-10-08, 10:04

Oops yes you're right Flatlander 2.3%

Gary go to nvidia's site and get Ntune. You can unlock the GPU fan from "Automatic" to "Direct control" and then set it to 100%

vaughan · 2010-10-08, 13:21

I managed to get the speed to display by going to the physical computer and plugging in a real monitor. Previously I was remotely logged in to the box using LogMeIn.

Next question, in a Windows 7 64-bit PC with two 9800GT GPUs how do you tell the cuda sieving application to use both GPUs not just GPU0?

frmky · 2010-10-08, 16:01

Add -t2 to the command line.

Ken_g6 · 2010-10-08, 16:17

Eep! No, don't do that! It used to work, but there are a lot of things that should be separate between threads, but which I haven't kept separate in a long time. With some work it could work again, but not now.

Edit: The proper way to do it is to run two copies of the app, one with --device 0 and the other with --device 1.

mdettweiler · 2010-10-08, 16:37

Quote:

Originally Posted by gd_barnes

Max, is there a way to cool my GPU? The CPU temps are running kinda high at 75C. I think I'll put an external fan on it.

Hmm, interesting...I hadn't considered that the GPU might be putting out enough heat to start to interfere with the CPU's cooling.

I'll give vaughan's suggestion a try as soon as I get the chance.

Meanwhile, though, in case you didn't get my email: I can't reach the GPU machine right now. I get "no route to host" indicating that it is offline or horribly frozen. In short, this means that the GPU is not crunching until I can get back on.

frmky · 2010-10-08, 16:44

Quote:

Originally Posted by Ken_g6

Eep! No, don't do that! It used to work, but there are a lot of things that should be separate between threads, but which I haven't kept separate in a long time. With some work it could work again, but not now.

Oh no...

The README.TXT file says:
-t4 Only multi-GPU systems. -tN will spread the work of the sieve over
N cores. Although it may be a little
LESS productive, it's easier to manage.
and I've been using it for days on a system with 4 GPUs. Are these results now suspect?

Edit: Once the current range finishes in a few hours, I'll try the test range using -t4.

Ken_g6 · 2010-10-08, 16:53

Hm. If they're all the same, like I think SLI requires, you might be OK. I'll look.

Edit: Yeah, if they're identical in compute capability and CUDA core count, you should be OK. I'll have to look at restoring the capability to work on multiple different GPUs later.

frmky · 2010-10-08, 20:08

The GPUs aren't SLI, but they are all identical C1060's. I reran 3100G-3101G with and without -t4 and they found the same 2235 factors. Likewise, 10000G-10001G with and without -t4 found the same factors, and all the tests in the first post of the ppsieve CUDA testing thread found the same factors. They are in a different order in the output file, but they are all there.

gd_barnes · 2010-10-09, 07:51

The mobo on my GPU machine has likely gone bad. It will be Monday before I can replace it. The ETC on my range after that will be ~3 days. If it looks like I'll be holding up completion of 10T-20T, I'll release it before Monday.

Thanks for the excellent contributions guys!

vaughan · 2010-10-09, 09:48

Did you download and install and run ntune? You can sometimes (see below) get it to control you GPU fan speeds so you can lower the temps when crunching 100%

Alternatively you can download and install and run Rivatuner. It triggers unsigned driver messages from Win 7 but will also control the fans and therefore the temps.

Google is your friend to find these apps.

On several of my machines, all Win 7 or XP 64-bit so far, ntune triggers a BSOD and forces a reboot. Win 7 says the system recovered from a serious error (yeah lousy programming by nvidia). On my 32 bit Windows machines ntune works as it should.

Both Rivatuner and ntune bring the GPU temps down at least 10C so are very useful utilities.

2010-10-08, 16:17	#49
Ken_g6 Jan 2005 Caught in a sieve 18B₁₆ Posts	Eep! No, don't do that! It used to work, but there are a lot of things that should be separate between threads, but which I haven't kept separate in a long time. With some work it could work again, but not now. Edit: The proper way to do it is to run two copies of the app, one with --device 0 and the other with --device 1. Last fiddled with by Ken_g6 on 2010-10-08 at 16:19

2010-10-08, 16:53	#52
Ken_g6 Jan 2005 Caught in a sieve 5·79 Posts	Hm. If they're all the same, like I think SLI requires, you might be OK. I'll look. Edit: Yeah, if they're identical in compute capability and CUDA core count, you should be OK. I'll have to look at restoring the capability to work on multiple different GPUs later. Last fiddled with by Ken_g6 on 2010-10-08 at 17:01

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
New PRPnet drive discussion	mdettweiler	Conjectures 'R Us	89	2011-08-10 09:01
Sieving drive Riesel base 6 n=1M-2M	gd_barnes	Conjectures 'R Us	40	2011-01-22 08:10
Bigger and better GPU sieving drive: k<10000 n<2M	mdettweiler	No Prime Left Behind	61	2010-10-29 18:48
GPU sieving drive for k<=1001 n=1M-2M	mdettweiler	No Prime Left Behind	11	2010-10-04 22:45
Sieving drive for k=301-400 n=1M-2M	MyDogBuster	No Prime Left Behind	42	2010-03-21 01:14

2010-10-08, 10:04	#46
vaughan Jan 2005 Sydney, Australia 517₈ Posts	Oops yes you're right Flatlander 2.3% Gary go to nvidia's site and get Ntune. You can unlock the GPU fan from "Automatic" to "Direct control" and then set it to 100%

2010-10-08, 13:21	#47
vaughan Jan 2005 Sydney, Australia 101001111₂ Posts	I managed to get the speed to display by going to the physical computer and plugging in a real monitor. Previously I was remotely logged in to the box using LogMeIn. Next question, in a Windows 7 64-bit PC with two 9800GT GPUs how do you tell the cuda sieving application to use both GPUs not just GPU0?

2010-10-08, 16:01	#48
frmky Jul 2003 So Cal 2×3⁴×13 Posts	Add -t2 to the command line.

2010-10-08, 20:08	#53
frmky Jul 2003 So Cal 2×3⁴×13 Posts	The GPUs aren't SLI, but they are all identical C1060's. I reran 3100G-3101G with and without -t4 and they found the same 2235 factors. Likewise, 10000G-10001G with and without -t4 found the same factors, and all the tests in the first post of the ppsieve CUDA testing thread found the same factors. They are in a different order in the output file, but they are all there.

2010-10-09, 07:51	#54
gd_barnes May 2007 Kansas; USA 3³×5×7×11 Posts	The mobo on my GPU machine has likely gone bad. It will be Monday before I can replace it. The ETC on my range after that will be ~3 days. If it looks like I'll be holding up completion of 10T-20T, I'll release it before Monday. Thanks for the excellent contributions guys!

2010-10-09, 09:48	#55
vaughan Jan 2005 Sydney, Australia 5×67 Posts	Did you download and install and run ntune? You can sometimes (see below) get it to control you GPU fan speeds so you can lower the temps when crunching 100% Alternatively you can download and install and run Rivatuner. It triggers unsigned driver messages from Win 7 but will also control the fans and therefore the temps. Google is your friend to find these apps. On several of my machines, all Win 7 or XP 64-bit so far, ntune triggers a BSOD and forces a reboot. Win 7 says the system recovered from a serious error (yeah lousy programming by nvidia). On my 32 bit Windows machines ntune works as it should. Both Rivatuner and ntune bring the GPU temps down at least 10C so are very useful utilities.