CUDA and openCL Benchmarks

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1907
Credit: 1438606145
RAC: 1222319

RE: Those long times are

Quote:

Those long times are from my GT640 not the GTX660, But that is running two
at a time.

The GT640 runtime for BRP4 was 8,543.83 two at a time so approx 11 - 1

BRP5 vs BRP4

Well that link says it is your GeForce GTX 660 tasks when I click on it Tom

(but in the Stderr it has "GeForce GT 640" )

Tom*
Tom*
Joined: 9 Oct 11
Posts: 54
Credit: 366729484
RAC: 0

Yeah, you have to look at the

Yeah, you have to look at the stderr to see which GPU is crunching.

Or just look at the runtimes:-)

I am calibrating SETI V7 on that system now and had to pull the GT640
as they send various app level cuda tasks (cuda 23, cuda 32, cuda 42, cuda 50)to see which one runs the best on any particular GPU, Well with that slow GT640 in the system it would never calibrate.

So it might stay out for awhile:-)

Once its calibrated I'll bring it back on Einstein

Alex
Alex
Joined: 1 Mar 05
Posts: 451
Credit: 508910244
RAC: 77687

Hi, new Intel cpu's are

Hi,

new Intel cpu's are there, AMD has updated their A8/A10 APU's, nVidia GTX 7xx is available.
Most discussions about new hardware take a closer look in gaming performance and deeper sleep states (which is, AFAIK, not the most important ability here), but no one talks about open-CL performance and crunchig speeds.

This is why I ask here: are there first results available which reflect the performance for what we do?

astrocrab
astrocrab
Joined: 28 Jan 08
Posts: 208
Credit: 429202534
RAC: 0

here are my results for BRP5

here are my results for BRP5 on ATI 7970 @ 1100MHz:
1WU: 75% gpu load, 5600 sec runtime (needs to be confirmed)
2WU: 89-91% gpu, ~7800 sec runtime. 3400 sec per WU
4WU: 97% gpu, ~13200 sec runtime. 3300 sec pe WU

astrocrab
astrocrab
Joined: 28 Jan 08
Posts: 208
Credit: 429202534
RAC: 0

5WU: 96% gpu, ~16330 sec

5WU: 96% gpu, ~16330 sec runtime. 3266 sec per WU

Jeroen
Jeroen
Joined: 25 Nov 05
Posts: 379
Credit: 740030628
RAC: 0

Now that Seti has both AP and

Now that Seti has both AP and v7 MB apps available for Linux AMD, I have been experimenting with running these tasks alongside BRP5 on the same GPU via a single system.

When I had two AP tasks and one BRP5 task running on a single GPU, the AP tasks had approximately the same run time as if I ran only a single AP task on the GPU and the BRP5 task is running approximately the same run time as if the GPU were only running BRP5 tasks.

I also had a case with a BRP5, AP, and v7 task running on a single GPU and each task ran okay. The v7 tasks have quite a bit of variation in run time from one task to another. I need to do some more testing to see how 1 task runs compared to 2 or 3 tasks per GPU and on the same GPU as the other applications.

GPU load is higher when running AP tasks alongside BRP5. Normally when I run only BRP5, the GPU load is 95% whereas with the AP tasks running, GPU load is 98-99%. Einstein is still my main project but I like to run some AP tasks here and there and I am glad the applications have been ported to Linux and run well alongside the BRP5 application.

Jeroen
Jeroen
Joined: 25 Nov 05
Posts: 379
Credit: 740030628
RAC: 0

RE: here are my results for

Quote:
here are my results for BRP5 on ATI 7970 @ 1100MHz:
1WU: 75% gpu load, 5600 sec runtime (needs to be confirmed)
2WU: 89-91% gpu, ~7800 sec runtime. 3400 sec per WU
4WU: 97% gpu, ~13200 sec runtime. 3300 sec pe WU

Great results!

Which driver are you running currently? I recently upgraded to 13.4 from 12.11 Beta 11.

astrocrab
astrocrab
Joined: 28 Jan 08
Posts: 208
Credit: 429202534
RAC: 0

13.4 too

13.4 too

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1907
Credit: 1438606145
RAC: 1222319

Well I just did a driver

Well I just did a driver update on a 650Ti GeForce 320.18 WHQL

At the same time I added another GPU cruncher to my grid using a quad-core that was just running Gamma Rays since Jan 2012

I never added a cuda card before because this is the only one I bought from a store and it is in that tiny box with no room for the GeForce card.

But I had a 650Ti under warranty in another box and I sent it back to this certain company using the "RMA" so they sent me a new card first and then I sent them the other card so they could replace the $178 they charged me for the replacement (credit card) since it was under the 3 year warranty.

Well instead when they received it they made up phony excuses and just sent it back to me so as I told them via email I could have repaired it myself but the warranty says they have to do that or replace it.

Well I got it back today and in about 5 minutes I repaired it myself and cut that tiny box open (had to remove the side panel and cut the top off so the 650Ti would fit) and downloaded the new driver and now I have 6 GPU crunchers running!

If I could go back in time I would have just had that certain company send me the over-priced 660Ti since it (and the 550Ti) are better and faster than the 650Ti (SSC 2GB) instead of just having to pay for another 650Ti.

I guess the only good thing is the 650Ti is supposed to run more "fuel efficient" and maybe my next light bill will not be too insane.

So right now this box is running 3 GRP's and one Test4Theory and a pair of BRP5's

http://einsteinathome.org/host/4519028

Betreger
Betreger
Joined: 25 Feb 05
Posts: 992
Credit: 1606905451
RAC: 673674

RE: maybe my next light

Quote:
maybe my next light bill will not be too insane


Where we live electricity costs are moderate compared to the rest of the world.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.