CUDA and openCL Benchmarks

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 0

I'm not seeing any speedup

I'm not seeing any speedup from 7.0.64

A sample of 20 BRP4 tasks from after averaged 1475 seconds. Two similar samples from before came in at 1479 and 1446 seconds/task. Standard deviations for the sets ranged from 123 to 157 seconds; so I'm almost certain the variation I've seen is just noise.

mountkidd
mountkidd
Joined: 14 Jun 12
Posts: 176
Credit: 12560542555
RAC: 8021528

Hi Dan, Thanks for your

Hi Dan,

Thanks for your comments. I presume your before was 7.0.28 as well.

I had a look thru a few days of your data and I see a huge range in the numbers - 1221-2995. And this is with overlooking several in the 4000-6000 range which are atypical. So an average of 20 might not be providing a true picture. The stdevs on my 50 sample data sets are averaging about 3% of the average value.

My numbers are holding (subject to hw malfunctions) at the lower values and it's now been about 12 days.

Gord

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 0

RE: Hi Dan, Thanks for

Quote:

Hi Dan,

Thanks for your comments. I presume your before was 7.0.28 as well.

I had a look thru a few days of your data and I see a huge range in the numbers - 1221-2995. And this is with overlooking several in the 4000-6000 range which are atypical. So an average of 20 might not be providing a true picture. The stdevs on my 50 sample data sets are averaging about 3% of the average value.

My numbers are holding (subject to hw malfunctions) at the lower values and it's now been about 12 days.

Gord

Yes to previously 7.0.28; my really high GPU numbers all involve boinc deciding to unsleep the GPU while I'm gaming and managing not to crash the task before I notice the framerate hit (the other half of the time I discover it crashed out a few dozen BRP4s before giving up and grabbing a less demanding primegrid/collatz task)

Eric_Kaiser
Eric_Kaiser
Joined: 7 Oct 08
Posts: 16
Credit: 25699305
RAC: 0

Hi all, I've looked on the

Hi all,

I've looked on the performance of my Radeon HD 7850 with 2 GB VRAM and a GPU clock set to 1050MHZ (overclocked in ccc).
Running solely 1 WU the GPU is around 45% used. With 2 WU in paralell GPU usage is around 90%.
Measured with 80 WU I've finished so far the average GPU time in seconds is 1330.
The values vary from 700s to 2000s but mostly around 1300s.
Temperature of GPU is ~50°C.

Card is running in combination with an and 64GB RAM

Edit: I'm using BOINC 7.0.64

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

I ordered a pair of Intel 64

I ordered a pair of Intel 64 GB SLC SSDs for my Linux AMD rigs. One system already had an Intel 520 but I wanted to move the BOINC data and log files over to the new SSD to save the flash life of the 520. The other system was on a spindle previously and had a 120mm fan to keep the disk cool. Since switching this system to SSD, I have seen a reduction in runtime of approximately 20 seconds per task per GPU. The system power consumption is down by approximately 15w from removing the hard disk and fan. These changes are not very significant but every bit helps.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 0

First BRP5 WU: 217m on a 560

First BRP5 WU: 217m on a 560 Ti, win7-64

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1886
Credit: 1406027941
RAC: 1164672

Anyone try the new nVidia

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

I am running 320.18 on my

I am running 320.18 on my i5-3570 with GT640 abd GTX660 it was a ho-hum

upgrade nothing changed , loading runtimes or power consumption, on Einstein.

of course the BRP5's were anything but ho-hum.

Upgrade was from 310.70 or 310.90 I have both directories in my NVIDIA/displaydrivers

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1886
Credit: 1406027941
RAC: 1164672

RE: I am running 320.18 on

Quote:

I am running 320.18 on my i5-3570 with GT640 abd GTX660 it was a ho-hum

upgrade nothing changed , loading runtimes or power consumption, on Einstein.

of course the BRP5's were anything but ho-hum.

Upgrade was from 310.70 or 310.90 I have both directories in my NVIDIA/displaydrivers

YOW I just looked at your 660 running the BRP5's and saw a couple of these.

http://einsteinathome.org/task/381597639

Yes that is a bit longer than the BRP4's on a GeForce 660

I loaded several on my 660Ti SC and my other hosts but they are still running the BRP4's they have so I haven't actually ran a BRP5 yet......now I wish I didn't load so many before trying just a couple on each host.

Thanks for the info Tom

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

Those long times are from my

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

Comment viewing options

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