CUDA and openCL Benchmarks

mmstick
mmstick
Joined: 6 Jun 12
Posts: 14
Credit: 2,066,411
RAC: 0

RE: RE: If you run with

Quote:
Quote:
If you run with Linux, AMD GPUs will perform faster, clock higher, and use lower voltages than in Windows in regards to OpenCL. Plus, it uses far less resources than Windows XP.

After doing some testing with my Radeon HD 7950 running Catalyst 13.11 @ 1200MHz and no voltage increase in Ubuntu and a GPU load of 93%, runtime per task is 800 seconds with three BRP4G-opencl-ati work units (as in, run three work units then divide their time by three for actual throughput rate). I'll soon have numbers for the longer BRP5 tasks

Okay, after averaging runtimes of 10 work units from BRP4 and dividing those results by three, because I run 3 work units on my 7950, I come to a result of 731 seconds per work unit.

After averaging runtimes of the last 5 BRP5 work units, the result is 2752 seconds per work unit.

tl;dr: Ubuntu is significantly faster than Windows with OpenCL for AMD using Catalyst 13.11. With a HD 7950 clocked to 1200Mhz, BRP4 work units complete within 731 seconds and BRP5 work units complete within 2752 seconds per work unit.

Jeroen
Jeroen
Joined: 25 Nov 05
Posts: 379
Credit: 738,423,020
RAC: 0

With a bit more tweaking, I

With a bit more tweaking, I was able to get my runtime down to 6338.10 seconds averaged running 3-tasks per 7970 GPU via the 13.10 driver. BRP5 tasks run slightly faster on one of my cards compared to the other. I think this is due to variance in PCI-E bandwidth between my two x16 3.0 slots. When I run AMD's BufferBandwidth tool, device->host bandwidth is reported as 12.36 GB/s on card #1 and 14.03 GB/s on card #2. Increasing base clock which in turn increases PCI-E frequency via Intel 2nd generation and newer CPUs also helps boost BRP performance.

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1,313
Credit: 427,164,907
RAC: 113,472

Well I just added this one

Well I just added this one since I have new parts laying around and decided to check out this new HD with Win8 since I never have even looked at it.

(and as usual changed several things so it was more like XP Pro and 7)

But as far as running the GPU tasks here this "CAL AMD Radeon HD 7500/7600/8500/8600 series (Devastator) (768MB) driver: 1.4.1848" is devastatingly SLOW......about 27 hours for a pair of Perseus Arm Survey tasks.

So I guesss I better figure out how to install a nVidia 550Ti I have extra here.

Found out getting in the bios is different too.

Maybe get in the Device Manager and see if it will recognize the nVidia card when I reboot.

I know it is at least 5X faster.

http://einsteinathome.org/host/9541222

 

Jeroen
Jeroen
Joined: 25 Nov 05
Posts: 379
Credit: 738,423,020
RAC: 0

I have been doing some

I have been doing some testing with the 780 Ti which has all 15 SMX units enabled via the GK110 GPU. With the card installed in a PCI-E 2.0 x16 slot and running three BRP5 tasks, the averaged run time for each task is 10,098 seconds. Hopefully NVIDIA will be releasing updated Linux drivers soon for the new card.

On the AMD front, the 13.11 Beta 1 and 6 Linux drivers run substantially slower than the 13.10 Beta 2 drivers when I tested with BRP5.

DF1DX
DF1DX
Joined: 14 Aug 10
Posts: 64
Credit: 1,418,197,359
RAC: 811,944

Anybody got a AMD R9 290 or

Anybody got a AMD R9 290 or 290X GPU?

My 7970 is broken :-(

Alex
Alex
Joined: 1 Mar 05
Posts: 449
Credit: 341,376,839
RAC: 92,118

from GPUGRID

John Jamulla
John Jamulla
Joined: 26 Feb 05
Posts: 29
Credit: 646,476,228
RAC: 79,171

Hi - I know this is an old

Hi - I know this is an old post, but have both a GTX 560 TI and a GTX 460 1 MB, I am running einstein@home with them.

I was wondering how you "benchmark" them, or see how long and individual task is taking, etc.
Are you digging into the .xml files?

I've been wondering how do I tell how well my cards are doing with different numbers of tasks running on them, since I hae 4 different types of cards, and probably getting some more.

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 804,698,613
RAC: 177,888

Either look at the finished

Either look at the finished tasks in Boinc before they are reported to the server or check on this webpage your account -> computers -> tasks for the computer your interested in. That's how I do it at least.

mikey
mikey
Joined: 22 Jan 05
Posts: 6,589
Credit: 604,613,764
RAC: 989,115

RE: Hi - I know this is an

Quote:

Hi - I know this is an old post, but have both a GTX 560 TI and a GTX 460 1 MB, I am running einstein@home with them.

I was wondering how you "benchmark" them, or see how long and individual task is taking, etc.
Are you digging into the .xml files?

I've been wondering how do I tell how well my cards are doing with different numbers of tasks running on them, since I hae 4 different types of cards, and probably getting some more.

Also get this:
http://downloads.guru3d.com/GPU-Z-0.6.4-download-2883.html

It is called gpu-z, and it's free, that lets you know lots of things about your gpu, I think the newest version 7.0.7.5, but the link should get you to a download after which you can then upgrade very easily. It will especially tell your current gpu load, which is very helpful in helping to decide if the gpu can handle a 2nd, 3rd or even 4th unit running at once on it.

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1,313
Credit: 427,164,907
RAC: 113,472

Time for a GeForce driver

Time for a GeForce driver update.

GeForce 334.89 Driver
Version: 334.89 - Release Date: Tue Feb 18, 2014

http://www.geforce.com/drivers/results/73214

New in GeForce 334.89 Certified drivers

Performance Boost – Increases performance by up to 19% for GeForce 400/500/600/700 series GPUs in several PC games vs. the previous NVIDIA Certified drivers (332.21). Results will vary depending on your GPU and system configuration.

 

Comment viewing options

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