Comparing GPU performance

Hype
Hype
Joined: 7 Oct 14
Posts: 5
Credit: 7,246,082
RAC: 52,634
Topic 220045

Hey guys,

I'm playing around with three different GPUs right now, 1050 Ti, 1080 and 1080 Ti.
I did a comparison of Einstein and Milkyway and there are quite some differences which I didn't expect.

5Gfei77.jpg

The 1080 for example is only 90% quicker than the 1050 Ti on Einstein, but it's 190% quicker on Milkyway.
That results in the 1050 Ti being the most efficient card on Einstein for me related to power consumed, it outputs the most points per kWh.
On Milkyway the differences are not that noticeable.
The watts are not 100% correct, as they are higher for Milkyway as the GPU usage is around 80%, whereas Einstein sits at around 65%.

Why is it that the 1080 and 1080 Ti are not considerably faster on Einstein than the 1050 Ti for example?

Cheers

mmonnin
mmonnin
Joined: 29 May 16
Posts: 265
Credit: 875,884,936
RAC: 1,508,268

The math functions are

The math functions are different. Don't expect performance to match. MW needs FP64 so GPUs with high FP64 compute perform great there. I never run any of my Pascal GPUs on MW for that reason.

MW requires multiple tasks to run in parallel to keep GPU util high since they run very quick. 

Some of the GPUs you've mention have different GDDR versions which can change compute.

mikey
mikey
Joined: 22 Jan 05
Posts: 5,314
Credit: 526,056,153
RAC: 43,079

Some of your cards are listed

Some of your cards are listed here with their FP64 rates:

http://www.geeks3d.com/20140305/amd-radeon-and-nvidia-geforce-fp32-fp64-gflops-table-computing/

 

Gaming does use FP64, dual precision, so it's been going down in the newer cards.

floyd
floyd
Joined: 12 Sep 11
Posts: 130
Credit: 185,674,651
RAC: 6,920

RoundFour wrote:Why is it

RoundFour wrote:
Why is it that the 1080 and 1080 Ti are not considerably faster on Einstein than the 1050 Ti for example?

I don't know the reason but I see the same for a GTX 970 and a GTX 750 Ti. Despite running near full load, the 970 only generates about 60% more output on the GW search. It should be capable of 150% and got into that range on the FGRP search last time I tried.

Hype
Hype
Joined: 7 Oct 14
Posts: 5
Credit: 7,246,082
RAC: 52,634

mikey wrote:Some of your

mikey wrote:

Some of your cards are listed here with their FP64 rates:

http://www.geeks3d.com/20140305/amd-radeon-and-nvidia-geforce-fp32-fp64-gflops-table-computing/

 

Gaming does use FP64, dual precision, so it's been going down in the newer cards.

Ah ok, seems to be Radeons are a lot quicker on MW then.

mikey
mikey
Joined: 22 Jan 05
Posts: 5,314
Credit: 526,056,153
RAC: 43,079

RoundFour wrote:mikey

RoundFour wrote:
mikey wrote:

Some of your cards are listed here with their FP64 rates:

http://www.geeks3d.com/20140305/amd-radeon-and-nvidia-geforce-fp32-fp64-gflops-table-computing/

 

Gaming does use FP64, dual precision, so it's been going down in the newer cards.

Ah ok, seems to be Radeons are a lot quicker on MW then.

Yes they are, some of mine can do a unit over there in 47 seconds!!

Hype
Hype
Joined: 7 Oct 14
Posts: 5
Credit: 7,246,082
RAC: 52,634

But they also use more power,

But they also use more power, right?

Comment viewing options

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