Sign in with
Sign up | Sign in
Your question

GTX 680 vs GTX580 performance scaling

Last response: in Graphics & Displays
Share
March 22, 2012 6:47:41 PM

Hi,

Just read the 680 review here.

And I am confused. :pt1cable: 

The 680 has thrice the number of shaders as the 580 but its performance doesn't scale in the same way.

For eg:

In Crysis 2 and BF3, @ 2560x1600, I see a 33-50% FPS gain over 580.

Am I missing something here? :ouch: 

I think its either the 256bit memory interface or its the ROP units (count = 32, same as the 580), thats bottlenecking the card.

What do you guys think?
a c 642 U Graphics card
March 22, 2012 7:45:05 PM

I think you hit the nail on the head. Also, there is the issue of CPU bottlenecking, because basically you're asking why 3x SLI'ed GTX 580's don't run 3x as fast as a single card.
m
0
l
March 22, 2012 8:31:15 PM

It's all architecture.

GTX 580 uses shader "hot clocks" that is twice that of the core, but instead the GTX 680 uses twice the amount of hardware while removing the hot clocks.

There are other tweaks to the arch as well, but that is the most significant.
m
0
l
Related resources
March 22, 2012 8:42:12 PM

Yeah, the cores don't scale one to one. Nvidia is trying a different architecture strategy. It isn't the memory, the bandwidth on the 680's is roughly the same as the gtx 580 at about 192 gb/s.

http://www.xbitlabs.com/articles/graphics/display/nvidi...

The design is a lot more like amd/ati's where they have always used a lot more parallel processing cores to get the same amount of frames out than nvidias design.
m
0
l
a b U Graphics card
March 23, 2012 7:14:11 PM

I for one with hazerath
And like every review has covered,the 680 shows striking similarities with a GF114 rather than GF110.And its built to be twice as fast or eqquivalent to 768 cuda cores of fermi.
But with this new generation processing node nvidia figured that they can go 2000Mhz and above on the shader clock,BUT thats too expensive so they simply doubled the number of cores and the functional units.This allowed them to run everything on core clock(half shader speed).
Thereby handing over to everybody more room for overclocking,scaling,maintaining power and temps,etc.
m
0
l
!