Asymmetric Crossfire 5770 x 5750

X

Xzyrus

Limp Gawd
Joined
Nov 5, 2009
Messages
290
http://www.xbitlabs.com/articles/video/display/asus_eah5770-powercolor_pcs_hd5750_7.html#sect0

I've always read that in crossfire the faster card downclocks to match the slower one and people usually say something along the lines of might as well just get 2 of the slower one instead of doing it asymetrically but then I found this today. It doesn't address the downclocking and a lot of the game results show the asymmetric crossfire of 5770x5750 is faster than the crossfire of 5750s.

Is this because of the extra shaders/TMUs on the 5770 which makes the assymetric crossfire more beneficial than just 2 5750s crossfired?

I'm not interested in doing crossfire since my mobo would only do x16 x4 and my PSU is only 500W but I just wanted to make sure I had the correct understanding.
 
It's perfectly possible to SLI/Crossfire two different boards together and get a performance increase.

There are two methods:

1. AFR - One card renders a frame, and the other card renders the next. Each frame can be rendered independently of the other card.

This works and can yield %100 scaling even with mismatched cards, but if the cards are not fairly close together in performance, the framerate produced will feel juddery (one card renders a frame faster than the other, giving you an uneven framerate). That is why in the past, they insisted that you clock the cards the same (to give you as smooth an experience as-possible). But for the 5x series, they've decided a little judder is okay, so they typically let you pair cards inside of a class (e.g. you can pair 5750 with 5770, 5850 with 5870).

2. Tiling or split-screen - draws a portion of the screen on each card. Geometry is precalculated for the scene by the master card.

This works, but scales well only if the game is shader-limited. The geometry can only be calculated by one card. This is not the default mode for 2-card Crossfire.

AFR was likely used in the review. I'm not sure how bad the judder would be with two cards that close in performance, I'd like to see some analysis. Unfortunately, the reviewers just ran a bunch of benchmarks. They could have sat down and written about their experience, and their take on the uneven pairing, but they didn't.
 
Last edited:
Thanks for the informative post, defaultuser. They do a poor job of documenting their methodology (or I'm a really sloppy reader). So I just have to assume they left the asymmetric crossfire at different speeds I guess.

Hypothetically speaking if they did downclock the 5770 to 5750 speeds, would 5770x5750 crossfire be faster than 5750s crossfired? i.e. extra shaders/TMUs make a difference? Or would it be too small an amount that it could be attributable to margin of error?
 
Xzyrus said:
Thanks for the informative post, defaultuser. They do a poor job of documenting their methodology (or I'm a really sloppy reader). So I just have to assume they left the asymmetric crossfire at different speeds I guess.

Hypothetically speaking if they did downclock the 5770 to 5750 speeds, would 5770x5750 crossfire be faster than 5750s crossfired? i.e. extra shaders/TMUs make a difference? Or would it be too small an amount that it could be attributable to margin of error?
Click to expand...

5770:

800 shaders
40 texture
16 ROPS
850 MHz core

5750:

720 shaders
36 texture
16 ROPS
700 MHz core

The texture/shader ratio is the same for both cores. You could clock them so they have approximately the same shader/texture rate

800 SP * x MHz = 720 SP * 700 MHz

x = 630 MHz = clock rate of the matched 5770.

And since you probably aren't ROP-limited, 70 MHz isn't going to make a lot of difference. They would perform nearly the same.
 
You must log in or register to reply here.
Back
Top