Sign in with
Sign up | Sign in
Your question

Crossfire, how does it work?

Last response: in Graphics & Displays
Share
July 5, 2008 10:47:29 PM

As I see that nVidia is now planning to go to DX 10.1 also, there is no question, I'll be getting ATI. as I'm going with ATI, I thought I might learn a bit about crossfire to take better advantage of it in the future with my next purchase.

my understanding is that crossfire allows 2 or more dissimilar cards to work together, but they must have the same amount of memory (or the lower amount will be what all use).

I imagine it works like SLI as in each card does a certain portion of the screen. with crossfire, it would seam that the slower card would render a smaller portion of the screen. my question is about different memory speeds, such as combining a 4870 and 4850. will the lower memory speed hold back the faster memory?

More about : crossfire work

July 5, 2008 10:53:42 PM

Yes, I think it may. They have made some improvements to Xfire recently that allowed things like the 780g+discrete Xfire, so that might have changed. I hear you will be able to mix 38xx with 48xx too...
a b } Memory
July 5, 2008 11:32:26 PM

SLI and crossfire work in similar ways, but they do not just do one thing. there are different ways to use multi cards together.

You can have each card render half(or close to) of the screen. its not always half as the sky takes less power so it depends on where the drivers decide to draw the line

have each card render opposite frames.

card 1 : 1-3-5-7-9-11-13 ect
card 2 : 2-4-6-8-10-12-14 ect

And there is away to break the screen into tiles.

I am not sure who does what actively, but i know where you need to look :) 

http://www.tomshardware.com/forum/245454-33-crossfire-f...
Related resources
July 6, 2008 6:58:57 AM

I believe xfire does the checkerboard, but I believe that is only some, not all cards.

a b } Memory
July 6, 2008 9:03:02 PM


I beleive the newer cards by ATI are set to do alternate rendering, not 100% sure though. This may help you get your head around the basics.
http://www.tomshardware.co.uk/foru [...] sfire-faqs

What about brands ?
Well the brand doesnt matter , again for example , you can use a XFX card with a EVGA card , or a SAPPHIRE card with a DIAMOND card , just make sure they have the same memory and same clocks.

What will happend if 2 cards have the same amount of memory but one has higher clocks than the other?
One of the best ways to learn is to make examples.
You have a 8800GT(Normal edition,no OverClocks) and you want to add another 8800GT(Which is OverClocked),then:

1_The OverClocked one will lower its clocks to reach the Non-OverClocked card.

2_You can OverClock your card(The one that has Stock frequencies) and reach the OverClocked card.
Remember, you cant ALWAYS OverClock the normal card to reach the other card,because it may not OverClock that much and may become unstable.

WARNING:There are many exceptions in Computer Hardware, for example "Mousemonkey" a member of this forum has told me that when he puts his ASUS 7900GT card (Which is at stock speeds) with a XFX 7900GT(A OverClocked card) the ASUS one raise its clock to reach the XFX one which is an exception and as i said exceptions are alot in Computer Hardware.

Mactronix
July 6, 2008 11:04:31 PM

In crossfire the clocks do not change in the faster card, they remain higher than the slower card. In SLI maybe they do...I don't know. I thought that the faster one downclocks too but if you look in xfire reviews, the faster one will indeed keep it's clocks. This has been confirmed by people also...no downclocking.


Fudzilla quote "However, this didn't give us much in terms of performance, but we're still pleased to see that you can mix these R670 based cards and set the clocks independently."



What I was told is that the faster card will complete it's workload quicker than the slower card but they will still render the frame at the same time. From that I gather that xfire doesn't work by alternate frames so in the end, you are limited by the slower card's performance. I have read a lot of different speculation on how xfire works but nothing solid.

Coming to the dual 4850 vs 4870/4850 xfire setup, I don't know of a site that has benchmarked this yet. Going off the 3800 series, Fudzilla did a benchmark http://www.fudzilla.com/index.php?option=com_content&task=view&id=5641&Itemid=1 where they xfired the 3870 and 3850 and to my knowledge was the ONLY site that compared that setup to a dual 3850 xfire setup. That is an important benchmark that NO ONE does and it sucks because that is the big question....is there any reason to do a xx70/xx50 over a xx50/xx50 xfire setup? They pretty much Fd it up though because their two 3850s had different amounts of memory (256 and 512). It just throws in a monkey wrench into the whole thing when trying to determine the dual 3850s performance. They should have used two 512MB 3850s so that there was no memory bottle neck in the 3850 xfire results.

They did however do a 3870/3850 setup with both 3850s. So if you look at the 3870/3850 256MB test (the bright yellow) you can see that there are pretty much no gains at all and in most tests it comes out a hair slower than the dual 3850 setup.

So until someone actually makes a worthwhile xfire benchmark, I don't see how it would be of benefit to have a 4870/4850 xfire setup. I would stick to identical cards. Someone needs to do that if they have two spare 4850s and a 4870 laying around....come on, who doesn't? :heink: 
!