facebook rss twitter

Review: CrossFire and SLI dongle-less / bridgeless performance analysis

by Steve Kerrison on 18 September 2006, 09:11

Tags: ATi Technologies (NYSE:AMD)

Quick Link: HEXUS.net/qagq3

Add to My Vault: x

A quick lesson...

Just in case you feel you need it, here's a quick recap on how both SLI and CrossFire work, along with why the bridges and dongles are necessary.

SLI

SLI, which stands for Scalable Link Interface, is a technology created by NVIDIA. It's the company that ended up with 3dfx's IP, who some of the old school gamers will remember for its SLI technology; also multi-GPU, but a different implementation.

SLI leverages two or more GPUs by distributing a workload over them in one of three ways. First we have Split-Frame Rendering (SFR). With SFR, the GPUs are each tasked with a part of the frame to render. With two GPUs, the split is, as you might expect, roughly half (top and bottom), although load-balancing means that the split can move up or down, depending what's happening in the scene.

Next we have Alternate-Frame Rendering (AFR). One GPU works on one frame while the other GPU works on the subsequent frame. Nice and simple to understand, that one.

Finally there's SLI Antialiasing. In this mode the GPUs share the load of the antialiasing work, with the intent of improving image quality rather than framerates.

CrossFire

ATI's answer to SLI is CrossFire. It does a few things differently, but with the same core goals.

Supertiling is the first CrossFire rendering method. A frame is divided up into little squares, like a checkerboard, where adjacent frames are rendered by alternate GPUs.

Next we have Scissor and AFR. For Scissor, think SFR, but for CrossFire (albeit without intelligent load-balancing). For AFR, think... AFR! Got it? Great.

Finally there's Super AA, where (and you'd never have guessed this) the GPUs share the load of the antialiasing work. Jeepers!

Bridges and dongles

OK, hopefully that recap was helpful to a few of you. So where do the dongles and bridges come in? The trouble with having more than one graphics card working on rendering something comes into play when it needs outputting to display. Only one card can be connected to the display, so the efforts of the other GPUs needs to reach that card, somehow.

For NVIDIA, the answer is an internal bridge link between the cards, through which the fruits of one card's labour can be sent to the other for output in all its glory on whatever cinema-like display is hanging off the head of the primary card.

ATI's chosen method has been to have the second card output its efforts over DVI to the primary card, which would then combine that output with its own, then send it out to the display. This creates the necessity for a CrossFire master card with the special input/output connector and a compositing chip.

Ditch the dongles or bridges and the PCI-Express bus can be used instead, but that's using up bandwidth on an interface that's already rather busy sending data to the cards. So, in this review, it's the impact of this that we'll be analysing. On to the system setup!