5870 Crossfire Bench.
For those of you who haven't seen this my jaw almost dropped at some of the numbers. Mostly at the 4890 quad fire set up, That scaling is a real bitch, can't believe how badly it sucks.
Remember the the 5870 cards just came out. The drivers for 2x 5870 Xfire are very immature...and the drivers that deal with Xfire are even more immature.
Also the 4890x4 is NOT a quad fire setup - it's just the name of a 4890 with a cooler. The cooler's name is "icooler x4" - it's just a single GPU: 4890x1.
Expect a substantial performance boost by the end of this year after mature drivers are released for the cards.
Folding only sucks because they didn't update the core from the HD38xx to the HD48xx series, so that the ATi cards have to do the force calculations twice. They were the only ones able to do the complex models/WU but they finally added them to the nV card which actually had worse results as a result of now doing similar tasks. But still that computation issue in the old client remains so the HD4K are nowhere near their potential.
Remember it's because ATi was first that they are unfortunately still being held back by code made to work on the HD2900, whereas nV has no such issue because they came to the game long afterwards and have the latest core build which can't be fixed by client optimizations. While there was talk of making another GPU core, the OpenCL one will be the way to go. Also remember early HD5K results aren't even using an HD5k client, so those are far less than potential too, should see about 25+% increase on top of that with just an optimization.
There's supposed to be a new OpenCL client in a few months, it should open up the performance on the HD4K & HD5K, but won't be able ot do much for the HD3K and earlier;
BTW, this gets done in stages because it's Stanford that adds this stuff to the client and they just put a few people on building the new client integrations. First it was available only on ATi cards who actually pioneered the effort to put F@H on GPUs, then the nV cards were given an optimized client, then a ATi tweak, then an nV tweak. Right now the advantage is in the nV optimizations in the client because they have no legacy hardware, whereas ATi has the HD2900 and even X1900 legacy stuff to support, but they are far from sucking in Folding @ home, they're just not good at the small WUs that people love for getting points.
Here's some insight for you guys;