It looks like the 5870 is basically two Radeon 4890s on the same die, sharing a memory controller. It is a much smarter design than your typical dual GPU solution since it behaves as a single GPU. Let's hope Nvidia releases something competitive so prices are low!

The beast they call the $5870 is only going to be $400 at launch, too.

Quite frankly, I'm scared of what the 5870X2 will be capable of. It's got more processing power than like 3.5 Play Station 3's, on one add-in PCI-Express card.

Oh Panda, you just had to mention the PS3. Prepare yourself.

It's a good metric for console gamers to be able to grasp on to.

If you're comparing GPU performance alone, 8800GTX was already 3x+ faster than the G71 based "RSX" and the R500 "Xenos" back in late 2006.

No, total PS3 system performance.

This GPU is 13x faster than the PS3's GPU.

They aren't really comparable when you bring the CPU into the equation though. That's like saying that the 4870 is faster than the Core i7 975, despite the fact that they perform different functions.

I'm using the total system output as my metric. FLOPS. FLoating-point operations per second. The PS3 has a peak output of about 1.9 to 2 teraFLOPS. And the 5870X2 should be able to peak at 6.5 teraFLOPS.

That PS3 figure is so grossly exaggerated it's not even funny, and the ATI one is exaggerated as well. ATI is still using their super scalar shader architecture so hitting their peak number of instructions per clock cycle per shader is very difficult, which is what they used for their multi-teraflop performance estimate.

When talking about OpenCL or CUDA like development, it is even less likely that performance can really be utilized. Folding@Home would still be insanely fast on the 5870, despite the fact that it wouldn't push the GPU to it's limits.

I agree that comparing GPU flops to CPU flops is, at present at least, pretty useless - different strengths, different application domains. It would be tortuous to get the original Unreal or Doom to run on nothing but this new GPU (though you would still need the CPU to interface with the OS and system hardware), yet 100 Mhz CPUs with about 1/50 - 1/100 the horsepower of modern CPUs were able to - without any GPU.

Suppose I design a computer (call it "Deep Thought") to compute the Meaning of Life, the Universe, and Everything - and set some FLOPS records... I could create an integrated circuit that consisted of 50,000 floating point adder circuits (at about 20k transistors each, and none needed for memory, this gives us 1 billion transistors, less than modern processors) that computed the same operation (using the same input, "20" and "22") in parallel. It should run anywhere from 5 Ghz to 10 Ghz (due to their simplicity). So we're looking at a minimum of 250-500 teraFLOPS of raw processing power! But will it run "Crysis"? Of course not, it only computes "42".

BTW, I am not an IC architect. I got the 20k and 5-10 Ghz numbers for my "design" from http://www.giacomotto.com/d... (scaling up the frequency due to improvements in process technology since 2007).

@ above...

lol

I wonder how many people will get it?

8800 Ultra was $750 CAD, it was a horrible time. We should be thankful that AMD/ATI forced Nvidia to drop their prices, otherwise we'd be paying for GFX Cards the price of complete Mid-range towers.