The HD4870 single chip will sell for $299. That means the HD4870X2 should be less than $599.

With effing 800 stream processors, 1GB of 3.9Ghz GDDR5 RAM and a core clock of 725Mhz, I would be surprised if it weren't more powerful than four PS3's duct taped together... at the very least...

Yeah, utility per dollar is pretty much what I was going for. I've long since stopped dropping my jaw at technology which is amazing, yet impractical (read: expensive).

I do appreciate that AMD is pursuing values beyond mere performance, though.

AMD has been playing a distant second to Nvidia as of late; playing the price/performance angle makes quite a bit more sense from a financial perspective.

While developing the highest specs have always been the driving force behind GPU R&D, it still boils back down to sales. Of course both major GPU manufacturers offer a range of cards to cover all budgets, there's just higher profit margin on top end cards initially. Nvidia only recently broke their pattern by introducing the consumer friendly 9600 series well before the high performance 9k models.

But if AMD can provide the highest performance relative to price, the the average savvy consumer should end up buying their latest offerings.

As is the case with the majority of PC gamers who generally try to find the best price/performance buy, while the specs on a $600 video card are always impressive relative to the last generation, the price premium doesn't exactly match the performance jump. You always end up paying the highest premium for that top 10-20% performance increase.

And their point? the 9800GX2 is too. And the GTX 280 is stronger then the 9800GX2 so.... ya...

I don't get it....is 1 trillion calculations per second 1 teraflop?  Can't 360 do 1 teraflop, and PS3 do 2 teraflops?

Or is the point that this thing is more powerful than all console *GPUs* combined?

FLOP calculation is hard to measure on a standardized way, specially since CPU and GPU architectures are quite different... and it's even hard to compare FLOP performance between GPUs that have traditional pipelines (PS3), unified shaders (360 and GF280) and stream processors (Radeon HD4). All companies measure FLOP performance in a different way, so judgment shouldn't be based on that.

In any case, it should be based in things like texture performance, fillrate, memory bandwidth, etc... and games performance of course.

^^ Cool.  I was just wondering if a Tflop was what they were talking about with the trillion calcs per second...if so, I thought they must be comparing the AMD card with all console GPUs combined, not all consoles combined.

Well, if they are indeed calculating FLOPS in the way I think they're doing it (by shader performance), the RSX on the PS3 does around 140GFLOPS and the Xenos does around 200GFLOPS. Even if you took the Cell and the Xenon into consideration, their statement wouldn't be a lie.

But then again, even the $199 Radeon HD4850 does that.

Actually GPUs have gotten so powerful and so fast in recent years that they are really overtaking CPUs in raw processing capability for certain operations.

Nvidia has started releasing CUDA software which basically allows you to program with the GPUs.

Some of the results from the latest cards are showing 6x the raw processing capability than top CPUs for certain operations.

Rumor is also saying that ATI's next graphics cards are designed towards the value to mid-value price performance ranges.

Estimated costs are $400 for the high cards and $200 for the lower cards. These are still top quality performance though, they are just ceding the super high performance market to Nvidia.