VR-Zone

Fudzilla has reported the existence of a dual-GPU version of Nvidia's upcoming GF100 in April, roughly one month after the single-GPU GF100 launch, if all goes well and A3 is the final stepping. However, is such a mammoth product, the dual-GF100, actually feasible?

Earlier rumours pointed to a TDP of 180-200W for the single GPU GF100, but the recent demonstration in CES 2010 saw a 8 pin + 6 pin PCI-e power configuration, suggesting a TDP of somewhere between 225W and 300W. Since then, SemiAccurate have reported the actual TDP to be a rumoured 280W.

We must keep in mind that the silicon demonstrated at CES 2010 is most certainly not the final stepping. Even so, barring a major miracle, the power draw of GF100 is likely to remain north of 225W. This kind of power draw is no surprise and something we would expect from a 3 billion transistor, 500+ mm² monster GPU. Unfortunately, this means that a dual-GPU GF100 is never going to make sense, irrespective of agressive downclocking and binning, at under 300W TDP. Technically, Nvidia can break the 300W barrier, but this will mean violating PCI-e regulations and foregoing PCI-e official compliance.

In essence, this is GT200 all over again. It was the last Nvidia single-GPU to carry a 6 pin + 8 pin configuration - and as history speaks, Nvidia was unable to make a dual-GPU iteration till the GT200b die shrink, simply because the single GPU ran too hot to begin with.

What this means is GF100's direct competitor is likely to be the ATI Radeon HD 5970, not the single-GPU Radeon HD 5870, despite rumours suggesting an existence of a dual-GF100 product. Once again, recalling the GTX 280 vs. HD 4870 X2 battle. However, GF100 looks to be more promising, albeit delayed, than the GTX 280. The question is, can it beat two of ATI's Cypress?

Reference: Fudzilla, SemiAccurate