The G70 graphics processor weighs in at a decidedly hefty 302 million transistors, according to NVIDIA's estimates. Contrast that to an estimated 222 million transistors on the NV40, and you get some sense of this chip's scope. The more important measure of a GPU is probably its die size, so let's take a look at that. To put things into perspective, I ripped the heat sink off of our GeForce 7800 GTX card and took a picture of the GPU next to a quarter:
That's big, folks. I measured it, with my highly accurate wooden ruler, at about 18.5mm by 18 mm, which works out to 333 mm2. For comparison, the dual-core Athlon 64 X2 and Pentium D processors are each in the neighborhood of 200 mm2, and the ATI R480 chip powering the Radeon X850 XT is about 297mm2.
The G70 is manufactured by foundry firm TSMC on its 110nm fab process, a "half-node" process that's a partial step down from the more common 130nm node.
Despite its size, the G70 shouldn't be too terribly bad about power consumption, thanks to the process shrink and some other important factors. NVIDIA is confident that they can handle the power and heat problems common to modern semiconductors thanks to the parallel nature of graphics, and the G70 is a poster boy for their approach. Rather than crank up the clock speed, they've just made the chip wider; it will run at a relatively leisurely 430MHz, only 5MHz faster than most PCI Express versions of the GeForce 6800 Ultra. The G70 employs more extensive clock gating than the NV40, turning off portions of the chip when not in use, and NVIDIA says it does lots of dynamic clock scaling in response to demand, as well. All told, the GeForce 7800 GTX requires no more power than the GeForce 6800 Ultraslightly less, in fact, the company claims. Given the 7800 GTX's higher performance, performance per watt is up, as well.
The GeForce 7800 GTX, which should be available from multiple vendors using the same basic board layout and cooler, is a PCI Express card with a single-slot cooler.
The board sports 256MB of GDDR3 memory clocked at 600MHz, and it has an SLI connector on top, in case just one 7800 GTX isn't good enough for you. The official power supply recommendation for a 7800 GTX card will be a 350W PSU, or a 500W PSU for an SLI rig. Not bad, all told. There's no AGP version yet, but NVIDIA says they'll consider making one if there's sufficient demand.
Here's the tricky part for your poor, abused credit card: these things will cost a whopping $599, and they should be available right now from several vendors, at least in North America. If you have the uncontrollable urge to always own the latest and greatest stuff, it's gonna cost you this time, because two times $599 ain't cheap.
I should note here that NVIDIA seems to have left itself room for an even higher end version of the 7800 possibly to be launched once they've seen how ATI's upcoming R520 product performs. The single-slot cooler, only 256MB of memory onboard, the GTX nameit all adds up. NVIDIA could easily cut the price on this puppy and introduce a GeForce 7800 Ultra with higher clock speeds, a dual-slot cooler, and 512MB of memory. It might not happen, though, if the company is happy with this product's competitive standing.
So how does a $599 graphics card perform? What about a pair of 'em in SLI? Let's have a look.