We’ve known for long that GT200 will be a huge chip even at 65nm, it was said to be somewhere in the vicinity of 600mm2 and pack about 1 billion transistors. It’s now been confirmed that it occupies an area of 576mm2 (24×24). That’s 75% larger than the 65nm G92 chip, 330mm2, and even larger than the already big 90nm G80 chip, which measures 484mm2. A picture of the GTX280 IHS and the card, although very small, have surfaced over at Chinese hardware forum ITOCP. The pictures looks genuine, even if they are of an engineering sample, the final card will look pretty much the same. The IHS says G200-300-A2.
Some new information mixed with what we’ve told you before;
The GTX280 will require one 8-pin and one 6-pin connector, while the GTX260 will suffice with two 6-pin connectors. GTX280 sports a 512-bit bus and 1GB GDDR3 memory, while GT260 has to settle for 448-bit and 896MB GDDR3 memory as NVIDIA will disable two memory chips and two clusters of shaders.
GTX280 will have a full setup of shaders, meaning all 240 till be active, while GTX260 has 192 shaders. This may sound like a lot already, but NVIDIA also claims that the shaders themselves are up to 50% faster than the shaders of G80/G92.
Alas, there’s no DirectX 10.1 support, which is kind of poor when you think about it, even if there are extremely few games (any?) out there that actually uses it. Instead NVIDIA has put a lot of work into incorporating the support for its newly acquired PhysX engine, but we still have little to no information about the exact works of that.
Prices are still in the vague area, but we estimate somewhere above $500, simply because the die is huge and the yields can’t be that good when you can only fit a very small number of chips on each wafer. Each die is estimated to cost more than $100 according to Theo’s sources. Also, this is an extreme high-end card, it’s for showing off mainly, not so much for making big sales.
