Finally! After a few very long months, a new "big" GPU has arrived. Nvidia’s GT200 pushes GeForce 8 architecture further and promises to replace the GeForce 9800 GX2 by a card equipped with a single GPU.An enormous GPU: not easy
Jen Hsun Huang has regularly reaffirmed Nvidia pledge in the continuing development of high end monolithic graphic solutions, or in other words, cards equipped with a single large GPU and not 2 more modest chips. However, this is a discourse which seems to be in contradiction with the GeForce 9800 GX2, Radeon HD 3870 X2, etc. While this strategy appears to be suited to AMD which unfortunately only has limited resources, Nvidia sees things differently.
Jen Hsun Huang, Nvidia’s CEO, isn't afraid of financing the development of enormous GPUs.A big GPU is profitable for Nvidia, first of all, because the GeForce brand has the wind in its sails and is selling well. Also, Nvidia can use the new GPUs on other very profitable markets with the Quadro and (if everything goes well) with Tesla cards – or at least this is Nvidia’s wager. For these reasons, why should they deny themselves?
The GT200 is thus an example of immoderation and is in fact the biggest GPU ever made with its 1.4 billion transistors engraved in 65 nanometers. Moreover, it’s a challenge for TSMC. The size of the chip should be around an unprecedented 600 mm², while the previous monster in terms of sheer size was the G80 of the first GeForce 8s. It wasn’t far from 500 mm².
The GT200’s die is enormous.Developing such a GPU isn't easy and for this reason there were delays. Jen Hsun Huang having indirectly announced it for late 2007 means a 6 month delay before the card is ready to be commercialized. Therefore the GT200 isn't a reaction to AMD’s RV770 as some might have said, even if it comes at the exact time to prevent AMD from taking the lead. The GT200’s delay can partly explain the disorder in Nvidia’s GeForce 8 and 9 lines as it seems evident that’s its absence posed a problem.
For this reason, Nvidia intends on brushing the slate clean of the names of its previous GPUs. Moreover, the GT200’s codename changed several times (and you will notice that ‘’G200’’ is written on the chip and no it’s not a Matrox chip). Graphic cards that integrate it will be named the GeForce GTX 200, starting with the 280 and 260 in the beginning.
So, what can be done with 1.4 billion transistors and GeForce 8 based architecture?
