The Inquirer, a British web site that is ground zero for computer industry gossip - Austin American Statesman
The firm would like to double up on G80 to a GX2 dual-chip version, if it can conquer the heat and power management problems that poses. After that it will be time for G9x flavour, codenamed G90, our senior industry sources tell us.
Those same sources say Nvidia needs a more cost-efficient way of producing G9x high-end chip based on the G80 marchitecture. Chances are that Nvidia is working with TSMC to produce the chips on a 65 nanometre process. The resultant shrunken chips would work with much higher frequencies, and powered with GDDR4. But we still have to confirm these details.
There are two challenges going down to 80 or 65 nanometres from the current 90. An optical shrink from 90 to 80 nanometre doesn't allow much space for clock speed increases. You cannot go much higher than 700+ MHz depending on the chip marchitecture. It
Going down to 65 nanometre is not an optical shrink and presents a major engineering challenge. That's why it took AMD ages to go from 90 nanometre CPUs to 65. And graphics chips are even more challenging to shrink. Nvidia will go 65 nanometre with G78 and if this goes well it will probably try to move all its chip to 65nm in order to make more money per wafer.
If Nvidia makes G80 smaller there may also be just enough space for 512-bit controller. If all goes as planned, the chip is scheduled for second part of 2007. We will keep our ears open. Nvidia never comments out forward-looking stories. µ
Sign up for INQbot – a weekly roundup of the best from the INQ