Some more details on Nvidia’s GK104 Kepler have been revealed which show that the chip would feature a hefty 1536 Cuda Cores and arrive in two variants. The specs of the GeForce GTX680 (Name not official Yet) were detailed by 3DCenter while Semiaccurate tells that the chip would be available in two variants.
The Specification shows that the Cuda Core Architecture would get a completely new design and would drop hot clocks being featured on previous generation cards. The Top GK-104 chip would be known as the GTX680/780 and would be manafactured using TSMC’s 28nm Fabrication Process.
GK-104 would pack 96 Stream Processors per cluster totaling at around 1536 Cuda Cores, 128 Texture Units, 32 ROP’s and a 2048Mb GDDR5 memory as standard, Core clock frequency would be set at 950Mhz.
|Nvidia Kepler GK-104:
The die size for both GK104 and GK110 have been revealed by Charlie on Semiaccurate which match the 3DCenter’s details. Semiaccurate has also release a new article which shows that the GK104 would be launched in two variants, GK104-400 and GK104-335 (Source). Such product codes were recently used on the Fermi based GTX560 and 560Ti cards where GTX560ti was marked GF114-400 and GTX560 was marked as GF114-325. This could suggest that specific GK104 based models would come with shader clusters disabled.
Other details include a 225W TDP and a price point of $299 for the performance segment cards, Performance is rumored to be faster than the AMD’s recently launched HD7970 Tahiti based card which is currently the fastest GPU on the planet!
Kepler is expected to launch in April 2012.
More details on Nvidia Kepler: