Following a series of reports that claimed Nvidia has decided to postpone the launch of its next-generation graphics cards, based on the Kepler architecture to the first quarter of 2012, a German website managed to get a hold of an Nvidia slide that seems to make this news official.
The slide was presented by Nvidia
during the International Supercomputing Conference and is entitled the “CUDA architecture roadmap.”
While we don't know if the roadmap refers only to the company's line of Tesla professional GPUs, or if it also covers consumer graphics cards, this clearly shows that Kepler isn't expected to arrive until 2012, while Maxwell is scheduled for a 2014 launch.
No information regarding the cause of the delay was provided, but most probably TSMC's still immature 28nm fabrication process is to blame.
Kepler is the name used by Nvidia for its next-generation graphics core and, just like AMD's Southern Islands GPUs, is a 28nm die shrink of the current Fermi architecture with a few minor tweaks and improvements meant to increase the performance of the chip.
According to the slide presented in this article, and to previous data supplied by Nvidia, these changes, combined with the new manufacturing process, should deliver 3 to 4 times the performance per Watt of the Fermi in double-precision 64-bit floating point operations.
Even though Kepler
was pushed back to Q1 2012, AMD is still on schedule to launch its 28nm chip in 2011.
According to some recent rumors, the first Southern Islands GPUs
will be released in September and the company's fastest core will go by the name of Tahiti.
This will be used for spawning both single-GPU and dual-GPU graphics cards, and the latter is code-named New Zealand. (via ComputerBase