Nvidia’s Kepler GPU should emerge in March 2012 with the budget version, the GK107, with a 128-bit memory interface, GDDR5 memory and support for PCI-Express 2.0 being the first out.
According to Japanese gaming site 4Gamer.net, the successors to Nvidia’s Fermi architecture and based on TSMC’s 28nm process, should have been out about now. However, production has been pushed back to early Q2 2012.
Kepler GPUs will have a triple dual-precision floating point performance of Fermi and can reach six dp Gflops if it is running downhill with the wind behind it.
4Gamer claims that Nvidia will release a mid-range GK196 GPU in the second quarter to replace the mature GTX 560. It will have a 256-bit memory interface, GDDR5 memory and support for faster PCI-Express 3.0.
It is expecting a higher end GK104 GPU with 384-bit memory interface and GDDR5 memory to appear in the early second half of 2012.
A dual-GPU version, the GK110, should be in the shops sometime in the third quarter of next year but Nvidia’s high end GK112 card, which will have a 512-bit memory interface, will not be seen until late 2012 or even early 2013.
Of course Nvidia is not saying anything and denied any delay to its Kepler launch schedule. It told Electronic Times that its transition to 28nm had gone a lot better than the 40nm move and yields were better than expected.
Fermi was notoriously delayed when they launched in the first half of 2010, owing to a complex design and large die size.
However, an Nvidia spokesperson more or less confirmed the delayed timeframe set out by 4Gamer, although was less specific.
If Kepler is delayed then it will have to launch against AMD’s Southern Islands GPUs which are due out earlier in the year. The Radeon HD 7900-series should come out with the higher end models first with Tahiti is due in January. This means that AMD will be two generations ahead of Nvidia for the first part of next year,