
NVIDIA has engineering samples out for its first chip based on the next-gen Kepler architecture. The chip is the same as revealed by SemiAccurate last week – GK107. As suggested by the nomenclature, this isn't the flagship GK100 we are all waiting for. Instead, NVIDIA is executing a bottom-to-top release, first releasing a smaller chip, GK107, which will find its way into notebooks.
NVIDIA has engineering samples out for its first chip based on the next-gen Kepler architecture. The chip is the same as revealed by SemiAccurate last week – GK107. As suggested by the nomenclature, this isn't the flagship GK100 we are all waiting for. Instead, NVIDIA is executing a bottom-to-top release, first releasing a smaller chip, GK107, which will find its way into notebooks.
GK107 is set to feature in four mobile GeForce SKUs – N13P-LP, N13P-GS, N13P-GT and N13E-GE. If these codenames sound familiar, it is because they have previously appeared in a leaked 28nm mobile GPU line-up. GK107 features a 128-bit memory interface, and supports DDR3 and GDDR5 memory. The first three GK107 based SKUs (with a "P" suffix) will likely succeed the GeForce GT 500M series, and will presumably be branded GeForce GT 600M series. The top GK107 part, N13E-GE, may succeed GTX 560M and be part of the GTX 600M series, as denoted by the "E" suffix (Enthusiast). In addition, GK107 will also be part of mobile Quadro SKUs – N14P-Q1 and N14P-Q3.
GK107 marks the first time NVIDIA has moved to a new process and a new architecture simultaneously in a long time – mostly due to the cancelation of TSMC's 32nm process. NVIDIA's transition to TSMC's 40nm process featured a bottom-to-top strategy as well, with smaller GT21x shrinks leading up to the big Fermi dies. However, this is the first time that NVIDIA is releasing a new generation of GPUs with a mainstream part. Between reports of AMD's next-gen GPUs being pushed back to 2012 to now NVIDIA starting off with a small die, all evidence points towards the same conclusion – TSMC's 28nm process is promising to be just as troublesome as the infamous 40nm (if not more so).