NVIDIA’s 28nm mobile line-up revealed
SemiAccurate have leaked a tentative line-up for NVIDIA's 28nm mobile GPUs. The first 28nm mobile GPUs are set to be sampled in Early October, with mass production commencing in End December if everything goes to plan. Availability is scheduled for Q1 2012. Crucially, these 28nm parts are not based on the next-gen Kepler architecture, but are die shrinks of the current Fermi architecture powering GeForce 400/500 series.
The new 28nm chips are codenamed the N13 family, succeeding the N12 series – the current GeForce 500M Series. The codenames follow a similar pattern to N12, with three categories – Performance (N13P), Mainstream (N13M) and Enthusiast (N13E). The first and only Mainstream part, the N13M-GS is most likely expected to be a GF119 shrink, replacing the GT 520M. The N13P-GV also appears to be based on GF119, but featuring GDDR5 memory, pushing it up to the P class. N13P-GL and N13P-LP appear to be GF108 shrinks, while N13P-GS, N13P-GT and N13E-GE could very well be based on GF116.
Finally, the two flagship parts – N13E-GS1 and N13E-GTX feature 256-bit memory and 2GB GDDR5 memory and certainly appear to be GF114 shrinks. The top N13E-GTX part is only expected to be in mass production in April/May 2012.
3DMark score estimates are also mentioned for the high performance parts, presumably 3DMark Vantage P scores. The current fastest mobile GPU, Radeon HD 6990M, scores in the range of P15500, while the fastest NVIDIA GPU, the GTX 580M scores P14500. This puts the N13E-GS1 right between the GTX 580M and the HD 6990M. The N13E-GTX, scheduled for Q2 2012, takes performance up a notch to P20000. It will also lower the TDP to 75W, down from GTX 580M's incredible 100W. However, major performance improvements will need to wait till the true next-gen GPUs in Southern Islands and Kepler.
Of course, this is only a tentative line-up, and many things could change. The question remains – if we get Fermi refreshes in Q1 2012, when can we expect to see Kepler based GPUs in the mobile arena? Also, could we see 28nm Fermi shrinks in the desktop market before Kepler releases?