We've known for several years that the next GPU architecture after Kepler would be codenamed Maxwell. NVIDIA
revealed as much in a roadmap
as far back as September 2010 when Fermi was the top dog, and though it will arrive just a tad behind the GPU maker's original schedule, Maxwell parts are right around the corner.
Back in 2010, NVIDIA expected to be pushing out 20nm GPUs by now, though the reality is TSMC isn't up to the task of transitioning to 20nm territory at an acceptable yield. If NVIDIA decided to push the issue, costs would likely jump up dramatically on the manufacturing side, which would result in higher priced graphics cards. With AMD already being so aggressive with its pricing strategy, the last thing NVIDIA needs is to price itself out of contention.
That said, some Maxwell-based parts will eventually transition to more advanced process nodes, but for now, the new architecture will be built on a 28nm process, according to Videocardz.com
, which posted the first pictures of what's supposed to be NVIDIA's GM107 GPU (Maxwell). According to the post, the GM107 is a hybrid between GK107 and GK106, with a die size that's been reduced by 30 percent (compared to GK106).
Reportedly, there's going to be two versions of Maxwell at the outset: GM107-300 (GeForce GTX 750) and GM107-400 (GeForce GTX 750 Ti). If what is being reported is true, GM107-300 will have 768 CUDA processors, 64 TMUs, 16 ROPs, and a 128-bit memory bus. It's not yet known what the rated TDP will be. As for the faster GM107-400, it will supposedly sport 960 CUDA processors, 80 TMUs, 16 ROPs, a 128-bit memory bus, and a 75W TDP. Though, as always, take unconfirmed rumors with a grain of salt.
Look for both parts to come out on February 18 (less than two weeks from now).