Originally, Kepler GPU architecture was scheduled to debut in 2011, with the successor codenamed Maxwell coming in 2013. We were not surprised to hear that Maxwell, tied to a 20nm process node – is getting delayed as well.
"The battle with manufacturing nodes continues." – This would be a summary of what our sources told us over the past couple of weeks. No manufacturer is having an easy time, and even Intel – known for its manufacturing excellence – pulled its 22nm Low Power node deeply into 2013, debuting about half a year later.
When it comes to Nvidia, the company is heavily engaged on increasing the yields for its products. With Kepler, Nvidia shifted to a new die strategy, and even though we've heard more rumors about the GK106 chips, the fact of the matter is that Kepler will remain a three GPU line-up: GK104 for performance, GK107 for mobile and entry-level desktop and the GK110 for high-end computational and visualization parts. Ultimately, GK104 served as the GTX 660/670/680/690, Quadro K5000 and Tesla K10, GK107 serves in numerous low-power mobile and desktop designs. GK110 will make a public appearance only in December as the Tesla K20, with a yet unannounced amount of video memory (according to our sources, NV is trying to secure different packaging to enable 12GB of GDDR5 memory, but realistically – prototype boards we saw utilized "just" 6GB).
Naturally, all of these movements don't hold much promise for the next generation GeForce cards. Given that the Maxwell, GM1xx parts won't be available until the first half of 2014, the GeForce and Quadro parts should continue to rely on refreshed/renamed GK104/107/110 parts. The GK110 is planned to expand and become available as the Quadro K6000 6GB, for those that require ultimate performance on a single piece of silicon.
In case you've doubted it, Maxwell is a 20nm part, capable of being manufactured in GlobalFoundries, IBM, Samsung and TSMC – 20nm Gate-Last HKMG will equalize between TSMC and Common Platform alliance – a move Nvidia and Qualcomm cannot wait to see come to frutition.
Availability of that part however, will not come sooner than the first half of next year. As far as the GeForce GTX 700 Series is concerned, the parts should repeat the same cadence as this year's lineup. However, the newer revised parts should offer further clock improvements, to the level where they can offer between 25-30% higher performance and power efficiency, while the company is working on parts based on the Maxwell GPU architecture, which should drive 2014 as one of key years in Nvidia's history. Maxwell will be the first top-to-bottom GPU architecture, powering everything from Tegra to Tesla. Furthermore, Maxwell should be the first GPU part to integrate the 64-bit ARM core which carries the codename "Project Denver". Putting the typically-bandwidth starved ARM cores onto an internal bus which in GPUs goes beyond 1.5TB/s should significantly change the playing game – a GPU capable of booting an operating system, regardless of what lies currently in public documents.
All in all, 2013 will see AMD's Sea Islands fight first versus Nvidia's Kepler refresh, and only then against the Maxwell. Real battle will come only in 2014.