Nvidia Pascal gets 16 GB HBM2 – manufactured on 16 nanometers by TSMC

Due to the fact that TSMC has not succeeded in developing a new advanced manufacturing process for high-performance circuits, AMD and Nvidia have been stuck at 28 nanometers for almost four years now and it will be some time next year before that changes. This has led to speculation that the players intend to turn to Samsung for the production of the next generation of graphics processors.

Most relevant in rumors has been Nvidia, not least when the company in a financial report confirmed that Samsung will produce a graphics processor for the company. However, a report from Business Korea now states that Samsung has not managed to win the contract for Pascal, the successor to the architecture Maxwell, but that Nvidia will remain with TSMC, which has been a partner for 20 years.

Pascal will thus be manufactured at TSMC on a 16-nanometer technology with so-called 3D transistors or “FinFET”, where doubled transistor density is accompanied by higher energy efficiency compared to today’s 28 nanometers.

The successor to the GM200 graphics processor, which is used in the Geforce GTX Titan X and GTX 980 Ti, is expected to go by the name GP100 and will, according to previous rumors, get no less than 17 billion transistors. This can be compared to 8 billion for the GM200 or as much as 8.9 billion for AMD Fiji, which in terms of number is currently considered to be the world’s most advanced circuit.

In addition to the new information, Nvidia is preparing to hold the GPU Technology Conference for developers in Japan on September 18. Prior to this, presentation photos have been published, which reveal more details about what to expect from Pascal.

It is already known that Pascal will use the memory standard HBM2, which doubles the memory bandwidth against HBM1 in the Radeon Fury series. This gives a memory bandwidth of one terabyte per second and now it appears that the capacity lands at 16 GB. It is worth mentioning that 32 GB will be possible with HBM2 and thus may be relevant in specific variants of Pascal, for example the Tesla and Quadro series.

The first graphics processor with architecture Nvidia Pascal is expected to appear in early 2016.

Source: Wccftech.