Nvidia’s new Volta GPU architecture has been delayed until 2018 according to a roadmap from Nvidia. There is some extra information on Nvidia’s Pascal architecture. It is found from the new leaks which pretty much confirm what many of us within the industry have come to expect is that Pascal will be created using TSMC’s 16nm FinFET PLUS node, offering around 2x the density, 65% higher speeds or 70% less power than the 28nm HPM technology which is currently used to create Nvidia’s already efficient Maxwell based GPUs.
Its been revealed by the new roadmaps that Nvidia’s Pascal based GPUs will support mixed precision compute and will contain up to 16GB of HBM 2.0 memory, which would be able to deliver an amazing 1TB/s of memory bandwidth, also been revealed recently that Nvidia will be ordering HBM 2.0 memory from SK Hynix and Samsung which meaning that they should have a large enough supply of their new memory in order to make up a sizable portion of their upcoming Pascal lineup.
Anyway the next year would be very interesting for GPUs as their will be a big architecture change for both AMD and Nvidia, a full on move to ultra high bandwidth HBM 2.0 memory and a large process shrink from 28nm to 16nm FinFET Plus.
The next generation of GPUs will have one of the largest leaps in performance ever seen before moving from one generations to another. Let’s hope that this proves to be true.