Report: Nvidia Next Generation Cards Could Be 16nm, with Up to 32GB RAM
While AMD readies its next top-of-the-line Fiji XT-based Radeon R9 Fury X and R9 300 series of cards, Nvidia is currently busy testing what will be there biggest and meanest GPU yet.
The GP100 from Nvidia will be a goliath, boasting between 4500 and 6000 CUDA cores. The ‘P’ in the GP stands for Pascal, which is the next in-line architecture for Nvidia GPU chips. Currently, Nvidia’s chips are of the GM series, with the ‘M’ standing for Maxwell.
Pascal will not only boast more cores, but will also be built on a smaller 16nm chipset, with the futuristic HBM2 memory, which could see insane memory bandwidth scalings all the way up to 1.2TB/sec. This means a possible 400% increase in bandwidth memory from Nvidia’s current flagship GTX 980 Ti.
The target window, according to Beyond3D Forums, will be the first quarter of 2016. There is a possibility of some (or all) of the information being false, but there seems to be some sort of consistency, and there has been a general consensus that Nvidia and AMD would ultimately move on to the HBM2 memory.
The added advantage of a possible move to 16nm architecture will also mean that Nvidia will have a much larger headroom without the great concerns of heavy power consumptions, resulting in the Pascal cards to feature insane HBM2-based VRAM values that go up to 32GB.
Of course, this should all be taken with a pinch of salt, but we do expect some sort of revelation towards the final quarter of this year (or even earlier) from Nvidia.