NVIDIA’s Maxwell GPU architecture has been with us for a year and a half now, which means that a new architecture is just around the corner and we even happen to know its name – Pascal. But what will it bring?
Well, as of now things look rosy. A few days ago NVIDIA announced that TSMC would make its Pascal GPUs on 16 nm tech process, although Pascal was previously expected to come on 14 nm FinFET process. This might seem like a setback but in fact it is not – the magic of Pascal will be inside the GPUs – the new architecture will feature HBM2 memory, right from the start, which will provide up to 1 TB/sec of memory bandwidth, which is more than impressive. The flagship Pascal video cards will also be equipped with up to 16 GB of memory (or even more), which will make Pascal-enabled cards 3D video card monsters. As a comparison AMD’s flagship Radeon R9 Fury X graphics card comes with 4 GB of HBM1 memory that provides 512 GB/sec of memory bandwidth. The NVIDIA NVLink interface will also be accelerated – it will reach 80 GB/sec of memory bandwidth. Finally Pascal will come with significantly improved FP64 performance, which will benefit various scientific applications.
Pascal is expected to debut in 2016 but the exact date, or even time period, is unknown.
Source: Digitaltrends.com