Well, well, well, NVIDIA – I guess it’s show off time. While PC gamers the world over enjoy the quiet satisfaction of just about getting the Witcher 3 to keep its head above 30FPS water with their 4GB graphics cards, the video giant reveals more plans about the architecture of its new generation of Pascal GPUs, replete with 16GB of memory and capable of 1TB/s bandwidth speeds.
Want faster graphics performance now, not tomorrow? Give our NVIDIA GeForce GTX 970 group test a look-see.
Pascal cards are manufactured using a new 16nm ‘FinFET’ manufacturing process, a considerable shift on the current 28nm process used to produce Maxwell cards that also marks the shift from ‘planar’ (2D, basically) transistors to ‘FinFET’ 3D. That should result in significant powers savings on the current NVIDIA architecture, say people much cleverer than I.
While Pascal cards will theoretically support 32GB of memory, the first wave GPUs will launch with 16GB of HBM2 memory. HMB2, in case you were wondering, is the second generation of stacked High-Bandwidth Memory produced by JEDEC. Architecturally, the 16GB model is packed with four 4GB modules, capable of 1TB/s bandwidth speeds.
There’ll be multi-GPU packages for server tech too, but let’s not depress ourselves even further by comparing our trusty GTX 970s to £5,000 server cards with 24GB of memory.
The tech world has known broadly about Pascal’s architecture for a few months now, after NVIDIA first revealed it at GTC (the GPU Technology Conference) back in March, but they’ve since delved deeper into Pascal at GTC Japan.
Interestingly, AMD’s upcoming new generation of GPUs is shaping up in a similar fashion – same 16nm manufacturing process and HBM2 memory standard. In fact, they’ve even announced that they’re working on CUDA compilers for their GPUs.
Pascal’s launch window is currently pencilled in for Q1/Q2 2016, so you have at least another few months to enjoy that 4GB card without feeling massively inferior. PC gaming, eh?