NVIDIA presentation was dedicated to accelerators NVIDIA Tesla computing. Let's start with the fact that the first generation of accelerators computing Volta appear at the end of 2017, NVIDIA will start implementing them in two supercomputer systems (Summit and Sierra).
In the consumer market graphical solutions generation Volta unlikely to appear before 2018.
Pascal architecture can host up to 16 GB of memory type HBM - we are talking about the second generation memory, but there is no way to predict what will be the maximum amount of memory in consumer graphics solutions in this family .
In a further development of its GPU NVIDIA company will pay special attention to improving the speed of interaction of multiple graphics processors to each other, as well as the implementation of the principles of unification of the memory address space.
Substantial assistance in speeding up the transfer of information will be NVLink interface that allows graphics processors to communicate with each other at speeds of up to 20 GB / s in one direction. A pair of GPUs will be able to communicate in four lines NVLink, with an aggregate bandwidth of 80 GB / s.
In speedup oriented systems , this optimization will make a huge difference. Related Products :
|