Nvidia reveals its shockingly powerful next-gen Volta graphics tech

Nvidia has revealed its new Volta technology, which follows on from the current-gen Pascal graphics cards, along with the first ever GPU to use the tech: the Tesla V100.

As those familiar with Nvidia’s naming conventions will realize, this is a heavyweight GPU targeted at data center usage and high-performance computing (as opposed to consumer PCs).

However, Nvidia envisages it bringing a whole new level of power to the arena of artificial intelligence and related AI services, which will help to seriously bolster the strength of things all of us will use like digital assistants, or advances in self-driving vehicles.

Plus the new GPU gives us a good idea of exactly what Volta is capable of, and in short, it’s massively powerful.

Volta is Nvidia’s seventh-generation GPU architecture, and represents nothing less than a fivefold performance improvement compared to Pascal in terms of peak teraflops. And in comparison to Maxwell, Pascal’s predecessor, we’re looking at a 15x boost.

Clever combination

The new technology combines CUDA cores with Volta Tensor Cores, the latter of which are specifically designed to speed up AI workloads very effectively. Indeed, a single Tesla V100 GPU is apparently capable of replacing ‘hundreds of commodity CPUs’ when it comes to high-performance computing.

The Tesla V100 is built on a 12nm process and boasts some 21 billion transistors, with 5,120 CUDA cores alongside 640 Tensor Cores. In terms of deep learning performance, Nvidia says it hits 120 teraflops. Single-precision compute performance is 15 teraflops.

The GPU also utilizes 16GB of HBM2 DRAM with 900GB/s memory bandwidth, which is 50% more bandwidth than the previous generation. Power consumption is 300W.

Nvidia’s fresh revelation here comes ahead of AMD’s next-gen Vega launch, with the new Radeon cards expected to be unveiled next month.