Not only has the new NVIDIA Quadro GV100 been unveiled, the incredible NVIDIA GDX-2 server has also been unveiled, which integrates a whopping 16 NVIDIA Tesla V100s.

The new NVIDIA Quadro GV100 has not arrived alone and it has come together with the NVIDIA DGX-2, a monster that arrives six months after the NVIDIA DGX-1. The DGX-2 is ten times the performance of the DGX-1, which is impressive, but the hardware explains it. This system implements a total of 16 Tesla V100s interconnected by NVSwitch technology. This allows all of them to work together, counting the system in total of 81.920 CUDA Cores, which means that we have a computing power of 2 PETAFLOPS.

In addition, it has a total of 512GB of HBM2 memory, with a total bandwidth of 14.4TB / s, which means that each GPU has a bandwidth of 300GB / s. All this set has a total weight of 159Kg, which is a lot and has an extremely low consumption, since it consumes only 10kW.

For the system to work, two Intel Xeon Platinum 8180 processors have been integrated, which offers us a total of 56 cores and 112 processing threads. The two processors work at a base frequency of 2.5GHz and in Boost mode they can reach 3.8GHz. The system also has 1.5TB of DDR4 RAM and a total of 30TB of NVMe SSD type has been integrated.

Jen-Hsun Huang has also said that this DGX-2 system is no less than five hundred times more powerful than two NVIDIA GTX 580s. The DGX-2 system costs $ 399.000, which may seem like a lot of money, but the large servers that would match in power to this system, they occupy much more, consume much more and their assembly cost is around 1.5 million dollars.

Experience new levels of AI speed and scale with the NVIDIA DGX -2, the first 2 PETAFLOPS system to combine 16 fully interconnected GPUs for 10x the deep learning performance. It's powered by NVIDIA DGX software and a scalable architecture built around NVSwitch, to meet the world's most complex AI challenges. '