NVIDIA announced in early May that several teams oriented to the professional sector are interested in their world’s first Deep Learning Supercomputer in box size. And from this week the green camp officially shipped the first of them, namely the NVIDIA DGX-1V. This mini-supercomputer is a beast in terms of performance thanks to employing eight NVIDIA Tesla V100, all connected together using the NVLink Hybrid Cube interconnect yielding a performance of 960 Tensor TFLOPs and backed by 128 GB of HBM2 memory.
To give us an idea, NVIDIA says that you won’t need to buy a rack of 400 servers if you have one high-powered NVIDIA DGX-1 supercomputer with a Volta GPU sitting on your desktop and all with the energy savings.
The first DGX-1V shipped to MGH & BWH Clinical Data Science Center (CCDS). In terms of specifications, equipped with the eight Tesla V100 backed by 2x Intel Xeon E5-2694v4 with 10 physical cores and 20 logic per CPU at a Base / Turbo frequency of 2.40 / 3.40 GHz. All this is watered cooled with no less than 512 GB of DDR4 @ 2133 MHz LRDIMM memory along with a RAID 0 of 4x 1.92 TB SSDs , dual 10GbE port, and a 3200W power supply.
“It comes out of the box, plug it in and go to work,” said Nvidia’s CEO Jen-Hsun Huang
The price of the NVIDIA DGX-1V is $ 149,000, something that could be still considered cheap, because you are getting the performance level of 400 servers in less space with low power consumption.