Nvidia has made it clear that it has moved on from the days where its main purpose was to manufacture products solely for gamers at its GPU Technology Conference this year. And what better way to make that statement then by introducing its first deep learning supercomputer called the Nvidia DGX-1.
So, what makes Nvidia’s DGX-1 special? Well, for starters, it comes with eight Tesla P100 GPUs, which were announced during the event as well. Storage-wise, the DGX-1 supercomputer is packed with 7TB of unspecified SSDs. Amazingly, this deep learning monster needs a whopping 3200W of power to run efficiently.
According to Nvidia, the DGX-1 is equipped with two Intel Xeon processors and has FP16 performance (half-precision floating point) of up to 170 TFLOPS. Nvidia describes the performance of the DGX-1 as being similar to that of 250 servers – all in a single box. Nvidia’s CEO, Jen-Hsun Huang, mentioned that the purpose of the DGX-1 was to “unlock the powers of superhuman capabilities and apply them to problems that were once unsolvable”.
Nvidia will initially make the DGX-1 available for “early adopters” of deep learning and its pioneers; this includes well known academic institutions and companies such as Google and Microsoft. If it wasn’t obvious before, then one should keep in mind that the DGX-1 is meant solely for enterprise and research usage, which is why it’s priced at US$129,000 (about RM507,776).
More information regarding the Nvidia DGX-1 as well as pre-order request can be found here.