NVIDIA CEO, Jen-Hsun Huang, showcased what he described as the “world’s biggest GPU” during the GTX 2018 keynote; the DGX-2. According to Huang, the DGX-2 is the first two petaFLOPS computer that comes with up to 16 Tesla V100 GPUs.
Adding on to the details, the DGX-2 sports comes with a whopping 32GB of HBM2 memory; aided by a healthy 1.5TB of on-board system memory. In regards to the aforementioned 16 Tesla V100, the DGX-2 scales the GPUs via the use of NVIDIA’s NVLink Interconnect technology or simply NVSwitch for short. To put it simply, NVSwitch allows all the GPUs in the world’s biggest GPU to “drive simultaneous communication” between each other at a blazing 300GB/s each.
The DGX-2 also comes equipped with two Intel Xeon Platinum CPUs, supports dual 10 and 25 GbE Ethernet, eight EDR Infiniband Ether with 1600GB/s bi-directional bandwidth and low-latency, and a massive 30TB internal storage capacity via a set of NVMe SSDs.
Naturally, all this hardware doesn’t come cheap; The NVIDIA DGX-2 is priced at US$399,000 (~RM1.5 million), and if you’re wondering why it’s expensive; that’s because this (obviously) isn’t your average GPU. NVIDIA designed its GPU supercomputer specifically for science and medical institutions that are currently or beginning to implement the benefits of Deep Learning into their scientific studies.
On that note, if you just so happen to be a member of particular scientific body looking to drive big data, you can now place an order for the DGX-2 via NVIDIA’s official site.