Nvidia launched its largest GPU supercomputer till date, DGX-2. The graphics processor is super big and even bigger than your expectations. Its not the typical GPU you can plug in your motherboard. Its a Devil.
The Nvidia DGX-2:
Its the second GPU in the DGX segment and offers 10x faster performance. The DGX-2 is basically 16 Volta GPUs stacked together using NVSwitch (a high performance external interconnect) driven by 2x Intel Xeon Platinums chips which pack a total of 56 cores. This latest GPU from Nvidia delivers a stunning performance with 2 petaflops processing power. The server packs a total of 81,920 CUDA Cores with 512 GB HBM2 memory (RAM) and a 14.4 TB/s aggregate bandwidth with 300 GB/s GPU to GPU transfer. It consumes a massive 10,000 watts and weighs 350 pounds (158.757 kg); our tiny hands won’t be enough even to lift this beast.
You may be thinking how fast can it go. Well, the world’s current fastest computer, LINPACK can process data at Rmax of 93 petaflops and Rpeak of 125 petaflops. Now, you do the math.
It has 8x EDR 100 Gigabit ethernet for connectivity. Onboard is 30 teraflops storage and 2 CPUs for fastest possible data processing. Some of you may think it as a GPU but the setup is actually a supercomputer. It’s built with a combinations of GPUs.
There is one very sad news. You can’t play your high-end games on DGX-2. Though it has a lot of GPUs, well you can’t. A simple explanation ? Because its not built for gaming. Even if you wanna give it a spin, it costs $399,000 (2599285.5 INR). Unless you are a saudi prince, don’t even think.
DGX-2 is purpose-built for data scientists for pushing the limits of deep learning research and computing. It has the updated suite of NVIDIA deep learning software. Well if you can afford it, you can have your own personal deep learning supercomputer.
Read other interesting announcements other than world’s largest GPU from Nvidia’s event on their blog.