NVIDIA Volta-Powered DGX-1 And DGX Station AI Supercomputers Debut At GTC 2017
Soon after unveiling the Tesla V100 data center GPU based on its next-generation Volta architecture at GTC 2017, NVIDIA CEO Jen Hsun Huang announced an updated, Volta-infused DGX-1 server appliance for deep learning data center deployment, along with a totally new product, the NVIDIA DGX Station, that’s being dubbed a ‘Personal AI Supercomputer’.
The new NVIDIA DGX-1 is similar to the previous generation offering based on
There are a total of 40,960 CUDA cores (5,120 Tensor cores) in the system, with 128GB of total GPU memory, spread across those Tesla V100 processors. All told, the Volta-infused DGX-1 offers up to 960 TFLOPs of FP16 compute performance, versus 170 TFLOPs for the original, with significantly more bandwidth on tap.
The new DGX-1 is priced at $149,000, and will be delivered in Q3. Customers who buy one today, can receive the Pascal-based version, and have the Tesla P100s swapped for V100s when they become available.
NVIDIA also unveiled the new DGX Station. The DGX Station is designed to be a quiet, standalone AI workstation, and features up to four Tesla V100 16GB cards, for a total of 20,480 CUDA cores and 2,560 Tensor cores.
The system offers up to 480 TFLOPS of compute performance (half of the DGX-1) and also packs a 20-core Xeon E5-2698 v4 CPU, 256GB of system memory, quad 1.92TB SSDs (one for the OS, and 3 in RAID 0 for data storage). The systems are water cooled and function within a 1500W envelope, so companies that don’t necessarily have access to environmentally controlled data centers, can work with bleeding edge AI and Deep Learning applications, employing this as a personal workstation specifically tasked for these kinds of workloads.
The NVIDIA DGX Station is priced at $69,000, and will also begin shipping in Q3.