Nvidia announced this week the release of a new 3U server, also referred as a supercomputer called DGX-1, as well as the release of a new Pascal based GPU accelerator called Tesla P100.
The DGX-1 is aimed to help businesses, researchers and scientist understand and analyze a huge amount of data.
This new server is considered a full-scale one because it works with the Tesla P100 in its maximum capacity. It is actually a server designed to implement the P100.
The DGX-1 is the only server so far that allows people to use and work with the new GPU.
The Tesla P100 is the largest GPU that is being produced nowadays, using FinFET Technology and being interconnected with Nvidia NVLink. The performance of this GPU is two times better than previous GPUs from Nvidia.
This new GPU is a graphics card designed for professionals and deep learning systems. Gamers and all types of consumers can also use this GPU for their own purposes, even though that is not the aim.
The DGX-1 main benefits are that, first, it can bring artificial intelligence technology into different business, and it also has access to cloud management tools, having also software updates.
Companies were starting to invest too much money into deep learning software, and Nvidia decided to be part of this new trend and release the DGX-1.
This server includes a 512GB memory, 8 Tesla P100 GPUs, 16-core Xeon CPUs, dual Ethernet of 10Gbps and 100Gbps InfiniBand ports, plus a 7TB solid-stage storage.
The Tesla P100 has four 40GB per second link which makes a total of 160GB per second of bidirectional bandwidth in between GPUs. It allows programs to access the memory in the systems CPU.
Tesla P100 will start shopping in the beginning of 2017, meanwhile Nvidia plans to release the DGX-1 in June.