Nvidia ‘Deep Learning’ Supercomputer To Tackle AI Challenges

Nvidia has bolstered its HPC (high-performance computing) credentials with the release of the Nvidia DGX-1, which it claims is the world’s first deep learning supercomputer.

And the company also announced the Nvidia Tesla P100 GPU, said to be the most advanced hyperscale data centre accelerator ever built.

Deep Learning

The Nvidia DGX-1 supercomputer however is said to offer the throughput of 250 x86 servers to meet the increasing tough computing demands posed by artificial intelligence applications.

The DGX-1 is designed specifically for deep learning, and will work with neural networks.

“Artificial intelligence is the most far-reaching technological advancement in our lifetime,” said Jen-Hsun Huang, CEO and co-founder of NVIDIA. “It changes every industry, every company, everything. It will open up markets to benefit everyone. Data scientists and AI researchers today spend far too much time on home-brewed high performance computing solutions. The DGX-1 is easy to deploy and was created for one purpose: to unlock the powers of superhuman capabilities and apply them to problems that were once unsolvable.”

Nvidia points out that the DGX-1 is helped because it is built on its Tesla P100 GPUs. These GPUs are based on the new NVIDIA Pascal GPU architecture and it provides the throughput of 250 CPU-based servers, networking, cables and racks – all in a single box.

And Nvidia has also squeezed out performance improvements thanks to the NVLink high-speed interconnect for maximum application scalability. The company claims this 16nm FinFET fabrication technology allows for unprecedented energy efficiency. Additionally, the chip on wafer on substrate with HBM2 is geared for big data workloads; and new half-precision instructions delivers more than 21 teraflops of peak performance for deep learning.

“NVIDIA GPU is accelerating progress in AI,” said Yann LeCun, director of AI Research at Facebook. “As neural nets become larger and larger, we not only need faster GPUs with larger and faster memory, but also much faster GPU-to-GPU communication, as well as hardware that can take advantage of reduced-precision arithmetic. This is precisely what Pascal delivers.”

The specs of the Nvidia DGX-1 are impressive. It offers uup to 170 teraflops of half-precision (FP16) peak performance. This is thanks to eight Tesla P100 GPU accelerators, 16GB memory per GPU, as well as NVLink Hybrid Cube Mesh, 7TB SSD DL Cache, and Dual 10GbE, Quad InfiniBand 100Gb networking.

The Nvidia DGX-1 supercomputer also comes equipped with a complete suite of deep learning software that is designed to help researchers and data scientists to quickly and easily train deep neural networks.

Tesla P100 GPU

Meanwhile the Tesla P100 GPU, which is used in the new supercomputer, should provide data centres with improved power to process large numbers of transactional workloads, such as web services.

“Our greatest scientific and technical challenges – finding cures for cancer, understanding climate change, building intelligent machines – require a near-infinite amount of computing performance,” said CEO Huang.

The first accelerator to deliver more than 5 and 10 teraflops of double-precision and single-precision performance. This, says Nvidia, means that the Tesla P100 provides a giant leap in processing capabilities.

The world is currently locked into a supercomputer arms race. Last July President Obama signed an executive order for a HPC initiative to built the world’s faster supercomputer.

The initiative, known as the National Strategic Computing Initiative (NSCI), will see the United States develop the fastest supercomputer in the world capable of running at 1,000 petaflops or one exaflop.

Think you know all about supercomputers? Try our superquiz!