Product and service reviews are conducted independently by our editorial team, but we sometimes make money when you click on links. Learn more.

New U.S. Supercomputers To Break 100 Petaflop Benchmark, Use Nvidia NVlink

By - Source: NVIDIA

U.S. laboratories today announced two new supercomputers. If all goes according to plan, they will become the world’s two fastest supercomputers, beating out China’s Tianhe-2. In fact, both machines are planned to double the previous performance record.

The two supercomputers, dubbed "Summit" and "Sierra," are to be built by the Oak Ridge National Laboratory and the Lawrence Livermore National Laboratory, respectively.

Oak Ridge’s Summit will have a theoretical max performance of 300 petaflops, though consistent performance will fall closer to 150 petaflops. The machine will be available for open science projects. Sierra is expected to deliver over 100 petaflops, and it will be used in Lawrence Livermore’s nuclear security research program.

To put these number in perspective, one petaflop is the ability to make one quadrillion calculations per second. The fastest supercomputer in the U.S., Oak Ridge’s Titan, delivers 27 petaflops of performance, while Tianhe-2 at the National Super Computer Center in China delivers 55 petaflops.

The two machines will be a part of a collaboration between IBM and Nvidia through the OpenPOWER Foundation. The new supercomputers will be built on an IBM POWER server architecture and utilize Nvidia’s Tesla GPU accelerator, as well as new NVLink GPU interconnect technology.

Nvidia’s NVLink will be used for communication between the CPU and GPU. Most GPU to CPU connectors now use PCIe interfaces that can limit bandwidth, particularly when multiple GPUs are used through a PCIe switch. NVLink aims to provide up to 200 Gb per second of bandwidth, effectively eliminating any restrictions to accessing CPU memory from the GPU.

Nvidia also plans on having the Volta GPU architecture as an integral part of the supercomputers. Volta features a stacked DRAM design, and according to Nvidia, could deliver up to 1 TB per second of bandwidth.

"Scientists are tackling massive challenges from quantum to global to galactic scales. Their work relies on increasingly more powerful supercomputers," said Jen-Hsun Huang, CEO and co-founder of NVIDIA. "Through the invention of GPU acceleration, we have paved the path to exascale supercomputing -- giving scientists a tool for unimaginable discoveries."

For more information on NVIDIA’s GPU technology head to High Performance Computing page.