​IBM and Nvidia has Completed the World’s Fastest Supercomputer

Jun.13.2018

Author :Justin Brunnette

Category: IT News

IBM and Nvidia has Completed the World’s Fastest Supercomputer

As we have seen a fierce chip competition between Ryzen and Intel, the affordability has been much appreciated by consumers. Chinese manufactures have also been developing their capabilities to produce in-house chip designs which have allowed them to create the world’s strongest and fastest supercomputer. In fact, China has held the title for the last five years while the US only clocked in at 4th place. US Department of Energy has reclaimed the title as on June 8th, they have announced they have completed construction of the fastest supercomputer in the world, Summit.
 
The capabilities of the machine is truly staggering as it can perform 200 petaflops, or 20 quadrillion calculations per second. This is over twice as fast as previous record holder, China’s Sunway TaihuLight Supercomputer which could perform 93.01 petaflops. Researchers at the Oak Ridge National Laboratory have been testing the system and have succeeded in running the world’s first exascale scientific calculation. Using comparative genomics calculations used for research in in bioenergy and human health, the machine has run a 1.88 exaops calculation.
 
The machine was built in a joint effort by IBM and Nvidia, which implements some of their most state of the art technology. The 4,608 servers each has two of IBM’s 22 core Power9 processors. These Power9 processor’s acceleration capabilities in GPU, FPGAs, DSPs and ASICs and memory coherence features give it the prime advantage in this architecture. Each server contain six Nvidia’s Tesla V100 GPUs which provide better performance than single-function ASICs.
 
Nvidia and IBM’s collaborated to integrate their chips together with NVLink to provide 300 gigabytes per second or ten times faster than any PCI express lanes. These components are interconnected with dual-rail Mellanox EDR 100Gb/s InfiniBand, giving it high bandwidth pathways. Parech Kharya, a director at Nvidia stated, “It’s the world’s largest GPU-accelerated supercomputer.”
 
The Summit supercomputer will be open to projects later this year. Some of the projects scheduled to be run on summit are:

  • An Astrophysics project that simulates exploding stars known as supernovas, to investigate how heavy elements such as gold and iron are seeded in the universe.
  • Cancer research that can automatically sort and analyze health data to see relationships between disease causing variables such as genes, environment and biological markers.
  • Systems biology research to analyze human proteins and cellular systems to identify patterns in their function and evolution.
 
As impressive as these achievements are, Summit is still another stepping stone in the overall Exascale computing race. Exascale computing is a system that is capable of an exaFLOPS or 10^18 calculations per second. The US Department of Energy along with China and Japan have all developed an initiative to develop a supercomputer of their own with this capability. We should anticipate more iterations and breakthroughs in this field in the coming years.

Original Article: https://www.ornl.gov/news/ornl-launches-summit-supercomputer