NVIDIA Expands HPC Footprint at SC17, as Study Shows GPU Acceleration Key for Scientific Computing

by Ian Buck

Top 15 and 70% of top 50 HPC applications now GPU accelerated; record number of GPU-accelerated systems join TOP500 list.

There’s no more vivid display of NVIDIA’s growing momentum in high performance computing than in the hallways of this week’s SC17 supercomputing show.

In a pre-show talk, NVIDIA CEO and founder Jensen Huang noted that every major computer maker and cloud service has turned to the NVIDIA Volta architecture to accelerate data-intensive workloads. The company had its best showing ever in the newly released version of the TOP500 list of the world’s fastest supercomputers.

And a new report by analyst firm Interesect360 Research describes NVIDIA as critical to the future of scientific computing, noting that all of the top 15 and 70 percent of the top 50 HPC applications are now GPU accelerated.

“GPU computing has reached a tipping point in the HPC market that will encourage continued increases in application optimization,” wrote Addison Snell and Laura Segervall of Intersect360.

Most New TOP500 Systems for NVIDIA

The latest version of the twice-yearly TOP500 supercomputer list showed that NVIDIA added a record 34 new GPU-accelerated systems — bringing its total on the list to 87. The company also increased its total petaflops on the list by 28 percent, and it captured 14 of the top 20 most energy-efficient supercomputers on the Green500 list.

The Changing HPC App Ecosystem

But the changing face of high performance computing was set out particularly clearly by the Intersect360 report, which details the growing importance of GPU acceleration.

It notes that GPUs currently accelerate:

  • Top 15 chemistry applications
  • Top two fluid dynamics analysis applications
  • Seven of the top eight structural analysis applications
  • All the top visualization analysis applications
  • All the most popular biosciences applications

The world’s top 15 HPC applications, all GPU accelerated, include GROMACS, ANSYS Fluent, Gaussian, VASP, NAMD, Simulia Abaqus, WRF, OpenFOAM, ANSYS, LS-DYNA, BLAST, LAMMPS, AMBER, Quantum Espresso and GAMESS.

The report also emphasizes the dynamic nature of the application ecosystem as a defining characteristic of HPC. Indeed, the most significant new trend is the coming together of AI and HPC. This year marked the first appearance of the GPU-accelerated deep learning framework, TensorFlow, on the list.

GPU-accelerated applications aren’t limited to the scientific community — the increasing crossover between HPC and business intelligence is demonstrated by SAP and Oracle applications joining the top 50 list for the first time with AI-optimized applications for enterprises.

“Today, one of the biggest market dynamics is the advent of AI,” according to Intersect360. “Many organizations are looking to deep learning techniques to bring AI advancements to their products, services, or operations. These algorithms often rely on GPUs, to the extent that AI has become a major growth driver for NVIDIA.”