That Was Fast: GPUs Now Accelerate Almost 600 HPC Apps

by Paresh Kharya

Just over 10 years ago accelerated applications didn’t exist. Today, almost 600 are accelerated by NVIDIA GPUs. The reason: GPU acceleration works. And that’s why it’s been put to work on the hardest computing jobs on earth.

These are apps that get work done in physics, bioscience, molecular dynamics, chemistry and weather forecasting. The world’s 15 most popular HPC applications are all GPU accelerated. In the last year, we’ve added more than 100 applications to our NVIDIA GPU Applications Catalog. More are coming.

A report by Intersect 360 research identified the key applications running in the data center. All the top 15 apps were GPU accelerated. It’s a murderer’s row of hard-core science apps. They include:

  • GROMACS (Chemistry) – Molecular dynamics application for simulating Newtonian equations of motion for systems with hundreds to millions of particles.
  • ANSYS (Fluid Dynamics Analysis) – Simulates the interaction of liquids and gases with surfaces.
  • Gaussian (Chemistry) – Predicts energies, molecular structures and vibrational frequencies of molecular systems.
  • VASP (Chemistry) – Performing ab-initio quantum-mechanical molecular dynamics simulations.
  • NAMD (Chemistry) – High-performance simulation of large biomolecular systems.
  • Simulia Abaqus (Structural Analysis) – Simulation and analysis of structural mechanics.
  • WRF (Weather/Environment Modeling) – Numerical weather prediction system designed for both atmospheric research and operational forecasting applications.
  • OpenFOAM (Fluid Dynamics Analysis) – Solver library for general-purpose CFD software
  • ANSYS (Structural Analysis) – Models 3D full-wave electromagnetic fields in high-frequency and high-speed electronic components.
  • LS-DYNA (Structural Analysis) – Simulation and analysis tool for structural mechanics.
  • BLAST (Bioscience) – One of the most widely used bioinformatics tools.
  • LAMMPS (Chemistry) – A classical molecular dynamics package.
  • Amber (Chemistry) – A molecular dynamics application developed for the simulation of biomolecular systems.
  • Quantum Espresso (Chemistry) – An integrated suite of computer codes for electronic structure calculations and materials modeling at the nanoscale.
  • GAMESS (Chemistry) – Computational chemistry suite used to simulate atomic and molecular electronic structure.

These tools don’t get incremental performance gains. GPU acceleration changes the economics of the data center. Servers with NVIDIA GPUs typically speed up the application performance by 10x or more.

And since the application performance does not scale linearly with the number of CPU servers, each GPU-accelerated server provides the performance of even more CPU servers than what just the speed-ups would imply. So you can meet the growing demand for computing — and save money.

Not bad for 10 years’ worth of work.

Predicting the Weather

Weather prediction looks hard. And it might be even harder than it looks. No surprise, then, that weather prediction is a big piece of HPC. Important, too. Reliable weather forecasts save lives. They also drive economic decisions in aviation, energy and utilities, insurance, retail and other industries.

But weather prediction requires massive computing resources. Two reasons: geometric scale (especially for global weather predictions), and the enormous number of variables that describe the state of the atmosphere.

Today weather prediction is limited by the amount of computing and application performance available. So today’s models are limited to low-resolution simulations, such as 12-km resolution.

But that leaves out important details, such as the impact of clouds, which play an important role in weather patterns by reflecting solar radiation. Going to 1-km cloud-resolving resolution can improve forecasting. But it requires 1,700x more application performance.

GPU acceleration can heft weather forecasts over that gap.

Accelerating Aerodynamics Simulations with FUN3D

SLS Block 1B booster separation flowfield simulated using NASA’s FUN3D code. Image courtesy of Jamie Meeroff, Henry Lee, NASA/Ames. FUN3D GPU development: E. Nielsen and A. Walden, NASA LaRC and FUN3D Development Team

Aircraft, spacecraft, automobiles. If it goes fast, large-scale aerodynamic simulations can help it go faster — and more efficiently.

The NASA Langley Research Center develops FUN3D computational fluid dynamics software to simulate fluid flow for a broad range of aerodynamics applications. This application consumes more cycles at NASA’s Pleiades supercomputer than any other. And GPU acceleration enables a server with six NVIDIA V100 Tensor Core GPUs to provide 30x higher performance than a dual-socket CPU server while running these simulations.

Takeaway: the performance on GPUs scales very well to enable efficient computation of the largest and the most complex simulations. NASA has shown that a thousand GPU servers on Summit supercomputer can do the work of over a million CPU cores. And for a fraction of the energy costs.

Performance That Keeps Growing

We have deep expertise in all accelerated computing domains. Combined with an ecosystem of more than 1 million developers, this results in a platform that’s constantly improving. This provides higher application performance over time on the same GPU-accelerated servers.

For instance, on a basket of 11 HPC applications, a server with 4 NVIDIA Tesla P100 GPUs now runs 2x faster compared to its performance from two years ago. Pair improvements in the software stack and GPU architecture advancements and you get even bigger performance gains.

With a single platform, you can now accelerate applications across a variety of HPC domains — scientific computing, industrial simulations, deep learning and machine learning. The harder the job, the bigger the payoff. So go ahead and accomplish wonders — or get your work done fast enough to see your kids — with GPU-accelerated applications.

To see the full list of GPU-accelerated applications, check out the NVIDIA GPU Applications Catalog.