Changing Times: How the World’s TOP500 Supercomputers Don’t Just Have to Be Fast, But Smart

changing-times:-how-the-world’s-top500-supercomputers-don’t-just-have-to-be-fast,-but-smart

The world’s quickest supercomputers aren’t just speedier than at any time. They’re smarter and assistance a larger variety of workloads, too.

Just about 70 percent of the devices, which include eight of the major 10, on the most recent Major500 checklist of the world’s speediest supercomputers, unveiled today at SC20, are driven by NVIDIA technologies.

In addition, four of the nominations for the Gordon Bell Prize, supercomputing’s most prestigious award — to be named this 7 days at SC20 — use AI to generate their discoveries.

The popular thread: our finish-to-conclusion HGX AI supercomputing system, which accelerates scientific computing, information analytics and AI workloads. It is a story that begins with a good chip and incredibly rapidly, sensible networking, but eventually is all about NVIDIA’s globally adopted data-centre-scale platform for doing great science.

The change to incorporating AI into HPC, and a platform that extends further than standard supercomputing facilities, signifies a considerable improve in a area that, given that Seymour Cray’s CDC 6600 was introduced in 1964, has focused on harnessing ever larger sized, more potent equipment for compute-intensive simulation and modeling.

The newest Top500 checklist is about far more than large-efficiency Linpack effects:

  • Velocity information: Calculated by the common benchmark of supercomputing general performance — the speed it requires to do operations in a double-precision floating-issue structure termed FP64 — NVIDIA technologies accelerate the world’s swiftest clusters, powering eight of the top 10 machines. This includes the No. 5 procedure — NVIDIA’s very own Selene supercomputer, the world’s most powerful commercial method — as properly as new additions like JUWELS (Forschungszentrum Jülich) at No. seven and Dammam-seven (Saudi Aramco) at No. 10.
  • “Smarts” records: When calculated by HPL-AI, the mixed-precision regular that is the benchmark for AI effectiveness, NVIDIA-powered devices captured top rated spots on the listing with Oak Ridge National Lab’s Summit supercomputer at .55 exaflops and NVIDIA Selene at .25 exaflops.
  • Inexperienced information: The NVIDIA DGX SuperPOD system captured the prime location on the Green500 record of most successful supercomputers, obtaining a new environment file in electricity effectiveness of 26.two gigaflops for each watt. General, NVIDIA-powered equipment captured 25 of the major 30 places on the checklist.

The Period of AI Supercomputing  Is in Higher Equipment

Probably the most amazing achievement: We have surpassed exascale computing nicely in advance of schedule.

In Oct, Italy’s CINECA supercomputing middle unveiled ideas to construct Leonardo, the world’s most highly effective AI supercomputer with an expected 10 exaflops of AI overall performance. It is joined by a wave of new EuropHPC AI programs in the Czech Republic, Luxembourg and Slovenia. Extra are coming not only in Europe but also throughout Asia and North The usa.

That is mainly because fashionable AI harnesses the incredible parallel processing energy of NVIDIA GPUs, NVIDIA CUDA-X libraries and NVIDIA Mellanox InfiniBand — the world’s only intelligent, fully accelerated in-network computing system — to pour extensive portions of info into advanced neural networks, developing advanced types of the earth all-around us. This lets scientists tackle a lot far more formidable jobs than would in any other case be doable.

Get the example of the group from Lawrence Berkeley National Laboratory’s Computational Investigate Division, a person of this year’s Gordon Bell Prize nominees. Many thanks to AI, the group was equipped to raise the scale of their molecular dynamics simulation by at the very least 100x compared to the most significant program simulated by former nominees.

It’s About Advancing Science

Of study course, it is not just how rapidly your technique is, but what you do with it in the authentic entire world that counts.

That is why you are going to come across the new breed of AI-driven being thrown into the entrance-line of the fight versus COVID.

A few of the 4 of the nominees for a particular Gordon Bell award concentrated tackling the COVID-19 pandemic depend on NVIDIA AI.

On the Lawrence Livermore Nationwide Laboratory’s Sierra supercomputer — No. three on the Best500 record — a group trained an AI ready to determine new drug candidates on one.six billion compounds in just 23 minutes.

On Oak Ridge’s Summit supercomputer — No. 2 on the Best500 checklist — a further team harnessed 27,612 NVIDIA GPUs to examination 19,028 probable drug compounds on two crucial SARS-CoV-2 protein constructions each and every 2nd.

Another staff used Summit to generate an AI-pushed workflow to model how the SARS-CoV-two spike protein, the primary viral an infection machinery, assaults the human ACE2 receptor.

Many thanks to the developing ubiquity of the scalable NVIDIA HGX AI supercomputing platform — which incorporates all the things from processors to networking and software package — scientists can run their workloads in the hyperscale information centers of cloud computing organizations, as properly as in supercomputers.

It is a unified system, enabling the fusion of high-efficiency computing, information analytics and AI workloads. With two.three million developers and supporting above 1,800 accelerated applications, all AI frameworks and common details analytics frameworks such as DASK and Spark, the system permits scientists and researchers to be promptly productive on GPU-powered x86, Arm and Power units.

In addition, the NVIDIA NGC catalog offers general performance-optimized containers for the most up-to-date versions of HPC and AI applications. So scientists and scientists can deploy quickly and continue to be centered on advancing their science.

Study a lot more in the stay NVIDIA SC20 Particular Handle at three p.m. PT currently.

Leave a comment

Your email address will not be published.


*