Nvidia units 16 new efficiency data in newest MLPerf AI benchmarks – SiliconANGLE

Nvidia Corp. mentioned at this time the most recent MLPerf benchmark test results show that its newest platforms ship the world’s quickest synthetic intelligence coaching efficiency amongst all commercially out there methods.

Nvidia’s new A100 graphics processing unit and its DGX SuperPOD system (pictured), which is a large cluster of A100 GPUs linked with its HDR InfiniBand know-how, each set eight new efficiency data for commercially out there methods within the third annual MLPerf benchmark exams, making a complete of 16 new data.

The exams have been organized by MLPerf, which is an business benchmarking group that was arrange in May 2018 and is backed by corporations that embody Amazon.com Inc., Baidu Inc., Facebook Inc., Google LLC, Intel Corp. and Microsoft Corp., in addition to Harvard and Stanford universities.

The outcomes mark a critical enchancment for Nvidia’s {hardware}, which beforehand set six data within the first MLPerf coaching benchmarks in December 2018, and eight records in July 2019.

Nvidia’s A100 GPU, launched in May, is the premise of the corporate’s third-generation DGX system, which is used to energy supercomputers similar to the University of Florida’s HiPerGator. The A100 chip can be available as a service on Google Cloud, focused at corporations that want the very best potential efficiency for information analytics, scientific computing, genomics, edge video analytics and 5G companies workloads.

Nvidia mentioned MLPerf’s newest benchmarks present that its DGX A100 system delivers a four-times efficiency enchancment over its authentic DGX system that was based mostly on its older V100 GPUs. However, it mentioned the exams present that the older system, generally known as DGX-1, is now twice as quick due to some new software program optimizations.

The newest MLPerf benchmarks included two model new exams in addition to one “substantially revised” check and Nvidia mentioned its {hardware} “excelled” in all of them. For instance, the A100 chip and the DGX SuperPOD system achieved the very best efficiency within the new suggestion methods check, which is an more and more common workload for AI methods.

Nvidia’s {hardware} additionally achieved prime scores within the Natural Language Processing class utilizing the Bi-directional Encoder Representation from Transformers, or BERT, neural community mannequin. It additionally set new data within the reinforcement studying check that makes use of Mini-go with a full-size 19×19 Go board. According to Nvidia, this “was the most complex test in this round involving diverse operations from game play to training.”


Nvidia mentioned the DGX SuperPOD, which options greater than 2,000 Nvidia A100 GPUs, swept each MLPerf benchmark class for at-scale efficiency amongst commercially out there merchandise.

Constellation Research Inc. analyst Holger Mueller advised SiliconANGLE that MLPerf’s benchmarks are necessary as we’re in the course of the race to AI, and that enterprises understand extra automation is healthier than much less.

“Covid-19 only increases the urgency, so companies are looking to platform vendors to help them with their next generation application AI loads,” Mueller mentioned. “Today it is Nvidia’s turn, setting new records on a number of MLPerf standards. What is remarkable is that Nvidia has been able to improve its performance by four times over the last one and half years. That kind of performance increase is what is needed play in the medal ranks of AI benchmarks.”

Nvidia wasn’t alone in setting new data. Google additionally participated within the MLPerf benchmarks with a few of its new {hardware}, and says they show it has constructed the world’s quickest machine studying coaching supercomputer after setting six efficiency data.


Google’s newest ML coaching supercomputer, based mostly on its latest Tensor Processing Unit, is 4 instances the dimensions of the Cloud TPU V3 Pod that set three data within the earlier benchmarks final yr. Made up of 4,096 TPU V3 chips and a whole bunch of CPU hosted machines, the system delivers a peak efficiency of greater than 430 petaflops, Google mentioned.

Photo: Nvidia

Since you’re right here …

Show your help for our mission with our one-click subscription to our YouTube channel (under). The extra subscribers we’ve, the extra YouTube will counsel related enterprise and rising know-how content material to you. Thanks!

Support our mission:    >>>>>>  SUBSCRIBE NOW >>>>>>  to our YouTube channel.

… We’d additionally wish to inform you about our mission and how one can assist us fulfill it. SiliconANGLE Media Inc.’s enterprise mannequin is predicated on the intrinsic worth of the content material, not promoting. Unlike many on-line publications, we don’t have a paywall or run banner promoting, as a result of we wish to hold our journalism open, with out affect or the necessity to chase visitors.The journalism, reporting and commentary on SiliconANGLE — together with dwell, unscripted video from our Silicon Valley studio and globe-trotting video groups at theCUBE — take a number of arduous work, money and time. Keeping the standard excessive requires the help of sponsors who’re aligned with our imaginative and prescient of ad-free journalism content material.

If you just like the reporting, video interviews and different ad-free content material right here, please take a second to take a look at a pattern of the video content material supported by our sponsors, tweet your support, and hold coming again to SiliconANGLE.


Please enter your comment!
Please enter your name here