New Gpu-Accelerated Supercomputers Alter The Ease Of Ability On The Top500

The 51st (25th anniversary) Top500 listing was released June 25, 2018 inwards Frankfurt Germany.
Here's some of Top500's thinking on the latest listing together with the province of the art.

From Top500, June 26:

For the showtime fourth dimension inwards history, most of the flops added to the TOP500 listing came from GPUs instead of CPUs. Is this the shape of things to come?

In the latest TOP500 rankings announced this week, 56 pct of the additional flops were a resultant of NVIDIA Tesla GPUs running inwards novel supercomputers – that according to the Nvidians, who bask keeping rails of such things. In this case, most of those additional flops came from iii transcend systems novel to the list: Summit, Sierra, together with the AI Bridging Cloud Infrastructure (ABCI).

Summit, the novel TOP500 champ, pushed the previous release 1 system, the 93-petaflop Sunway TaihuLight, into 2nd house amongst a Linpack score of 122.3 petaflops. Summit is powered past times IBM servers, each 1 equipped amongst 2 Power9 CPUs together with 6 V100 GPUs. According to NVIDIA, 95 pct of the Summit’s tiptop functioning (187.7 petaflops) is derived from the system’s 27,686 GPUs.

NVIDIA did a like calculation for the less powerful, together with somewhat less GPU-intense Sierra, which at in 1 lawsuit ranks every bit the 3rd fastest supercomputer inwards the footing at 71.6 Linpack petaflops. And, although real like to Summit, it has iv V100 GPUs inwards each dual-socked Power9 node, rather than six. However, the 17,280 GPUs inwards Sierra yet stand upwards for the lion’s percentage of that system’s flops.

Likewise for the novel ABCI machine inwards Japan, which is at in 1 lawsuit that country’s speediest supercomputer together with is ranked 5th inwards the world. Each of its servers pairs 2 Intel Xeon Gold CPUs amongst iv V100 GPUs. Its 4,352 V100s deliver the vast bulk of the system’s 19.9 Linpack petaflops.

As dramatic every bit that 56 pct release is for novel TOP500 flops, the reality is in all likelihood fifty-fifty to a greater extent than impressive.  According to Ian Buck, vice president of NVIDIA's Accelerated Computing concern unit, to a greater extent than than one-half the Tesla GPUs they sell into the HPC/AI/data analytics infinite are bought past times customers who never submit their systems for TOP500 consideration. Although many of these GPU-accelerated machines would qualify for a spot on the list, these detail customers either don’t assist well-nigh all the TOP500 fanfare or would rather non advertise their hardware-buying habits to their competitors.

It’s also worth mentioning that the Tensor Cores inwards the V100 GPUs, amongst their specialized 16-bit matrix math capability, endow these iii novel systems amongst to a greater extent than deep learning potential than whatever previous supercomputer. Summit lone boasts over iii tiptop exaflops of deep learning performance. Sierra’s functioning inwards this regard is to a greater extent than inwards the neighborhood of 2 tiptop exaflops, spell the ABCI release is to a greater extent than or less one-half an exaflop. Taken together, these iii supercomputers stand upwards for to a greater extent than deep learning capability than the other 497 systems on the TOP500 listing combined, at to the lowest degree from the perspective of theoretical performance.

The add-on of AI/machine learning/deep learning into the HPC application infinite is a relatively novel phenomenon, simply the V100 appears to hold upwards acting every bit a catalyst. “This year’s TOP500 listing represents a clear shift towards systems that back upwards both HPC together with AI computing,” noted TOP500 writer Jack Dongarra, Professor at University of Tennessee together with Oak Ridge National Lab....MORE

Working backward, hither is the Top500 press unloosen announcing the novel list, June 25:

US Regains TOP500 Crown amongst Summit Supercomputer, Sierra Grabs Number Three Spot 

FRANKFURT, Germany; BERKELEY, Calif.; together with KNOXVILLE, Tenn.—The TOP500 celebrates its 25^th anniversary amongst a major shakeup at the transcend of the list. For the showtime fourth dimension since Nov 2012, the U.S. claims the most powerful supercomputer inwards the world, leading a meaning turnover inwards which iv of the v transcend systems were either novel or substantially upgraded.

Summit, an IBM-built supercomputer at in 1 lawsuit running at the Department of Energy’s (DOE) Oak Ridge National Laboratory (ORNL), captured the release 1 spot amongst a functioning of 122.3 petaflops on High Performance Linpack (HPL), the benchmark used to rank the TOP500 list. Summit has 4,356 nodes, each 1 equipped amongst 2 22-core Power9 CPUs, together with 6 NVIDIA Tesla V100 GPUs. The nodes are linked together amongst a Mellanox dual-rail EDR InfiniBand network.

Sunway TaihuLight, a organisation developed past times China’s National Research Center of Parallel Computer Engineering & Technology (NRCPC) together with installed at the National Supercomputing Center inwards Wuxi, drops to release 2 afterward leading the listing for the past times 2 years. Its HPL grade of 93 petaflops has remained unchanged since it came online inwards June 2016.

Sierra, a novel organisation at the DOE’s Lawrence Livermore National Laboratory took the release iii spot, delivering 71.6 petaflops on HPL. Built past times IBM, Sierra’s architecture is quite like to that of Summit, amongst each of its 4,320 nodes powered past times 2 Power9 CPUs plus iv NVIDIA Tesla V100 GPUs together with using the same Mellanox EDR InfiniBand every bit the organisation interconnect.

Tianhe-2A, also known every bit Milky Way-2A, moved downward 2 notches into the release iv spot, despite receiving a major upgrade that replaced its five-year-old Xeon Phi accelerators amongst custom-built Matrix-2000 coprocessors. The novel hardware increased the system’s HPL functioning from 33.9 petaflops to 61.4 petaflops, spell bumping upwards its mightiness consumption past times less than iv percent. Tianhe-2A was developed past times China’s National University of Defense Technology (NUDT) together with is installed at the National Supercomputer Center inwards Guangzhou, China.

The novel AI Bridging Cloud Infrastructure (ABCI) is the fifth-ranked organisation on the list, amongst an HPL grade of 19.9 petaflops. The Fujitsu-built supercomputer is powered past times 20-core Xeon Gold processors along amongst NVIDIA Tesla V100 GPUs. It’s installed inwards Nihon at the National Institute of Advanced Industrial Science together with Technology (AIST).

Piz Daint (19.6 petaflops), Titan (17.6 petaflops), Sequoia (17.2 petaflops), Trinity (14.1 petaflops), together with Cori (14.0 petaflops) motility downward to the release 6 through 10 spots, respectively.

General highlights
Despite the ascendency of the U.S. at the transcend of the rankings, the province at in 1 lawsuit claims solely 124 systems on the list, a novel low. Just 6 months ago, the U.S. had 145 systems. Meanwhile, China improved its representation to 206 full systems, compared to 202 on the concluding list. However, thank yous mainly to Summit together with Sierra, the U.S. did acquire by to withdraw hold the atomic number 82 dorsum from China inwards the functioning category. Systems installed inwards the U.S. at in 1 lawsuit contribute 38.2 pct of the aggregate installed performance, amongst China inwards 2nd house amongst 29.1 percent. These numbers are a reversal compared to 6 months ago.
The adjacent most prominent countries are Japan, amongst 36 systems, the United Kingdom, amongst 22 systems, Federal Republic of Federal Republic of Germany amongst 21 systems, together with France, amongst eighteen systems. These numbers are nearly the same every bit they were on the previous list....MORE

And the list:

June 2018
The TOP500 celebrates its 25^th anniversary amongst a major shakeup at the transcend of the list. For the showtime fourth dimension since Nov 2012, the U.S. claims the most powerful supercomputer inwards the world, leading a meaning turnover inwards which iv of the v transcend systems were either novel or substantially upgraded...."NVIDIA Chief Scientist Bill Dally on How GPUs Ignited AI, together with Where His Team’s Headed Next" (NVDA)