The newest replace to the listing of the world’s 500 strongest supercomputers noticed just one new entry within the prime ten, confirming that though these gadgets are nonetheless enhancing their efficiency, the tempo of innovation is slowing down.
Perlmutter, a US-based supercomputer positioned on the Division of Power’s Lawrence Berkeley Laboratory, entered the June version of the Top500 listing in fifth place, bumping Nvidia’s Selene machine to sixth place.
At 64.6 petaflops, Perlmutter is essentially the most notable change to the listing; it additionally fared properly within the Green500 listing, which focuses on the vitality effectivity of supercomputers, claiming quantity six spot due to an influence effectivity of 25.55 gigaflops per watt.
In whole, there have been solely 58 new entries within the Top500 listing. This isn’t far off the record-low of 44 new entrants registered final November, and signifies a slow-down in comparison with earlier years. In 2007, for instance, 300 new gadgets made it to the rating.
That is largely blamed on the influence of the COVID-19 disaster, which drove funding in business high-performance computing programs to an all-time low.
The authors of the listing admitted that the newest version “noticed little change”, with Japan’s Fugaku supercomputer nonetheless retaining a powerful lead within the number-one spot.
Fugaku got here on-line final March and boasts a record-breaking 442 petaflops – that means that it’s thrice extra performant than the closest competitor, IBM’s Summit, which is hosted on the Oak Ridge Nationwide Laboratory and claims 148.8 petaflops.
A product of a partnership between Riken and Fujitsu, Fugaku makes use of Fujitsu’s customized ARM processor, which has enabled the machine to achieve peak efficiency of over an exaflop. “Such an achievement has triggered some to introduce this machine as the primary ‘exascale’ supercomputer,” mentioned the Top500 authors.
Exascale computations can carry out a quintillion calculations every second, and are anticipated to considerably velocity up purposes starting from precision drugs to local weather simulation. China, the US and the EU have all unveiled aggressive targets to realize exascale computing programs within the subsequent few years.
In Japan’s Analysis Group for Data Science and Expertise (RIST), Fugaku is ready to advance a particular 74 scientific initiatives, a lot of which intention to speed up analysis to fight the COVID-19 pandemic, similar to early detection of illness or high-speed and high-precision drug discovery simulations.
Competing behind Fugaku, the remainder of the listing’s prime ten remained largely the identical, with IBM’s Sierra in third place following the corporate’s Summit supercomputer.
China noticed a big drop within the variety of gadgets it could actually declare on the listing. From 212 Chinese language machines that includes on the earlier iteration, the nation now solely accounts for 186 supercomputers on the Top500 listing.
“There hasn’t been a lot definitive proof of why that is taking place, however it’s definitely one thing to notice,” mentioned the listing’s authors.
Regardless of the drop, China nonetheless dominates the Top500: the following greatest competitor, the USA, lags behind with 123 programs on the listing. This has led some researchers to conclude that the hole between the 2 nations is quickly closing.
It stays, nonetheless, that the efficiency of Chinese language supercomputers is way outstripped by different programs. The USA has an mixture efficiency of 856.8 petaflops per second, whereas China’s machines produce on common 445.3 petaflops per second.
In one other noteworthy improvement, the Top500’s authors highlighted the marked enhance in the usage of AMD processors, particularly among the many few new entries. The corporate’s EPYC processors energy half of the 58 new contestants, and three of the gadgets included within the prime ten – Perlmutter, Selene and Germany’s Juwels Booster Module.
In comparison with the identical time final yr, AMD famous that EYMC programs additionally energy 5 instances extra supercomputers all through the complete listing.
“We’re dedicated to enabling the efficiency and capabilities wanted to advance scientific discoveries, break the exascale barrier, and proceed driving innovation,” mentioned Forrest Morrod, senior vice chairman of knowledge middle and embedded programs group at AMD.
By way of system interconnects, the authors noticed little change, with ethernet utilized in round half of the programs. A 3rd of the supercomputers leveraged Infiniband and fewer than a tenth relied on OmniPath. Customized interconnects accounted for 37 programs.