Supercomputing modified within the late 90s. It was once that the largest and quickest computer systems have been all utterly customized jobs, with proprietary every little thing and unique {hardware} working esoteric architectures. In the late-90s, somebody discovered that you would obtain supercomputer efficiency on some duties just by wiring collectively a big cluster of off-the-shelf PCs. Ever since then, many of the world’s quickest supercomputers have primarily been overgrown commodity servers, generally with customized interlinks or different peculiarities.

So it goes with the most recent machine to high the Top500 record. For these unfamiliar, Top500 is an inventory of the highest five-hundred quickest supercomputers on this planet—or at the least, people who submit their outcomes to be ranked, anyway. The most-recent Top500 record simply got here out, and resting atop the pile is the brand-new Frontier system, owned and operated by the USA’s Oak Ridge National Laboratory.

Frontier was constructed by HPE Cray, and makes use of an unbelievable variety of third-generation AMD EPYC processors outfitted with AMD Instinct MI250X accelerators. On the usual HPL benchmark, it managed to crank out 1.1 EFLOPs, making it the very first machine on this planet to surpass the exaFLOP barrier. For perspective, 1.1 EFLOPs is 1.1×10^18 floating level operations per second, or across the efficiency of 36,000 GeForce RTX 3080 playing cards.
AMD’s Instinct compute accelerators are doing a lot of the heavy compute lifting.

Despite the large measurement of the machine, its unbelievable efficiency signifies that it’s truly not simply the quickest, but in addition essentially the most environment friendly: Frontier topped the Green500 record, too. Each cupboard of the machine delivers 62.68 GFLOPs of compute per watt consumed, though whenever you take into consideration all the networking required for the total system its effectivity drops to “just” 52 GFLOPS/watt—nonetheless excessive sufficient to take the highest spot.

HPE Cray’s new AMD-powered machine knocked Japan’s Fugaku system off the #1 spot. That machine is powered by ARM A64X processors, and it too has a peak theoretical efficiency over 1 EFLOP, nevertheless it hasn’t been capable of notice that efficiency on the Linpack benchmark that Top500 makes use of. It’s held the highest spot for 2 years with “just” 442 PFLOPs, or just below half of what Frontier is able to.

To illustrate how far forward Fugaku and Frontier are from the remainder of the record, a really related system to Frontier—additionally constructed by HPE Cray and utilizing the identical fundamental AMD elements—slotted in freshly at place #3. That system is named LUMI, and it lives in Finland, where it produces 152 PFLOPs for that nation’s IT Center for Science, generally known as CSC.

In truth, AMD proudly notes in its press launch that it lays declare to 94 methods out of the Top500 record, which does not sound that spectacular on the face of it, however that quantity has practically doubled since final 12 months. The firm additionally gloats that it dominates the Green500 record, holding the highest 4 positions, eight of the highest ten, and 17 of the highest 20.

A cynic would possibly level out that that is just because the AMD methods are additionally the latest methods, and that future machines utilizing Intel, NVIDIA, IBM, or ARM {hardware} would possibly equally leapfrog the pink crew’s gear. That’s probably sooner or later, although, and never a assured. As such, let’s let AMD rejoice a bit, as a result of they’ve earned it.