The podium at the top of supercomputers has changed

The dominance of the top machine of the top500 list is unshakable, but life is very much going on behind it.

A top500.org updated the ranking of supercomputers in the world, which finally brought excitement in terms of the podium. The first place is still held by the Frontier, and its specifications have not changed compared to the middle of the year. Another large Exascale project from the USA, known as Aurora, immediately jumped into second place. It is powered by Intel Xeon Max 9470 processors and accelerators codenamed Ponte Vecchio. Unfortunately, for the time being, the machine is operating at roughly half steam, as it was not possible to build it in time. The maximum theoretical performance of the current configuration reaches 1059.3 PFLOPS, of which it achieves 585.3 PFLOPS in the Linpack test, and this is coupled with a consumption of 24.6 megawatts. From these data, you can see that the Aurora is a long way from the energy efficiency of the Frontier.

The bottom rung of the podium was taken by Eagle, which is part of Microsoft Azure. It uses Intel Xeon Platinum 8480C processors and NVIDIA H100 accelerators. The entire system is capable of 561.2 PFLOPS of the theoretical 846.8 PFLOPS in the Linpack test, but its consumption is not public, so nothing can be known about its efficiency.

Although they slipped off the podium, the former second, third, fourth and fifth placed Supercomputer Fugaku, LUMI, Leonardo and Summit foursome now occupy fourth, fifth, sixth and seventh place. It is important to mention that LUMI has expanded somewhat, so that its maximum theoretical performance has reached 531.5 PFLOPS, of which it achieves 379.7 PFLOPS in the Linpack test, with a consumption of 7.1 megawatts.

Eighth place was rented by MareNostrum 5 ACC, which was built in Spain and uses a Xeon Platinum 8460Y+ and NVIDIA H100 configuration. This is capable of 138.2 PFLOPS in the Linpack test out of a theoretical 265.57 PFLOPS, while requiring roughly 2.56 megawatts.

The ninth position was taken by NVIDIA’s own system, the Eos NVIDIA DGX SuperPOD, paired with Xeon Platinum 8480C and NVIDIA H100. Its consumption is also secret, but it achieves 121.4 PFLOPS of the theoretical 188.65 PFLOPS in the Linpack test.

With the above changes, the sixth Sierra slipped back to tenth position in the middle of the year, but its capabilities did not change.