Recently, NVIDIA announced its super computing systems based on DGX A100 nodes. Now, the company is using dozens of these nodes in a super machine called Selene, which is none other than the seventh most powerful supercomputer in the world, which uses AMD EPYC processors, intended for servers.

Supercomputer Selene, with 27.58 petaflops

It seems a bit confusing, but NVIDIA, AMD’s archrival in the graphics chip segment, bet on its competitor’s technology to create one of the largest supercomputers in the world, for its own use.

Each DGX A100 node consists of two AMD EPYC 7742 processors – each with 64 cores and 128 threads – and eight NVIDIA Ampere A100 GPUs. The Selene supercomputer uses 140 DGX A100 nodes, adding 280 EPYC processors and 1,120 A100 GPUs, totaling 277,760 processing cores, equivalent to 27.58 computational power petaflops. Completing the configuration, we still have 560 thousand GB of RAM and the interconnection technology Mellanox HDR Infiniband, from NVIDIA, which makes the connection between all this hardware.

AMD returns to the Top 10 of supercomputers

After about a year, AMD is returning to the top 10 on the list of the world’s largest supercomputers. Before Selene, another supercomputer based on AMD processors was the Titan, equipped with Opteron 6274 CPUs and NVIDIA K20x graphics chips.

The Titan became the most powerful supercomputer in the world in 2012 and is currently in 12th place in the top 500, with 17.6 petaflops and consumption of 8.2 MegaWatts.

See Also
Nvidia reschedules GTC 2020 and teaser announces Ampere architecture

Selene, number seven in the ranking, has 27.58 petaflops and a consumption of 1.3 MegaWatts, which means that it surpasses the Titan with 57% more performance, for only 16% of its consumption, being up to 10 times more efficient.

Currently, of the 500 largest supercomputers in the world, eight are based on AMD EPYC Rome processors, making it the 10th most popular chip among these machines.

In the near future, the US Department of Energy intends to break the barrier of exaflops with two supercomputers based on AMD EPYC chips: the Frontier, with 1.5 exaflops, and El Capitan, with 2 exaflops. They will be launched, respectively, in 2021 and 2023, and will consume between 30 and 40 MW.

LEAVE A REPLY

Please enter your comment!
Please enter your name here