Official announcement! AMD will be responsible for building the world's strongest supercomputer Frontier: 150 billion times per second

AMD today announced that it has joined forces with Cray, based on EPYC Opteron processors (Zen 3 or Zen 4 architecture), Radeon Instinct accelerator cards, and ROCm open source software to create the highest performing supercomputer ever for the US Department of Energy Oak Ridge Lab. "Frontier", peak delivery of up to 1.5 exaflops per second (billions of times).

It is reported that the communication between the CPU and the GPU on each node is a customized Infinity Fabric interconnection technology, which can achieve high bandwidth and low latency.

AMD said that this super-calculation will be delivered in 2021, covering two basketball courts, which are said to be priced at 9 digits. Help ORNL researchers build, simulate, and improve understanding of weather science, subatomic structural science, genomics, physics, and other important scientific phenomena with unprecedented computing power and next-generation artificial intelligence.

According to last year's super-calculation of 500, the current super-performance is the Summit deployed in Oak Ridge, with 2.5 million cores, 2.8PB of memory, and a peak of 200PFLOPS (20 billion times); China's Shenwei Taihu Lake ( 90PFLOPS) and Tianhe 2A are ranked three or four.

In addition, in 2005, AMD created the first “Jaguar” for the US Department of Energy. In 2012, Titan was built.

Interestingly, in 2021, there is also a supercomputer "Aurora" to be delivered to the US Department of Energy Argonne laboratory, based on Intel Xeon processor, Xe graphics card, but performance is slightly inferior to Frontier, 1 EFLOPS (1 billion billion times).

Credit: mydrivers


Post a Comment