The AMD-powered Frontier supercomputer will help power the Oak Ridge National Laboratory.
Our friends over at AMD have been making big strides in the field of supercomputing for some time. Just now the Frontier supercomputer—one of several machines powered by the red team’s custom-built components—has gained the top spot in not only the Top500, and Green500 lists, but also the HPL-AI performance list. That’s an immense feat for the company, and speaks to the raw power of the silicon AMD is producing nowadays.
The supercomputer is powered by AMD’s 3rd Gen, AI Optimized, 64-core, 7A53s EPYC processors, and four purpose built Radeon Instinct MI250x Accelerator GPUs.
Just to get a handle on the kind of power we’re talking here, AMD’s silicon provides 1.1 exaflops of compute power, which at peak performance equals 1.1 quintillion calculations per second. For comparison, an Nvidia GeForce RTX 3090-powered machine only offers around 35 (and a bit) teraflops.
All that makes the Frontier supercomputer the the world’s fastest right now, and officially the first to break the exascale barrier.
“The simple way to think about it, is that Frontier will give us a boost of acceleration by a factor of eight,” an official video explains. In other words we’ll be able to “tackle problems that are eight times more complex” in the same timeframe it takes to do those calculations now.
Aside from this incredible achievement, Frontier has also claimed silver as the second-most efficient supercomputer in the Green 500 list. In fact, this thing can deliver 52.23 gigaflops per watt of power-efficiency from a single cabinet, which is a good 11.33 gigaflops per watt over the only non-AMD machine in the top 5.
In the top spot is the Frontier TDS, also powered by AMD optimized 3rd gen EPYC processors, and AMD Instinct MI250X GPUs.
The red team’s notes that the next steps for Frontier involve “continued testing and validation of the system, which remains on track for final acceptance and early science access later in 2022.” The machine will then be open for “full science” at the beginning of 2023.
Just imagine all the good we can do in the fields of science and medicine with computing power (and efficiency) like that behind us.