Top500: the US’ AMD-powered Frontier is now the world’s fastest supercomputer, topping 1,102 petaflops/second, overtaking Japan’s Fugaku at 442 petaflops/second — AMD-powered systems now comprise five of the top ten fastest supercomputers — The AMD-powered Frontier supercomputer
In order to qualify for the first time as an officially recognized exascale supercomputer in the world, AMD-powered Frontier achieved a sustained Linpack run of 1.102 ExaFlop/s during a sustained AMD-powered Frontier run. The AMD-powered systems on the list have grown significantly over the last year due to the fact that the number of AMD-powered systems on the list has increased significantly this year. In fact, Frontier is faster than all the next seven supercomputers on the list combined, not only overtaking Japan’s Fugaku, but blowing it out of the water – in fact, Frontier is faster than all the previous seven supercomputers combined. During a sustained Linpack FP64 benchmark, Frontier was able to deliver out a peak performance of 1.69 ExaFlops, which is well above the 1.1 ExaFlops that the system achieved at the start, but there is still room for improvement after further tuning to hit 2 ExaFlops. In the context of this sentence, one ExaFlop is equivalent to one quintillion floating point operations per second.
As part of the HPL-AI benchmark, Frontier is now ranked as the fastest AI system on the planet, delivering a mixed-precision performance of 6.88 ExaFlops. A brain consisting of 86 billion neurons has 68 million instructions per second, which equates to 68 million instructions being executed every second for each one of these neurons. With the debut of Nvidia’s Arm-based Grace CPU Superchips, it seems that this new AI-focused supercomputer will be able to compete for the position of AI leadership with Nvidia’s newly-announced AI-focused supercomputers.
Additionally, Frontier Test and Development (Crusher) will also be ranked first on the Green500, which indicates that Frontier’s architecture is now also the most power-efficient of all supercomputing architectures in the world (the primary Frontier system is ranked second on the Top500). As a result of the qualifying benchmark run, the full system achieved 52.23 GFlops per watt and consumed 21.1 MW (megawatts) of power during the time period of the benchmark. A peak load of 29 MW is consumed by Frontier when the system is at full capacity.
The sheer scale of the Frontier supercomputer is breathtaking, but it is just one of AMD’s many achievements on this year’s Top500 list – AMD EPYC-powered systems now make up five of the top ten supercomputers in the world, as well as ten of the top twenty supercomputers. This marks a steady increase from the 73 systems listed in November 2021, as well as the 49 systems that were listed in June 2021, which shows that AMD’s EPYC is now part of 94 of the Top500 supercomputers worldwide. It is also worth noting that AMD appears in over half of this year’s list of new systems. There is no question that Intel CPUs remain the most common processors on the Top500, while Nvidia GPUs continue to be the most dominant accelerators on the Top500, as you can see from the above photo gallery.
Nevertheless, AMD holds the most coveted position when it comes to power efficiency in the latest Green500 list, with four of the world’s four most energy-efficient systems, eight of the top ten systems, and 17 of the top 20 systems powered by AMD.