AMD-Powered Frontier Supercomputer Breaks Exascale Barrier to Become World’s Fastest Supercomputer
Image: ORNL
The AMD-powered Frontier supercomputer became the world’s first officially recognized exaflops supercomputer, reaching speeds of 1,102 ExaFlop/s during Linpack’s long run. This is the first place in the recently published Top500 list of the world’s fastest supercomputers, as the number of AMD-based systems in the list has increased significantly this year. Frontier not only overtook the previous leader,
, but also beat it – in fact, Frontier is faster than the next seven supercomputers on the list combined. It is noteworthy that during the long testing of Linpack FP64, the Frontier system reached 1.1 ExaFlops, and the peak performance of the system is 1.69 ExaFlops, but after additional tuning, it can reach 2 ExaFlops. For reference, one ExaFlop is equal to one quintillion floating point operations per second.
Frontier is also the fastest AI system on the planet, delivering 6.88 ExaFlops mixed-precision performance in the HPL-AI benchmark. This equates to 68 million instructions per second for each of the brain’s 86 billion neurons, underscoring the enormous computing power. It looks like this system will compete for AI leadership with recently announced AI-focused supercomputers powered by Nvidia’s Arm-based Grace CPU superchips.
In addition, the Frontier Test and Development (Crusher) system also ranked first in the Green500, which means that the Frontier architecture is the most power-efficient supercomputing architecture in the world (the Frontier main system is ranked second in the Top500). During qualification testing, the complete system delivered 52.23 Gflops per watt while consuming 21.1 MW (megawatts) of electricity. At peak load, the Frontier system consumes 29 MW.
Image: ORNL
The scale of the Frontier supercomputer is impressive, but it’s just one of AMD’s many significant accomplishments on this year’s Top500 list – AMD EPYC-based systems now feature in five of the top ten supercomputers in the world and ten of the top twenty. In fact, AMD’s EPYC is now used in 94 of the world’s Top500 supercomputers, which represents a strong increase from 73 systems listed in November 2021 and 49 systems listed in June 2021. In addition, AMD is used in more than half of the new computer systems included in the list this year. As you can see from the album above, Intel processors still make up the majority of systems in the Top500, and Nvidia GPUs also continue to be the dominant factor in overclocking.
However, in terms of energy efficiency, AMD leads the latest Green500 list with four of the most efficient systems in the world, as well as eight places in the top ten and 17 places in the top twenty.
The Frontier supercomputer is built by HPE and installed at the Department of Energy (DOE) Oak Ridge National Laboratory (ORNL) in Tennessee. The system includes 9,408 compute nodes, each equipped with one 64-core AMD “Trento” processor paired with 512GB of DDR4 memory and four AMD Radeon Instinct MI250X GPUs. These nodes are spread across 74 HPE Cray EX cabinets, each weighing 8,000 pounds. Overall, the system has 602,112 processor cores associated with 4.6 petabytes of DDR4 memory.
In addition, 37,888 AMD MI250X GPUs have 8,138,240 cores and 4.6 petabytes of HBM memory (128 GB per GPU). The CPU and GPU are interconnected using the HPE Cray Slingshot-11 Ethernet-based network fabric. The entire system uses direct water cooling to reduce heat generation, with 6,000 gallons of water being moved through the system by 350-horsepower pumps – these pumps can fill an Olympic-sized pool in 30 minutes. The water in the system is 85 degrees, which contributes to energy efficiency because the system does not use chillers to reduce the temperature of the water.
The entire system is connected to an incredibly powerful storage subsystem with a capacity of 700 petabytes, a throughput of 75 TB / s and a throughput of 15 billion IOPS. The metadata tier is spread across 480 NVMe SSDs providing 10 PB of total capacity, while 5400 NVMe SSDs provide 11.5 PB of capacity for the primary high-speed storage tier. At the same time, 47,700 PMR hard drives provide a capacity of 679 PB.
Image: ORNL
Assembling the Frontier was a daunting task, as 60 million parts with 685 different part numbers had to be purchased to create the ORNL system. During construction, there was a chip shortage that affected 167 of those part numbers, so ORNL faced a shortage of two million parts. AMD also ran into trouble as 15 part numbers for its MI200 GPUs were in short supply. To circumvent the shortage, ORNL partnered with ASCR to secure a Defense Priorities and Allocation System (DPAS) rating for these parts, meaning the US government used the Defense Act to procure parts due to Frontier’s importance to national defense.
While the system currently peaks at 29 MW, Frontier’s mechanical unit can cool up to 40 MW of computing power, the equivalent of 30,000 US homes. The plant can be expanded up to 70 MW, leaving room for future growth.
While Frontier gets the title of the world’s first officially recognized Exascale supercomputer, China is believed to have two Exacscale supercomputers, Tianhe-3 and OceanLight, that broke the barrier a year ago. Unfortunately, these systems were not submitted to the Top500 committee due to political tensions between the US and China. However, the lack of official Top500 filings—Gordon Bell was a proxy—has led to some doubts that these are truly exascale systems, at least in terms of FP64 workload.
The Frontier is now officially the world’s fastest supercomputer and the first to officially break the exascale barrier. The near-mythical, long-delayed Intel-based Aurora is expected to arrive later this year or early next year and deliver up to 2 exaflops of performance, vying with Frontier for the top supercomputer spot.
What’s next for AMD? El Capitan, a 2+ ExaFlop performance machine, which, according to the latest data, will appear in 2023. When completed, this Zen 4-based supercomputer will compete with the Intel-based Aurora for the title of the fastest supercomputer in the Top500.
Some advertising
Thank you for staying with us. Do you like our articles? Want to see more interesting content? Support us by placing an order or recommending to friends,
cloud VPS for developers from $4.99
,
a unique analogue of entry-level servers, which was invented by us for you:The whole truth about VPS (KVM) E5-2697 v3 (6 Cores) 10GB DDR4 480GB SSD 1Gbps from $19 or how to share a server?
(available with RAID1 and RAID10, up to 24 cores and up to 40GB DDR4).
Dell R730xd 2 times cheaper in the Maincubes Tier IV data center in Amsterdam? Only here 2 x Intel TetraDeca-Core Xeon 2x E5-2697v3 2.6GHz 14C 64GB DDR4 4x960GB SSD 1Gbps 100 TV from $199 in the Netherlands! Dell R420 – 2x E5-2430 2.2Ghz 6C 128GB DDR3 2x960GB SSD 1Gbps 100TB – from $99! Read about how to build infrastructure corp. class with the use of Dell R730xd E5-2650 v4 servers worth 9,000 euros for a penny?
