Frontier, the groundbreaking supercomputer at Oak Ridge National Laboratory in Tennessee, continues to claim its title as the world’s most powerful non-distributed supercomputer. As noted in the latest biannual Top500 list, Frontier, also known as OLCF-5, has held this prestigious position since June 2022. This advanced AMD/HPE system can achieve a staggering performance of 1.194 Exaflops per second. To put this into perspective, one exaflop is the equivalent of a single human performing one calculation every second for a remarkable 31.7 billion years. With over 8.6 million cores, Frontier is expertly engineered for high-performance computing and artificial intelligence tasks. Its capabilities support critical projects, including comprehensive mappings of nuclear reactor lifecycles and significant breakthroughs in genetics related to disease research. Presently, it remains the only exascale supercomputer featured on the Top500 list, although this might soon change.
### Intel’s Emergence
Intel is rapidly closing in on Frontier with its Aurora supercomputer, developed in collaboration with Argonne National Laboratory. Aurora has recently edged out Fujitsu’s Fugaku for the second spot in the rankings. Although still in the commissioning phase, Aurora is projected to deliver an impressive 2 Exaflops per second upon completion. Located at the Argonne Leadership Computing Facility in Illinois, the system consists of 10,000 servers that integrate over 21,000 CPUs and an extraordinary 60,000 GPUs. Notably, the results submitted for this latest ranking only showcased half of Aurora's full capabilities, suggesting that it could surpass Frontier in future assessments.
One of the groundbreaking projects being conducted on Aurora involves generative AI, where the supercomputer is tasked with running a one-trillion-parameter version of OpenAI's leading large language model, GPT-3. This massive model will be trained on a wealth of scientific literature, coding sources, and various scientific datasets, making it a valuable resource for researchers in fields such as cancer studies, climate science, and cosmology. Leveraging Intel’s Max Series GPU architecture, Aurora exhibits the capability to handle vast models with just 64 nodes. With successful demonstrations of running multiple instances on 256 nodes, Argonne National Laboratory has proven Aurora’s potential to accelerate the training of colossal models using trillions of tokens, all across an extensive network of 10,000 nodes.
Though other Intel contributions to the list did not achieve the height of Aurora, they still deserve mention. Crossroads, from Los Alamos National Laboratory, ranks at 24th, while the newly deployed Dawn Phase 1, the fastest AI supercomputer in the U.K. for the time being, comes in at 41st. Additionally, SuperMUC-NG Phase 2, located at the Leibniz Supercomputing Center in Germany, secures the 52nd position. The Top500 list also highlights the emergence of 20 new systems utilizing Intel's Sapphire Rapids CPUs, which positions it as the leading CPU among new entries. However, of the 45 newly added systems, only four are equipped with Intel GPUs.
### Microsoft Rises to Power
Microsoft has also made a substantial leap forward. The Azure-based Eagle supercomputer has soared into the top three, displacing Fugaku to the fourth position. Eagle boasts 1.12 million computing cores, which include Nvidia’s cutting-edge H100 chips alongside Intel’s Xeon Platinum 8480C processors. Launched in August, Eagle is tailor-made to provide access to expansive large language models, including OpenAI's GPT-4, hosted on Microsoft Azure. It also supports NDv5 virtual machines, making Eagle the highest-ranked cloud-based system ever recorded in the Top500 list, as well as the fastest H100-equipped system among the top ten.
### Additional Insights from Top500
Among the top 10 supercomputers, processors from Intel, AMD, and IBM are preferred choices for high-performance computing systems. In this elite group, five utilize Intel Xeons, while two feature AMD hardware and another two incorporate IBM processors. The United States maintains its lead with 161 supercomputers, far ahead of China, which has 104 systems. Notably, North America has increased its count from 160 to 171 supercomputing systems, whereas Asia's total has decreased from 192 to 169. Europe, however, has seen growth, rising from 133 to 143 systems. Despite sliding down to fourth place, Japan's Fugaku retains its status as the highest-ranked supercomputer outside the U.S., demonstrating its ongoing significance in the global supercomputing landscape.
Ranking
