Over the past couple of years, the world’s most powerful supercomputers have experienced a sizable leap in performance. The combined processing power of the computers on the TOP500 list surged from 5.24 exaflops in June 2023 to 11.72 exaflops in November 2024, representing a 123.7% increase.
Meanwhile, the anticipated Colossus supercomputer from Elon Musk’s xAI—if it truly scales to one million GPUs—could dwarf even these top-tier HPCs, suggesting a future where HPC’s upper limit stands far beyond the current exaflop frontier.
This chart visualizes estimated total cores (CPU + GPU) for El Capitan and xAI’s Colossus at different phases (100k, 200k, and a future 1 million GPUs).
A wave of new exascale systems—capable of at least a quintillion calculations per second—has propelled the latest surge in supercomputing performance. The top slot belongs to Lawrence Livermore National Laboratory’s El Capitan, weighing in at 1.742 exaflops of Rmax (the maximum performance a computer achieves). Oak Ridge National Laboratory’s Frontier follows closely, its score jumping from 1.194 exaflops in June 2023 to 1.353 exaflops in the most recent ranking. Making a strong first showing, Argonne National Laboratory’s Aurora claims third place with 1.012 exaflops in its initial benchmark submission.
The following data points are from a review of data from Top500.
1. The US still leads the HPC landscape
Key takeaway: America’s dominance remains intact, but Europe and Asia each boast multiple exascale-class installations.
The performance-labeled map and rank columns demonstrate continued U.S. leadership, especially in the top 10 (six American systems totaling ~4 exaflops). Europe’s combined systems surpass 1 exaflop, thanks to LUMI in Finland, Leonardo in Italy, and Alps in Switzerland. Anchoring Asia’s prominence are large-scale supercomputers in China (Sunway TaihuLight) and Japan (Fugaku). Further down the list, smaller ranks highlight Australia’s Setonix (GPU partition at #45) and Saudi Arabia’s Shaheen III (#38).
The U.S. maintains a commanding lead in HPC, with China ascending.
[Data courtesy of: https://top500.org/]
2. Industry in the lead: Nearly half of TOP500 systems serve commercial needs
Key takeaway: HPC is no longer just for large government labs or academic research—commercial usage has become the single biggest slice.
According to the latest TOP500 data, nearly half of all listed systems (48.6%) serve industrial needs—ranging from automotive crash simulations to oil and gas reservoir modeling—indicating a strong commercial demand for HPC resources. Another 36.5% primarily support research-oriented workloads, such as astrophysics, fluid dynamics, and genomics. Government installations make up 7.1% of the chart, and 5.4% are hosted by technology vendors who often showcase HPC as part of their in-house R&D or service offerings. The remaining slice falls under “Others,” including smaller organizations and specialized sectors.
Although the chart above appears interactive, it does not support hover annotations. Instead, we reference CSV data that breaks down each system by usage category. The same data also highlights the growth of cloud-based HPC, which—while still a slim fraction—lets smaller R&D teams rent on-demand computing muscle that was once the exclusive domain of large research labs or enterprise data centers.
Research and industry applications dominate HPC use.
[Data courtesy of: https://top500.org/]
3. Three US systems break exaflop barrier in 2024
Key takeaway: The exaflop era is firmly here, with multiple U.S. supercomputers breezing beyond 1 EF.
A glance at the bar chart below underscores El Capitan’s breakthrough >1.7 exaflops (Rmax). Frontier, though slightly behind, remains a core exascale system. Simulations once deemed “unthinkable”—complex multi-physics or large-scale AI—become routine at these speeds.
El Capitan HPC surpasses 1.7 exaflops of Rmax performance.
[Data courtesy of: https://top500.org/]
4. Most top systems achieve 60%+ of peak performance
Key takeaway: Modern HPC software and hardware design converge to make real-world performance consistently high relative to theoretical max.
Recent data shows that most TOP500 systems exceed 60% of their theoretical peak (Rpeak) in practice. For instance, HPC6 (#5) delivers nearly 79% efficiency (478 vs. 607 petaflops), and Tuolumne (#10) runs at about 72% (208 vs. 289 petaflops). Overall, 311 of the 500 listed HPCs achieve at least 60% efficiency, with 199 surpassing the 70% mark.
5. Top-tier HPC systems draw 25 MW, or more…
Key takeaway: Power usage has escalated alongside exascale performance, pushing engineers to devise new cooling strategies and more efficient architectures.
At the high end of the TOP500 list, several exascale HPC systems surpass 0.5 exaflops of Rmax—most notably, El Capitan (#1) at 1.74 EF, Frontier (#2) at 1.35 EF, and Aurora (#3) at 1.01 EF. While El Capitan and Frontier each operate around the 25–30 MW range, which is enough to power tens of thousands of U.S. households, Aurora tops even that at 38.7 MW—making it the highest power draw among current exascale machines. For context, Aurora consumes electricity on the order of what a small city of roughly 31,000 households would use (38,698.36 kW × 24 hours = 928,760.64 kWh per day, which divided by the average U.S. household usage of 30 kWh per day equals 30,959 homes).
Beyond the aforementioned systems leaders, other HPCs also enter the “tens of megawatts” territory. The Japanese supercomputer Fugaku (#6), for instance, delivers 442 PF at 29.9 MW, while Sunway TaihuLight (#15) and Tianhe-2A (#24) each exceed 15 MW. Even some older CTS-1 clusters at LLNL—Jade (#411) and Quartz (#412)—still hover above 13 MW, though their Rmax values are orders of magnitude smaller than the latest generation’s multi-petaflop to exaflop benchmarks.
Rmax vs. power usage among HPC systems.
[Data courtesy of: https://top500.org/]
6. Four systems cross 0.5 Exaflop threshold
Key takeaway: There aren’t just one or two exascale systems—at least four HPCs surpass half an exaflop. In 2025, that number is likely to begin to grow.
Recent data confirms that El Capitan (#1), Frontier (#2), Aurora (#3), and Eagle (#4) all surpass 0.5 exaflops (EF) of Rmax, with El Capitan hitting 1.74 EF and Frontier at 1.35 EF. Both Frontier and Aurora also appear lower in the rankings in other system configurations—Frontier (#68) at just 0.019 EF and Aurora (#119) at 0.010 EF. Despite that, their top entries achieve full exascale performance. Eagle’s primary listing (#4) nears 0.56 EF, confirming it crosses the 500-petaflop threshold.
A bar chart highlights these exascale machines’ performance:
- El Capitan (#1): 1.74 exaflops Rmax
- Frontier (#2): 1.35 exaflops Rmax
- Aurora (#3): 1.01 exaflops Rmax
- Eagle (#4): ~561 petaflops, nearing the exaflop threshold
Exascale HPC systems above 0.5 exaflops Rmax.
[Data courtesy of: https://top500.org/]
Tell Us What You Think!