Eleven million cores. That’s not a typo. The new El Capitan supercomputer at Lawrence Livermore National Laboratory packs more processing power than most people can comprehend. This beast delivers 2.79 exaflops of peak performance, making it the fastest supercomputer on Earth as of 2024.
The machine’s architecture reads like science fiction. Each compute node houses four AMD MI300A APUs, cramming CPU and GPU chiplets together using 3D stacking technology. It’s dense, efficient, and probably worth more than your house. Actually, definitely worth more than your house. With 5.475 petabytes of HBM3 memory spread across the system, El Capitan has enough storage to make your laptop cry.
Here’s where it gets wild. The entire system runs fanless. Direct liquid cooling keeps those 11 million cores from melting into expensive silicon pudding. Peak power consumption hits 35 megawatts, which sounds like a lot because it is. But for a machine solving problems that would take regular computers centuries, it’s surprisingly efficient. The system spreads across 7,500 square feet of floor space, roughly the size of two tennis courts.
The Slingshot interconnect technology links everything together, minimizing latency between nodes. Each MI300A device gets 128 GB of shared memory for both CPU and GPU tasks. The result? 250.8 teraflops per node, adding up to 2,792.9 petaflops system-wide. Numbers that make physicists giddy.
El Capitan isn’t just another expensive toy. Lawrence Livermore, Los Alamos, and Sandia National Laboratories share this monster for both classified and unclassified research. Nuclear stockpile stewardship, climate modeling, materials science, astrophysics – the works. This collaborative approach proved essential in achieving the exascale computing milestone. Problems that need massive parallel computation finally have the horsepower they deserve.
The machine enables simulations at speeds and detail levels that were pipe dreams just years ago. Fluid dynamics, quantum materials research, and other compute-intensive fields suddenly have a new best friend. Time-to-solution for complex models? Slashed. Scientific discovery? Accelerated.
Built with modularity in mind, El Capitan can handle future hardware updates. Smart move, considering how fast technology evolves. For now, though, those 11 million cores represent the bleeding edge of computational physics. Not bad for a fanless computer.
References
- https://en.wikipedia.org/wiki/El_Capitan_(supercomputer)
- https://asc.llnl.gov/exascale/el-capitan
- https://www.hpe.com/us/en/compute/hpc/lawrence-livermore-national-laboratory.html
- https://hpc.llnl.gov/documentation/user-guides/using-el-capitan-systems/hardware-overview
- https://www.nextplatform.com/2024/11/18/el-capitan-supercomputer-blazes-the-trail-for-converged-cpu-gpu-compute/
