Featured in this episode of Chaos Lever
Intel has announced that the Aurora supercomputer installation is complete. The massive computing cluster is a collaboration between Intel, HPE, and the Department of Enegry, composed of 10,246 compute blades containing 63,744 Intel Data Center GPUs and 21,248 Xeon Max Processors.
At the moment, this is the largest GPU cluster in the world. The theoretical peak performance of the system is 2 exaflops. Current record holder on the TOP 500, an HPE Cray EX system named Frontier, hits a measly single exaflop. Just pathetic really.
Aurora is now going into the acceptance testing phase where the system will be stress-tested to identify bugs before full blown workloads are pushed to the system. What type of workloads? At the moment they are planning to run large, open-source generative AI models for science.
Based on Moore’s law, we should expect a similar amount of compute power to be on our phones in about a decade. Unfortunately, Siri will still be ducking terrible.