On the current Worldwide supercomputing convention known as ISC 2024, Intel’s latest Aurora supercomputer put in at Argonne Nationwide Laboratory raised a couple of eyebrows by lastly surpassing the exascale barrier. Earlier than this, solely AMD’s Frontier system had been capable of obtain this stage of efficiency. Intel additionally achieved what it says is the world’s finest efficiency for AI at 10.61 “AI exaflops.”
Intel reported the information on its weblog, stating Aurora was now formally the quickest supercomputer for AI on the earth. It shares the excellence in collaboration with Argonne Nationwide Laboratory and Hewlett Packard Enterprise (HPE), which each constructed and homes the system in its present state, which Intel says was at 87% performance for the current assessments. Within the all-important Linpack (HPL) take a look at, the Aurora laptop hit 1.012 exaflops, which means it has virtually doubled the efficiency on faucet since its preliminary “partial run” in late 2023, the place it hit simply 585.34 petaflops. The corporate then stated it anticipated to cross the exascale barrier with Aurora finally, and now it has.
The Aurora supercomputer homes 10624 blades like this one, together with 84,992 HPE Slingshot ethernet interconnects.
Credit score: Intel
Intel says for the ISC 2024 assessments, Aurora was working with 9,234 nodes. The corporate notes it ranked second general in LINPACK, which means it is nonetheless unable to dethrone AMD’s Frontier system, which can be an HPE supercomputer. AMD’s Frontier was the first supercomputer to interrupt the exascale barrier in June 2022. Frontier sits at round 1.2 exaflops in Linpack, so Intel is knocking on its door however nonetheless has a solution to go earlier than it may topple it. Nevertheless, Intel says Aurora got here in first within the Linpack-mixed benchmark, reportedly highlighting its unparalleled AI efficiency.
Intel’s Aurora supercomputer makes use of the corporate’s newest CPU and GPU {hardware}, with 21,248 Sapphire Rapids Xeon CPUs and 63,744 Ponte Vecchio GPUs. When it is totally operational later this yr, Intel believes the system will finally be able to crossing the 2-exaflop barrier.
Nevertheless, AMD is not simply sitting round counting its flops. It is also planning to deploy its personal next-generation supercomputer this yr known as El Capitan, which is anticipated to supply 2 exaflops of efficiency. Nvidia has additionally joined the celebration and introduced right this moment that 9 supercomputers all over the world are actually working its Grace Hopper CPU+GPU combo. So, the supercomputer race is about to get fairly spicy within the coming years, particularly as AI efficiency turns into a paramount consideration.
