Comments on: AWS And Nvidia Reboot “Ceiba” Supercomputer For Blackwell GPUs https://www.nextplatform.com/2024/03/20/aws-and-nvidia-reboot-ceiba-supercomputer-for-blackwell-gpus/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Mon, 01 Apr 2024 18:52:04 +0000 hourly 1 https://wordpress.org/?v=6.5.5 By: Calamity Jim https://www.nextplatform.com/2024/03/20/aws-and-nvidia-reboot-ceiba-supercomputer-for-blackwell-gpus/#comment-222106 Fri, 22 Mar 2024 00:02:50 +0000 https://www.nextplatform.com/?p=143854#comment-222106 Very impressive … this upgraded Ceiba (278x NVL-72 racks — each with 72 double-Hoppers) should bat between 2.1 and 2.6 EF/s at FP64! Quite a performance and density improvement over the originally planned 0.549 EF/s from 512x NVL-32 racks (each with 32 Hoppers). This liquid-cooled kapok-tree of Mayan mythology (and vigesimal numerals) looks to be an Aurora Sleeping Beauty-class machine, and a worthy contender to the swashbuckling stronger-swagger of El Capitan.

A bit like city buses then, you miss the first exaflopper, wait 2 years in vain for another one, and then wam! 4 double-deckers arrive at once (Aurora, ElCapitan, Ceiba, and Meta’s two single-deckers from the TNP’s 03/13/24)!

]]>