Might be a little smaller but i think HPE's emphasis on memory is spot on: https://www.theinquirer.net/inquirer/news/3034429/hpe-reveals-worlds-largest-arm-based-supercomputer-with-23-petaflops
They likely want to make an Exa scale competitor too: https://www.theregister.co.uk/2018/06/07/hpe_exaflops_memory_sharing/ >Might such a model not be better for future J-like processors hoovering up >and leaving in place large chunks of memory? ~greg http://krsnadas.org -- from: Ian Clark <[email protected]> to: [email protected] date: 20 June 2018 at 05:16 subject: Re: [Jchat] superdooper computer >What caught my eye was a configurable pool of heterogenous nodes, each with massive (to me) amounts of local storage and well-adapted to matrix operations. I got a picture in my mind's eye of J-like primitives linked in a "tacit" program. >The NVIDIA contribution was hazy in the Livescience article. A better clue is given here: https://en.wikipedia.org/wiki/Summit_(supercomputer) . I guess the IBM Summit isn't just a bunch of graphics cards. :-) ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
