News : SpiNNcloud Systems Announces First Commercially Available Neuromorphic Supercomputer

SpiNNcloud board
Sandia National Laboratories is Among the First Customers Using SpiNNaker2 for Research and Development in Nuclear Deterrence Field and Other AI Applications.

Today, in advance of ISC High Per­for­mance 2024, SpiNN­cloud Sys­tems announ­ced the com­mer­cial avai­la­bi­lity of its SpiNNaker2 plat­form, a super­com­pu­ter-level hybrid AI high-per­for­mance com­pu­ter sys­tem based on prin­ci­ples of the human brain. Pio­nee­red by Steve Fur­ber, desi­gner of the ori­gi­nal ARM and SpiNNaker1 archi­tec­tures, the SpiNNaker2 super­com­pu­ting plat­form uses a large num­ber of low-power pro­ces­sors for effi­ci­ently com­pu­ting AI and other workloads.

First-gene­ra­tion SpiNNaker1 archi­tec­ture is curr­ently used in dozens of rese­arch groups across 23 count­ries world­wide. San­dia Natio­nal Labo­ra­to­ries, Tech­ni­cal Uni­ver­sity of Munich and Uni­ver­sity of Göt­tin­gen are among the first cus­to­mers pla­cing orders for SpiNNaker2, which was deve­lo­ped around com­mer­cia­li­zed IP inven­ted in the Human Brain Pro­ject, a bil­lion-euro rese­arch pro­ject fun­ded by the Euro­pean Union to design intel­li­gent, effi­ci­ent arti­fi­cial systems.

»Brain-like com­pu­ta­tion requi­res pro­gramma­ble dyna­mics, event-based com­mu­ni­ca­tion, and extreme scale,« said Fred Roth­gan­ger from San­dia Natio­nal Labs. »SpiNNaker2 is the most fle­xi­ble neu­ral super­com­pu­ter archi­tec­ture available today. At San­dia, we are exci­ted to build appli­ca­ti­ons on this ama­zing system.«

Tech­ni­cal spe­ci­fi­ca­ti­ons for SpiNNaker2 include:

  • The SpiNN­cloud ser­ver board con­sists of 48 SpiNNaker2 chips. Each chip con­ta­ins a low-power mesh of 152 Arm-based cores with a wide variety of acce­le­ra­tors inclu­ding dis­tri­bu­ted GPU-like units desi­gned to boost neu­ro­mor­phic, hybrid, and main­stream AI model computations.
  • The hybrid capa­bi­li­ties of SpiNNaker2 allow it to com­bine prac­ti­cal algo­rithms from Machine Lear­ning, energy effi­ci­ency from an event-based ope­ra­tion, and the trust­wort­hi­ness of rule-based lay­ers gover­ning the backend of applications.
  • The com­mer­cial sys­tem is inten­ded to be sca­led up through a multi-rack setup, each com­po­sed of 90 SpiNN­cloud ser­ver boards to enable at least 10 bil­lion inter­con­nec­ted neu­rons firing in real-time. As a super­com­pu­ter desi­gned to also allo­cate Machine Lear­ning workloads, the pre-exas­cale sys­tem can sup­port up to 0.3 exaops.
  • SpiNNaker2 dif­fe­ren­tia­tes from GPU solu­ti­ons by being more ver­sa­tile, having a bet­ter per­for­mance, having a lower cost to run and having a bet­ter availability.
  • In con­trast to GPU solu­ti­ons, SpiNNaker2 uses a large num­ber of asyn­chro­nous low-power units to tackle workloads. This leads to more ver­sa­ti­lity, energy-pro­por­tio­nal ope­ra­tion, effi­ci­ent com­mu­ni­ca­tion, and lower ope­ra­tion cost. SpiNN­cloud is the first to make this neu­ro­mor­phic super­com­pu­ting archi­tec­ture broadly available and at sca­les that exceed any other neu­ro­mor­phic system.
  • SpiNNaker2 goes bey­ond neu­ro­mor­phic archi­tec­ture by offe­ring hybrid AI acce­le­ra­tion, cru­cial for desig­ning sys­tems »in the third wave of AI,« defi­ned by DARPA as sys­tems that under­stand the con­text and envi­ron­ment in which they ope­rate, and over time build under­ly­ing expl­ana­tory models that allow them to cha­rac­te­rize real-world phenomena.
  • SpiNNaker2 will be initi­ally available as a cloud plat­form solu­tion, enab­ling early cus­to­mers to get access to the tech­no­logy as cost effi­ci­ently as possible.

»Our vision is to pio­neer the future of arti­fi­cial intel­li­gence through brain-inspi­red super­com­pu­ter tech­no­logy, and to create sys­tems that are not only powerful but also trans­pa­rent and explainable,« said Hec­tor Gon­za­lez, co-foun­der and co-CEO at SpiNN­cloud Sys­tems. »We’re buil­ding the most advan­ced brain-like super­com­pu­ting plat­form on the mar­ket, which posi­ti­ons us as the mar­ket lea­der in hybrid AI HPC. And it enables us to drive the deve­lo­p­ment of relia­ble, effi­ci­ent hybrid AI sys­tems, ope­ning up enti­rely new mar­kets in next-gene­ra­tion defense, drug dis­co­very, quan­tum emu­la­tion, smart city appli­ca­ti­ons and more.«

The com­pany’s cloud ser­vice will be available in the second half of 2024, and full pro­duc­tion sys­tems will be available for ship­ment in the first half of 2025.

Learn more about SpiNN­cloud here.