NVIDIA has unveiled a brand new suite of space-focused computing platforms centred on the House-1 Vera Rubin Module, because it seems to allow extra orbital information centres.
Introduced in the course of the firm’s presentation at its annual GPU Expertise Convention, referred to as GTC, the House-1 Vera Rubin Module is designed for orbital information centres, geospatial intelligence and autonomous house operations, with the Rubin GPU on the module claimed to ship as much as 25 occasions extra AI compute for space-based inferencing than the NVIDIA H100 GPU.
There was rising curiosity in orbital information centres over the previous few years, with many seeing it as a golden alternative to construct capability with out having to cope with a number of the giant constraints which can be holding again information centres on earth. That’s as a result of they don’t require the prolonged planning course of related to land-based information centres, whereas in addition they profit from an abundance of renewable power – which could be generated 24/7 from on-board photo voltaic panels.
Orbital information centres usually are not there to solely change land-based information centres, nevertheless. Actually, one of many key advantages of getting an AI information centre in house is solely to permit it to course of the info that’s already being generated in house – from different satellites, sensors and spacecraft. The sensible upside is quicker selections, decrease reliance on scarce downlink bandwidth, and the flexibility to ship again solely probably the most helpful outcomes relatively than big volumes of uncooked information.
There’s one main draw back to orbital information centres, nevertheless. Whilst you may not have to search out an enormous plot of land, an orbital information centre isn’t going to have the identical computing energy as a facility situated on planet Earth. That’s as a result of they’re inherently constrained by, relatively mockingly, house. Not solely is there a restrict within the measurement of the info centre you may ship up into house, however there’s additionally a strict restrict on weight – with SpaceX’s Falcon 9, a standard launch platform for sending satellites into orbit, restricted to a most payload of twenty-two,800kg.
It’s with that measurement constraint in thoughts that has seen NVIDIA announce its suite of space-focused computing platforms. It provides operators a full compute stack from orbit to floor, whereas its House-1 Vera Rubin is designed to convey data-centre-class AI into orbit, with 25x extra AI compute for space-based inferencing than the H100 chip. It’s additionally vital to notice that it is a chipset that has been particularly engineered for size-, weight- and power-constrained environments, so it ought to make constructing an orbital information centre for AI considerably simpler and more cost effective.
The launch of NVIDIA’s space-based suite comes on the proper time for the market too. There have already been quite a few corporations clamouring to launch orbital information centres, with Axiom House having beforehand introduced its intentions to launch proof-of-concept information centres into house. It finally achieved that goal earlier this year, but it surely actually received’t be the final.
It also needs to be famous that Axiom was title dropped throughout NVIDIA’s GTC presentation, with the corporate mentioned to already be utilizing NVIDIA’s accelerated computing platforms for next-generation missions. Aetherflux, Kepler Communications, Planet, Sophia House and Starcloud are amongst the opposite corporations already utilizing NVIDIA’s platforms.
