Nvidia has unveiled the Rubin platform, a brand new suite of six chips designed to help AI supercomputing infrastructure at scale.
The platform consists of the Vera CPU, Rubin GPU, ConnectX-9, NVLink 6 Change, SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet Change.
Entry deeper trade intelligence
Expertise unmatched readability with a single platform that mixes distinctive information, AI, and human experience.
Nvidia plans to deploy the platform within the second half of 2026, with hyperscalers together with AWS, Google Cloud, Microsoft and Oracle Cloud all asserting plans to combine Rubin-based techniques.
Microsoft will undertake the Vera Rubin NVL72 rack-scale techniques for its next-generation Fairwater AI information centres.
Rubin’s structure is predicated on shut codesign between compute, networking and storage {hardware}.
In keeping with Nvidia, this strategy reduces token era prices as much as tenfold, and cuts the variety of GPUs required for mixture-of-experts mannequin coaching by an element of 4 in contrast with its predecessor, Blackwell.
The corporate has named the platform after astronomer Vera Florence Cooper Rubin.
Two principal system configurations are detailed: the Vera Rubin NVL72, containing 72 GPUs and 36 CPUs per rack, and the HGX Rubin NVL8 server board with eight interconnected GPUs for x86-based generative AI.
The Rubin GPU features a third-generation Transformer Engine with hardware-accelerated adaptive compression and delivers as much as 50 petaflops of NVFP4 compute for inference workloads.
Nvidia’s sixth-generation NVLink presents 3.6TB/s bandwidth per GPU and 260TB/s at rack scale, enabling high-throughput communication for big fashions.
The Vera CPU options 88 customized cores with Armv9.2 compatibility and direct NVLink-C2C connectivity.
Safety and reliability are strengthened with third-generation Confidential Computing, deployed at rack scale on Vera Rubin NVL72 techniques, to maintain information protected throughout all processing domains.
The platform implements real-time well being checks and fault tolerance by way of a second-generation RAS engine.
BlueField-4 DPU introduces Superior Safe Trusted Useful resource Structure (ASTRA) for provisioning large-scale AI environments.
Networking advances embody the Spectrum-6 Ethernet Change constructed on 200G SerDes know-how with co-packaged optics optimised for AI workloads.
Spectrum-X Ethernet Photonics will increase reliability and energy effectivity whereas supporting long-range distributed operations as a single AI atmosphere.
Early adopters equivalent to CoreWeave plan to deploy Rubin-based infrastructure beginning in late 2026.
Server distributors Cisco, HPE, Lenovo, Dell Applied sciences, and Supermicro will supply Rubin-compatible techniques.
Analysis labs together with Anthropic, OpenAI, Meta, and Mistral AI intend to make use of the platform for bigger fashions and decrease latency.
OpenAI CEO Sam Altman stated: “Intelligence scales with compute. After we add extra compute, fashions get extra succesful, remedy tougher issues and make an even bigger affect for individuals. The Nvidia Rubin platform helps us hold scaling this progress so superior intelligence advantages everybody.”
Crimson Hat has prolonged its collaboration with Nvidia to offer an optimised software program stack masking Crimson Hat Enterprise Linux, OpenShift and Crimson Hat AI merchandise for Rubin deployments.
Join our each day information round-up!
Give your corporation an edge with our main trade insights.
