Particle.news
Download on the App Store
8 ARTICLES
·
yesterday

Nvidia Debuts Rubin CPX, a Long-Context GPU to Split LLM Inference Workloads

Built for inputs over one million tokens, the part is slated to ship in about a year with claimed throughput gains for video and code workloads.

An NVIDIA logo and a computer motherboard appear in this illustration taken August 25, 2025. REUTERS/Dado Ruvic/Illustration/File Photo
Image
Image
Jensen Huang, co-founder and chief executive officer of Nvidia Corp., during the Nvidia GPU Technology Conference (GTC) in San Jose, California, US, on Tuesday, March 18, 2025. Nvidia Corp., the chipmaker that’s become synonymous with the artificial intelligence boom, said it will help General Motors Co. develop self-driving vehicles and work with the telecom industry on new 6G networks. Photographer: David Paul Morris/Bloomberg

Overview

  • Nvidia introduced Rubin CPX as a context-encoding processor that offloads the compute-heavy prefill phase from generation-focused GPUs.
  • CPX will ship in two forms: an integrated Vera Rubin NVL144 CPX compute tray for new builds and standalone CPX racks that attach to existing Rubin deployments.
  • Vendor slides cite up to roughly 6.5x faster performance on very large-context tasks versus GB300/Blackwell Ultra, a claim that awaits independent benchmarks.
  • Nvidia highlights large codebases and multi-frame video as early use cases, aiming to cut time-to-first-token and boost overall inference throughput.
  • Availability is targeted for roughly a year from now, with Reuters reporting launches by the end of next year, and Nvidia pitching improved token economics for data centers.