#graceblackwell — Public Fediverse posts

Supermicro’s GB300 Super AI Station puts NVIDIA Grace Blackwell hardware in a deskside tower, but the price and infrastructure needs make it feel more like a compact data-center no

#tech #technology #supermicro #nvidia #aiworkstation #graceblackwell

https://techshowup.com/News/Article/019ff0ce-34cf-7755-bc1c-b753f412e6f8/supermicro-gb300-super-ai-station-brings-data-center-ai-hard

#Nvidia’s #VeraRubin Platform Obsoletes Current AI Iron Six Months Ahead Of Launch
Nvidia’s top brass says Vera-Rubin #NVL72 #rackscale system, with 72 GPU sockets and 36 CPU sockets as well as NVSwitch fabric linking the, delivers a 10X reduction in #inference cost per token for MoE #AI models and 4X reduction in number of GPUs (but maybe not a 4X reduction in cost, be careful) to train those models compared to the prior generation of #GraceBlackwell NVL72 systems.
https://www.nextplatform.com/2026/01/05/nvidias-vera-rubin-platform-obsoletes-current-ai-iron-six-months-ahead-of-launch/

#Arm wrestles away 25% share of #server market thanks to Nvidia's home-grown #CPU
That’s up from 15% of the server market a year ago but still far short of 50% market share Arm infra chief was hoping for. Driving force behind this trend is adoption of Nvidia’s #GraceBlackwell #rackscale compute platforms like GB200 and GB300 #NVL72. Each of the 120-kilowatt machines is equipped with 72 Blackwell GPUs and 36 of Nvidia’s Grace CPUs based on the Arm #NeoverseV2 architecture.
https://www.theregister.com/2025/09/11/nvidias_graceblackwell_drives_arms_cpu/