Particle.news
Download on the App Store
7 ARTICLES
·
4w ago

AWS and NVIDIA Expand AI Partnership With NVLink Fusion, New Trainium3 Servers and AI Factories

The expanded pact brings NVIDIA interconnects into AWS to speed large-scale training, offering sovereign, dedicated AI capacity.

NVIDIA logo is seen near computer motherboard in this illustration taken January 8, 2024. REUTERS/Dado Ruvic/Illustration
Image
Image
AWS said its new chip is lower cost than rivals and delivers over four times the computing performance of its predecessor while using 40 percent less energy

Overview

  • AWS confirmed support for NVIDIA NVLink Fusion in a future Trainium4 and plans to extend the interconnect to Graviton CPUs and the Nitro System with NVIDIA MGX rack integration.
  • New Trainium3 servers are available now with 144 chips per system, delivering more than four times the previous generation’s compute while using about 40% less power, according to AWS.
  • AWS broadened GPU choices with NVIDIA Blackwell hardware, including HGX B300 and GB300 NVL72 systems, providing immediate access for training and inference.
  • The companies launched AWS AI Factories to provide dedicated, sovereign AI infrastructure operated by AWS inside customer data centers.
  • NVIDIA Nemotron open models are now available on Amazon Bedrock, and Amazon OpenSearch Service added serverless GPU vector indexing via NVIDIA cuVS that NVIDIA says can be up to 10x faster at a quarter of the cost.