AWS Wrap: Blossoming AI Services and NVIDIA Alternatives
LAS VEGAS, NV — At its AWS re:Invent 2024 conference this week, Amazon Web Services showcased new mass-scale AI chips and systems as well as new solutions to streamline the way enterprises wrangle and manage data for use in generative AI (GenAI) applications.
A centerpiece of this week’s announcements is the Trainium2 (Trn2) chip, which is powering new services for GenAI training and inferencing tasks and is meant to compete with NVIDIA’s GPUs. Trainium2 was announced at last year's re:Invent.
Specifically, AWS has announced general availability of Amazon Elastic Compute Cloud (Amazon EC2) Trn2 instances, which feature clusters of 16 Trn2 chips that AWS claims “offer 30-40% better price performance than the current generation of GPU-based EC2 P5e and P5en instances.” Those instances are based on NVIDIA H200 GPUs! The new Trn2 instances are available in the US East (Ohio) AWS Region.
In an interesting twist that shows it's become even more focused on high-performance hardware, AWS announced some very powerful hardware clustering solutions for mass-scale AI training. A separate Trn2 UltraServer, now in preview, features a cluster of 64 Trn2 chips across four Trn2 instances, which according to AWS quadruples the resources of a single node and is meant to serve as a building block for supercomputers and AI factories. One of the more interesting additions AWS has introduced is its own proprietary chip-to-chip interconnect, NeuronLink, to connect 64 Trainium2 chips across four Trn2 instances. This is a similar approach to NVIDIA using own NVLink interconnect.
To access the rest of this article, you need a Futuriom CLOUD TRACKER PRO subscription — see below.
Access CLOUD TRACKER PRO
|
CLOUD TRACKER PRO Subscribers — Sign In |
