Look ma, no #GPUs

AWS is building a machine with hundreds of thousands of #Tranium2 processors to achieve performance of ~65 ExaFLOPS for #AI

AWS also unveiled its Trainium3 processor which will quadruple performance compared to the Trainium2

www.tomshardware.com/tech-industr...

#HPC

The Claude Haiku 3.5 running on Tranium2 chips is interesting. I need to rewatch but i think he said 60% faster. Impressive benchmark.

BTW, Claude is my preferred LLM at the moment. Its results and analytics of data given to it are quite impressive.

Apple says it's using Amazon's Trainium and Graviton chips to serve search services and is evaluating if Tranium2 chips can be used to pre-train its AI models (Kif Leswing/CNBC)

Main Link | Techmeme Permalink

AmazonがAWS向けのプロセッサ「Graviton4」とAIトレーニング用チップ「Trainium2」を発表
https://gigazine.net/news/20231129-amazon-graviton4-trainium2/

Amazon EC2 gets Tranium2 instances. AWS silicon made for AI inference workloads. Can also be used for training and inference workloads. "Naming isn't always our strong suit," jokes Garman. #reinvent

One of the key takeaways for me from last night’s AMZN call was the discussion around Trainium - Jassy flagged that they are sold out of Tranium2 and have had discussion with “large and medium” sized players about deploying Tranium3.

That would be a huge win for AWS and the bet they made

Tranium2 ULTRASERVERS - connecting 4 instances together for 62 Tranium2 chips for unparalleled power

re:Invent keynote: Tranium2 instances now GA - game changing for training genai apps