At this year’s re:Invent conference, Amazon Web Services (AWS) unveiled its latest Graviton processor chip and a new machine learning chip called Trainium.

The Graviton 3 is the newest generation of Amazon’s ARM-based processor and is designed to compete with Intel and AMD central processors.

Heralded as 25 percent faster than the last generation, AWS promises the new chip has 2x faster floating-point performances, a 3x speedup for machine-learning workloads, all whilst using 60 percent less power.

The new Trainium chip is designed to train machine learning models to compete against similar NVIDIA chips. The company expects it to do so at a cost 40 percent lower than NVIDIA’s top chip.

AWS has been working on developing its own chips since buying the start-up Annapurna Labs in 2015.

AWS VP and chief evangelist Jeff Barr noted the variety of workloads the new chips could be used for, including “HPC, batch processing, electronic design automation (EDA), media encoding, scientific modelling, ad serving, distributed analytics, and CPU-based machine learning inferencing”.

For now, it is likely that only experienced early adopters who can re-write their software for new processors will try the AWS chips. Decades of software has been written for Intel and NVIDIA chips, and thus more will need to be done to inspire mass adoption.

Despite this, competing firms should take Amazon’s ambitions seriously in the long-term. When AWS first launched in 2006 it was a small fish in a big pond, but now it is on track to overtake massive traditional firms like Cisco Systems in revenue.

The annual re:Invent conference, which is now in its tenth year, began in Las Vegas, Nevada on November 29 and will conclude on December 3.

Looking to learn how to establish a strategic hybrid cloud? Learn more about the virtual Hybrid Cloud Congress, taking place on 18 January 2022 and explore how to optimise and unleash the power of your hybrid cloud.

Tags: Amazon, AWS, Graviton 3, Intel, machine learning, NVIDIA, processors, re:Invent, Trainium

Source link


Please enter your comment!
Please enter your name here