Amazon Web Services said it will deploy simplified electrical and mechanical designs, liquid cooling, new rack designs and updated control systems to handle AI workloads sustainably.

The news, outlined at re:Invent 2024 in Las Vegas, landed ahead of CEO Matt Garman's keynote on Tuesday. AWS said the new flexible data center components will enable it to provide 12% more compute power while boosting availability and efficiency.

More re:Invent 2024:

AWS, like other hyperscale data center operators, is revamping designs and offering custom silicon to become more efficient to handle AI workloads and hit sustainability goals. AWS said the components will be modular and retrofit existing infrastructure. These additions will also support GPU-based servers, which will require liquid cooling.

Here's a look at the changes:

  • Simplified electrical distribution systems that minimize downtime and the number of racks impacted by electrical issues by 89%. AWS said it has reduced the number of failure points by 20%. AWS also brought backup power closer to the rack and reduced the number of fans.
  • AWS added configurable liquid-to-chip cooling in new and existing data centers. Updated systems will integrate air and liquid cooling for AI chips including AWS Trainium 2 and Nvidia GB200.
  • The company changed how it positions racks in a data center and optimized for high-density AI workloads. Software additions will predict the most efficient ways to place servers.
  • AWS is building out its control systems to standardize monitoring, alarms and operating tools.

As for sustainability, AWS said that it has been able to cut mechanical energy consumption by 46% with a 35% reduction in carbon used in concrete.

Business Research Themes

Tech Optimization

Roles

Chief Information Officer

Tags

#AWS, #AWSreinvent, hyperscale data centers