Last week, the community team went to ACM/IEEE Supercomputing ‘23 – the world’s largest HPC (High Performance Computing) conference. The show floor was packed with supercomputers from different vendors, liquid cooling tech, a student competition, and more. It was pretty awe inspiring to walk through the floor and hear the hum of these massive machines that our day-to-day AI workloads run on!

Here are our takeaways:

Aurora named #2 on Top 500 list:

Aurora is a brand new supercomputer being installed at Argonne National Laboratory and is built in partnership with HPE.

The Top 500 list is a ranking of the most powerful supercomputers in the world. The list is determined by how well a supercomputer performs on the LINPACK benchmark, which is based on solving a dense system of linear equations.  Read more about the Top 500 here.

The top 3 on the Top 500 list this year are:

1. Frontier – HPE Cray EX

2. Aurora – HPE Cray EX

3. Eagle – Microsoft NDv5

Frontier is currently the only Exascale computer on the list.  

What is Exascale computing anyway?

Exascale computers are able to perform 1,000,000,000,000,000,000 floating point operations per second (FLOPS). Simulation of the fundamental forces of the universe, training AI models, and other types of modeling are all accelerated by this insanely powerful computer, located at Oak Ridge National Laboratory.

Although Aurora is #2 on the list, it will eventually be able to deliver over 2 exaflops of computing power, surpassing even Frontier. Read more about Aurora here.

NVIDIA announces the H200 GPU

According to HPCwire, the H200 is an incremental improvement over the H100 – it’s 1.6 times faster than the H100 on GPT-3 and 1.9 times faster on Llama. It’s also two times faster than the A100. Sadly, we did not see the H200 at SC23, but here’s a pic of the H100 from the HPE booth:

While we’re here, here’s the inside of a Cray XD system:

And a Cisco system by the University of Texas at Dallas booth:

And here’s some immersion-cooled hardware. It’s surreal to see in person.

GE Aerospace Research wins Editor’s Choice “Top SuperComputing Achievement”

HPCwire hosts a competition called the HPCwire Readers’ and Editors’ Choice Awards. “Selected by the readers and the editorial team, these awards recognize the technologies, use cases, companies, and leaders that have contributed to the continued growth of the HPC market and community” - HPCwire.

Here’s a quote from the article:

GE Aerospace Research and GE Aerospace have pioneered the use of Oak Ridge National Laboratory’s Frontier, an HPE Cray EX supercomputer. They’re employing computational fluid dynamics simulations to model complex turbulent flow, studying the performance of open fan engine architecture for the next-generation aircraft engine (RISE) potentially cutting CO2 emissions by over 20% and redefining the future of flight.

Read about other highlights from the show here.