Microsoft Azure’s Eagle system has more computing power than all but two of the world’s most powerful supercomputers, and it’s open to developers to run their applications on it – albeit at a hefty price.
The Frontier and Aurora systems, the two most powerful supercomputers in the world, do both crossed the exascale barrierare closed to commercial interests and instead form the core of, for example, scientific research.
However, Eagle differs in that Microsoft has opened up its use via the Azure cloud platform to those who want to take advantage of the massive levels of performance and run high-performance computing (HPC) workloads.
Maximizing wealth, but at what cost?
With a maximum performance of 561 petaFLOPS, Eagle is a new entry on the TOP500 list and comes in third place. It’s the highest any cloud system has ever reached in this ranking of the world’s 500 most powerful supercomputers, with another Azure system previously breaking the top ten two years ago.
The 1,123,200-core Eagle is powered by the Intel
Although dependent on technology from partners such as Intel and Nvidia, there is a good chance that future systems that Microsoft builds will have the best GPUs and the best CPUs built by others in favor of custom silicon. This is especially relevant in light of a recent collaboration with Arm.
The Azure Cobalt CPUBased on the Arm Neoverse CSS architecture, it is the company’s first custom silicon built for the cloud, and may be the first in a line of chips that mean future supercomputers will be 100% powered by Microsoft.
But with Eagle, the fact that such a powerful supercomputer is available to, in theory, anyone with the capacity to pay could be a source of anxiety when it comes to cybersecurity. Because security is often seen as a barrier to achieving staggering levels of performance, there are concerns that vendors will compromise on certain measures in favor of maximizing performance.
While this may not seem like a big deal for systems like Frontier or Aurora, which have specific cases and very limited access to a handful of researchers, the same may not be true for cloud-centric systems like Eagle.