The Long and Winding Road

There’s a sentence I’ve waited years to write, so I’m going to write it carefully. The Department of Energy’s Aurora supercomputer has entered production at the Argonne Leadership Computing Facility, and the exascale system is now open to the broader research community.

If you don’t work in this corner of the field, that may read as a routine announcement. A machine was built; the machine was switched on. I’d like to gently push back on that, because Aurora was never routine, and the road that led here was neither short nor straight.

Here are the facts I’m proud of. Aurora was built in partnership with Intel and Hewlett Packard Enterprise. It integrates roughly 60,000 Intel Data Center GPU Max Series accelerators across its 10,000 nodes, and it now stands as the second U.S. exascale system available to open science, alongside Oak Ridge’s Frontier. Those are genuinely staggering numbers. But numbers have a way of hiding the years behind them, and it’s the years I want to talk about.

A system like this does not simply arrive. It is coaxed into existence. There were several years of system bring-up, of scaling tests, of Early Science Program campaigns that pushed the machine in ways its earliest plans never anticipated. There were long stretches where progress was measured in small, unglamorous increments, and stretches where the most honest status update we could give was “not yet.” Anyone who has built something genuinely hard knows that particular feeling: the milestone keeps receding, and you keep walking toward it anyway. Aurora asked that of a lot of people for a long time.

This next part is the one I most want to get right, because it’s the part that doesn’t fit in a press release. The reason Aurora got here is not a vendor and not a budget line. It’s a team. It’s people who showed up through the uncertainty, who debugged at scale when the scale itself was the problem, who learned to treat each setback as information rather than as defeat. I have the enormous privilege of working alongside them, and I want to be plain about my own part in it: mostly, I got to watch extraordinary people refuse to give up. A facility is not its hardware. It is the patience and the stubbornness and the plain care of the people who run it. I have never been more certain of that than I am this week.

So what happens now. Allocations on Aurora, including those awarded through INCITE, ALCC, and the ALCF Director’s Discretionary program, move into routine production use. The system will carry a portfolio of science campaigns spanning materials, climate, fusion, drug discovery, cosmology, and frontier AI. Researchers who have waited their turn now get the machine. That is the entire point of the long road. We didn’t build Aurora in order to have built Aurora. We built it so that someone, somewhere, can ask a question that was previously unaskable and actually get an answer back.

And so, for our deployment team, the production milestone is the start of the story rather than the end of it. The hard, ordinary, beautiful work of science begins. The road was long and it was winding, and I would not trade the people I walked it with for anything. Coverage of the milestone is available at HPCwire.