It’s only 25 miles between NCAR and NREL, so I have time to fuel up, grab a sandwich, and even do a car wash before my NREL tour stops. Feeling refreshed, I stop at their campus in Golden, CO, and begin the interview with Ray Grout, the director of NREL’s Computational Science Center.
NREL is all about energy efficiency and renewable energy, period. To support their mission, they mainly rely on Eagle, an 8 PFlop supercomputer, as well as a few smaller clusters. They are eagerly awaiting the arrival of Kestrel, which will weigh around 44 PFlops. It will be more heavily accelerated than Eagle, sporting a large number of GPUs. Which GPUs and how many will likely be released during SC22.
In the interview, we discuss how the lab is embracing AI and more specifically machine learning. In line with the mission of the lab, they are also working on the implementation of green AI, with the aim of reducing the energy used by AI/machine learning computation.
We had a nice transition in the trade-offs between specialty and general-purpose hardware. We also touched on a very recent meme that I heard recently – where scientists and researchers express interest in how much energy their specific runs will consume. As I have already mentioned, this point has come up time and time again during my recent travels.
We had a fun and thoughtful interview and I really appreciated that they took the time to organize this leg of the tour. We also appreciate our sponsors who made this possible: HPE, Cornelis Networks, Giga IO and Dell Technologies. Plus our friends at HPCwire, of course.
Next stop: Los Alamos National Laboratory in beautiful Los Alamos, NM, which is about 350 miles down the road.
Stay tuned for more news!