FAQ Friday

I gotta tell you, we used to get this question a lot more because, when we started the company six years ago, we really believed in power efficiency to deliver maximum density.

But not everyone saw it that way. Now fast forward a couple of years and of course with all of the compute and AI build out, people are much more power conscious because their DCs are out of power. So they wanna understand how do I get the most amount of compute density per watt? And the answers might surprise you.

One of them is super basic which is to have a DC busbar, big honkin' piece of copper up and down the back of the rack, with rectifiers in the middle. Get rid of this nonsense of AC power supplies, two AC power supplies per server, power cables, fans, it’s an absolute mess. Going to an actual DC busbar, we know we wanted to go do that. Once you are at that level of rack design, you are free to do some things, and in particular, we changed the geometry. So instead of having these kind of, this, tyranny of the 1U, 2U server, we expanded that out.

We actually made the these things a little bit higher. We made them narrower, and then we made the whole thing a lot taller. Because if you go into a lot of DCs and you look up, there’s often unused space, there’s empty space. Why? Because you run out of power when the rack is only halfway built. Ironically, if you want to get the best density, compute density, that is to say you want to get the most amount of compute on that floor tile, you actually need to make the thing physically less dense.

By doing that, you open up airflow. You make it much more efficient. With that DC busbar in the back and power management up and down the rack, you add all those up and the differences become really, really significant. We get twice the power efficiency in an Oxide rack. That is to say you can get twice the compute cores in an Oxide rack that you can get with a traditional rack and stack 1U, 2U kind of approach.

So these differences are really, really significant. And for a, for a DC operator, for someone who needs to, to deploy either in a colo or their own data center, it makes a real difference because now you can actually compress that compute footprint, open up that footprint for other things you may want to use, or you can actually fit some new compute into your data center. Either way, power really, really matters in the data center today more than ever, and that’s why we took the first principles approach at Oxide. Thanks and see you next time.