Mythic builds AI compute platforms around large numbers of small, energy-efficient accelerators rather than a handful of very large ones. Serving frontier-scale models on that kind of architecture pushes the hardest engineering problems out of the die and into the rack: how many thousands of devices are packaged, connected, powered, and cooled in a very small volume. We are looking for a System Architect to own the rack-level reference design for these platforms.
You will lead rack and platform pathfinding across board and chassis layout, scale-up and scale-out fabric topology, connector and cable plant, power delivery, and the cooling required at high rack densities. None of it can be settled apart from software: how a model is partitioned across devices sets the bandwidth required between layers, and the bandwidth the fabric can economically carry sets which partitionings are viable. The role sits at the intersection of electrical, mechanical, and software engineering, and the goal is an architecture quite different from today’s GPU racks that delivers a comparable result at rack level.