Elastic compute fabrics for training.
Active work includes topology-aware placement, mixed-precision collectives, and inference schedulers that co-locate cache-hot workloads. The lab’s continental inference case study grew from operating these systems under real traffic.
Benchmark farms spanning GPU generations, fault-injection harnesses, and shared kernel libraries. Researchers and platform engineers pair on the same queues.
Winning ideas become Inference Infrastructure modules and Distributed Cloud features. The lab exists to collapse the gap between systems papers and production clusters.
Distributed Computing Lab shares compute, evaluation, and publication tooling with every other Zansoc lab — so discoveries compound across the stack.
Visit all labs