We develop Morphlux, a server-scale programmable photonic fabric to interconnect accelerators within servers. We show that augmenting state-of-the-art torus-based ML data-centers with Morphlux can improve the bandwidth of tenant compute allocations by up to 66%, reduce compute fragmentation by up to 70%, and minimize the blast radius of chip failures. We develop a novel end-to-end hardware prototype of Morphlux to demonstrate these performance benefits which translate to 1.72X improvement in training throughput of ML models. By rapidly programming the server-scale fabric in our hardware testbed, Morphlux can replace a failed accelerator chip with a healthy one in 1.2 seconds.
@article{arxiv.2508.03674,
title = {Morphlux: Transforming Torus Fabrics for Efficient Multi-tenant ML},
author = {Abhishek Vijaya Kumar and Eric Ding and Arjun Devraj and Darius Bunandar and Rachee Singh},
journal= {arXiv preprint arXiv:2508.03674},
year = {2025}
}