d-Matrix has unveiled d-Matrix Squadron, a scalable fleet of multi-rack computing systems designed to deliver megawatt-level inference performance with exceptional energy efficiency. Built to execute AI models at ultra-high speeds, the solution brings powerful, energy-conscious inference computing to large-scale data center environments.
Benefiting from d-Matrix’s unique memory-centric compute architecture, Squadron clusters are ideal for AI labs, enterprises, and sovereign organizations with latency-sensitive AI applications including coding, personal agents, cybersecurity, and high-frequency trading.
Developed in partnership with Arista, Broadcom, and Supermicro, Squadron has already been deployed by d-Matrix in its Arizona data center for internal research and development and is now shipping to priority customers for deployment in their own data centers. Additionally, Squadron’s building blocks — d-Matrix Corsair servers and d-Matrix Corsair SquadRacks — are now generally available.
“The journey to enable customers with our memory-centric compute advantage coincides with d-Matrix’s evolution from a specialized chip company to a data center compute infrastructure business. With Squadron shipping, and our Corsair servers and SquadRacks now generally available, the wait is over. Customers now have access to the world’s most efficient data center-scale inference infrastructure available for building and ramping their ultra-low-latency AI applications.”
– Sid Sheth, Founder and CEO of d-Matrix, who unveiled Squadron during his keynote on the opening day of OCP Global Summit at the San Jose Convention Center
“Bringing next-generation AI infrastructure to market at scale requires deep engineering, systems integration, and manufacturing expertise. Our collaboration with d-Matrix demonstrates Supermicro’s ability to deliver innovative AI technologies through our proven Data Center Building Block Solutions (DCBBS). We enable customers to deploy efficient, high-performance inference infrastructure while providing accelerated time-to-online supporting the most advanced AI workloads and heterogeneous computing environments.”
-Charles Liang, president and CEO of Supermicro
Squadron is available in a wide range of configurations, scaling up to 16 d-Matrix SquadRacks — d-Matrix single eight-server racks — into a fully integrated compute cluster with up to 1,024 d-Matrix Corsair XPUs. The clusters are seamlessly interconnected with d-Matrix JetStream NICs and optimized for ease of deployment with d-Matrix Aviator software. Because Squadron clusters are air-cooled and built on industry standards, customers can integrate them directly into the data centers they already operate, bypassing the need for liquid cooling or costly and time-consuming data center buildouts.
Corsair servers and SquadRack rack-scale systems are built with Corsair XPUs, JetStream NICs, and Aviator software. Customers can deploy a single Corsair server, a full SquadRack, or scale all the way to a multi-rack Squadron cluster, all on the same air-cooled, industry-standard architecture that drops into existing data centers.
Optimized to deploy AI with speed and efficiency, Squadron clusters can run full models exclusively on Corsair XPUs and are also designed for heterogeneous environments where work is split between GPUs and XPUs, an approach called disaggregated compute.
During his OCP keynote, Sheth revealed that d-Matrix has successfully implemented Attention-FFN Disaggregation (AFD) using its own Squadron system paired with GPU servers.
This work demonstrated that, when offloading Experts/FFN to Corsair racks, the heterogeneous system was able to take advantage of Corsair’s efficiency attributes to generate 845 tokens per second per user. With 64 concurrent users, the heterogeneous AFD setup delivered a total throughput of 20,415 tokens per second.
While AFD has been discussed across the industry in theory as a promising disaggregating compute application, d-Matrix is the first to show it running on real silicon that’s available today.
