We’ve collaborated with SMART Modular Technologies to showcase our Leo CXL Smart Memory Controller in a real-world setup that delivers up to 2TB of additional CXL memory using SMART Modular CXL cards.
In the demo, we ran a series of Large Language Model inferencing tasks with FlexGen. Using CXL, the demo achieved 5.5x higher throughput and 90% GPU utilization. The demo also showcased COSMOS for unprecedented data center observability.
