Nvidia has introduced NVLink Fusion at Computex, allowing its high-speed interconnect technology to be used with custom CPUs and non-Nvidia accelerators. The new technology promises significantly higher bandwidth for CPU-to-GPU communications compared to PCIe 5.0, though it remains exclusive to Nvidia's ecosystem. Meanwhile, Nvidia launched its DGX Cloud Lepton platform for GPU workload deployment, likening it to a ridesharing app for developers seeking GPU resources.
NVIDIA NVLink™ Fusion offers advanced AI scale-up and scale-out performance by integrating NVIDIA technology with semi-custom ASICs or CPUs. It facilitates rapid communication among accelerators, enables hyperscalers to deploy custom silicon efficiently, and maintains a robust ecosystem for enhanced AI infrastructure management.