Servers and accelerators.

Technologies

• Short-depth and ruggedised edge servers Servers sized for telecom cabinets with front access, wide temperature range and DC power.
• GPUs General-purpose accelerators for AI inference, training and software-defined RAN.
• NPUs and inference accelerators Low-power accelerators for vision and sensor models at the device and far edge.
• FPGAs Reconfigurable logic for L1 acceleration, signal processing and deterministic pipelines.
• DPUs and SmartNICs Network processors that offload packet processing, storage and security from host CPUs.
• Memory and interconnect High-bandwidth memory and coherent interconnects that set model size and throughput at the edge.
• Chiplets and die-to-die interconnect Modular silicon built from chiplets, changing how edge accelerators are sourced and specified.
• Out-of-band management Standard remote management of servers for power, firmware and health at unattended sites.