NVIDIA switches scale AI with silicon photonics
NVIDIA’s Spectrum-X and Quantum-X silicon photonics-based network switches connect millions of GPUs, scaling AI compute. They achieve up to 1.6 Tbps per port and up to 400 Tbps aggregate bandwidth. NVIDIA reports the switch platforms use 4x fewer lasers for 3.5x better power efficiency, 63x greater signal integrity, 10x higher network resiliency at scale, and 1.3x faster deployment than conventional networks.
Spectrum-X Photonics Ethernet switches support 128 ports of 800 Gbps or 512 ports of 200 Gbps, delivering 100 Tbps of total bandwidth. A high-capacity variant offers 512 ports of 800 Gbps or 2048 ports of 200 Gbps, for a total throughput of 400 Tbps.
Quantum-X Photonics InfiniBand switches provide 144 ports of 800 Gbps, achieved using 200 Gbps SerDes per port. Built-in liquid cooling keeps the onboard silicon photonics from overheating. According to NVIDIA, Quantum-X Photonics switches are 2x faster and offer 5x higher scalability for AI compute fabrics compared to the previous generation.
NVIDIA’s silicon photonics ecosystem includes collaborations with TSMC, Coherent, Corning, Foxconn, Lumentum, and SENKO to develop an integrated silicon-optics process and robust supply chain.
Quantum-X Photonics InfiniBand switches are expected to be available later this year. Spectrum-X Photonics Ethernet switches will be coming in 2026 from leading infrastructure and system vendors. Learn more about NVIDIA’s silicon photonics here.
Find more datasheets on products like this one at Datasheets.com, searchable by category, part #, description, manufacturer, and more.
The post NVIDIA switches scale AI with silicon photonics appeared first on EDN.
