NVIDIA 900-9X7AX-004NMC0 | ConnectX-7 | 400GbE QSFP112 | PCIe Gen5 x16 | RoCEv2 | SR-IOV | GPUDirect | PTP 1588v2 AI Hot‑swappable and interoperable with IEEE

Hot‑swappable and interoperable with IEEE 802

080 CUDA cores

and is the unmanaged variant requiring an external Subnet Manager (e

low-profile form factor

this switch is engineered for extreme-scale AI training clusters

NVIDIA 900-9X7AX-004NMC0 | ConnectX-7 | 400GbE QSFP112 | PCIe Gen5 x16 | RoCEv2 | SR-IOV | GPUDirect | PTP 1588v2 AI Hot‑swappable and interoperable with IEEENext generation Ethernet adapter offering up to 400GbE over QSFP112 for dense compute, storage, and AI fabrics. With PCIe Gen5 x16 and extensive offloads (NVMe oF TCP, OVS, TLS IPsec, RoCEv2), it delivers extreme throughput while reducing CPU overhead. Supports IEEE 1588v2 PTP, advanced telemetry, and breakout modes such as 4x100G where supported. Compatible with QSFP family optics and DAC AOC matched to the selected speed.