Description
The NVIDIA Quantum-2 QM9790 is a high-performance InfiniBand switch purpose-built for large-scale AI training clusters and high-performance computing (HPC) fabrics. Featuring 64 ports of 400Gb/s NDR InfiniBand in a compact 1U chassis, the QM9790 delivers an aggregate bidirectional throughput of 51.2 Tb/s and packet processing capacity of 66.5 billion packets per second (Bpps).
Built on Quantum-2 architecture, the QM9790 supports SHARPv3 in-network computing, adaptive routing, and advanced congestion control, enabling ultra-low latency and efficient collective operations for distributed AI workloads. It is validated for DGX SuperPOD architectures, making it ideal for hyperscale AI deployments, generative model training, and scientific simulations.
Specification Table — 400Gb/s NDR InfiniBand Switch
Key Feature | Specification |
---|---|
Port Configuration | 64 × 400Gb/s NDR InfiniBand (OSFP) |
Switching Capacity | 51.2 Tb/s bidirectional throughput |
Packet Processing | ~66.5 Bpps |
Latency | Sub-microsecond switching |
Form Factor | 1U rackmount, 14.5 kg |
Power Consumption | Typical: 720W (with passive cables) |
Management Options | Onboard subnet manager (up to 2,000 nodes) or external UFM control |
Protocol Support | Native InfiniBand, RDMA, SHARPv3, adaptive routing, congestion control |
AI/HPC Optimization | SHARP offloads for collective ops, self-healing networking, DGX SuperPOD compatibility |
Supported Topologies | Fat Tree, SlimFly, DragonFly+, Torus |
Reviews
There are no reviews yet.