Limited Time Offer: Save $350 on your Ai Cluster → Shop Now

--

Days

--

Hours

--

Mins

--

Secs

Sale Ended

NVIDIA Quantum QM8790 HDR InfiniBand Switch, 40× 200G

Vendor:NVIDIA

NVIDIA Quantum QM8790 HDR InfiniBand Switch, 40× 200G

$20,441.00

The NVIDIA Quantum QM8790 is a 1U HDR InfiniBand switch built for AI, HPC, and high-performance data center fabrics. It provides 40 non-blocking 200Gb/s QSFP56 ports, up to 16Tb/s of aggregate throughput, and NVIDIA SHARP in-network computing to accelerate communication-intensive workloads. Adaptive routing, congestion control, and redundant hot-swappable power and cooling support reliable, low-latency cluster connectivity

Out of stock
Product datasheet

Pairs well with

30-Day AISuperStore Guarantee 

TECHNOLOGY AT A GLANCE

Explore the capabilities

16Tb/s HDR InfiniBand fabric

Forty non-blocking 200Gb/s ports deliver high-bandwidth connectivity for compute, storage, and cluster fabrics.

Accelerate collective operations

NVIDIA SHARP performs supported reductions and aggregations within the network to accelerate distributed workloads.

40× 200G or 80× 100G

QSFP56 ports support HDR 200G connectivity or splitter-cable deployments with up to 80 HDR100 links.

1U cluster-ready infrastructure

Hot-swappable redundant power and N+1 cooling support serviceable, high-density rack deployment.

Scale your AI fabric with HDR InfiniBand

The Quantum QM8790 brings dense 200G HDR InfiniBand connectivity and in-network computing into a compact 1U platform. Build low-latency cluster fabrics for AI, HPC, simulation, and high-performance data processing.

Dense 200G connectivity for cluster fabrics

The QM8790 concentrates 40 non-blocking HDR InfiniBand links into a compact 1U system, with the bandwidth and intelligence needed for modern AI and HPC clusters.

  • 40× QSFP56 ports: Connect compatible compute, storage, and fabric infrastructure at up to 200Gb/s per port.
  • HDR100 port splitting: Use compatible splitter cables to support up to 80× 100Gb/s InfiniBand connections where higher endpoint density is needed.
  • In-network acceleration: NVIDIA SHARP helps process supported collective operations within the fabric, reducing unnecessary data movement between nodes.
  • Externally managed deployment: The MQM8790-HS2R is designed for externally managed InfiniBand fabrics and includes an I²C management connection.

TECHNICAL DETAILS

Built into the system

40-Port HDR 200G InfiniBand Fabric

The QM8790 provides 40 non-blocking QSFP56 ports at up to 200Gb/s each. Its 16Tb/s aggregate throughput supports high-bandwidth communication between compute nodes, storage, and fabric infrastructure.

Flexible HDR100 Connectivity

With compatible splitter cables, the QM8790 supports up to 80 HDR100 100Gb/s InfiniBand links. This flexibility helps create high-density top-of-rack deployments while retaining a clear path to 200G connectivity.

NVIDIA SHARP In-Network Computing

NVIDIA SHARP performs supported aggregation and reduction operations as data moves through the network. This reduces repeated endpoint-to-endpoint data exchanges and can accelerate collective communication patterns used by MPI, SHMEM, AI, and scientific workloads.

Low-Latency Cluster Communication

Sub-130ns port-to-port latency, adaptive routing, congestion control, and enhanced virtual-lane mapping help reduce bottlenecks and sustain effective bandwidth across modern cluster topologies.

Externally Managed Fabric Design

The MQM8790-HS2R is an externally managed switch, intended for use with compatible external InfiniBand fabric-management tools. It includes an I²C management port and a system reset button.

Reverse-Airflow 1U Infrastructure

This model uses connector-to-power, or reverse, airflow. Dual 1+1 hot-swappable AC power supplies and N+1 hot-swappable fans support reliable operation and easier serviceability in high-density racks.