Top
Author
Published
24 Jun 2026Form Number
LP2452PDF size
19 pages, 2.0 MBSubscribed to LP2452.
Thank you for your feedback.
Abstract
NVIDIA InfiniBand switching continues to set the standard for high-performance, low-latency interconnects in modern data centers, powering demanding workloads such as AI training, HPC simulations, and large-scale analytics. With successive generations, from HDR to NDR and now XDR, these switches deliver increasing bandwidth, radix, and efficiency while maintaining the performance and advanced in-network computing capabilities required for tightly coupled clusters.
This document introduces the switches and provides the Lenovo ordering information. It is intended for technical specialists, sales specialists, sales engineers, IT architects, and other IT professionals who want to learn more about the NVIDIA InfiniBand Switches and consider their use in IT solutions.
Introduction
NVIDIA InfiniBand switches deliver industry-leading, low-latency, high-bandwidth interconnects designed to power modern AI, high-performance computing (HPC), and data-intensive enterprise workloads. From HDR at 200 Gb/s to the latest Quantum NDR and XDR generations, these switches provide the scalable fabric foundation required for demanding applications such as large-scale AI training, real-time analytics, and scientific simulation. With advanced features such as adaptive routing, congestion control, and hardware-accelerated collective operations, NVIDIA InfiniBand ensures predictable performance and optimal resource utilization across clustered environments.
This product guide covers the following NVIDIA InfiniBand switch offerings, most available in both Port Side Exhaust (PSE) and opposite Port Side Exhaust (oPSE) airflow configurations:
- Mellanox QM8700 – 40-port managed HDR 200 Gb/s switch (PSE and oPSE variants)
- Mellanox QM8790 – 40-port unmanaged HDR 200 Gb/s switch (PSE and oPSE variants)
- NVIDIA QM9700 – 64-port managed Quantum NDR switch (PSE and oPSE variants)
- NVIDIA QM9790 – 64-port unmanaged Quantum NDR switch (PSE and oPSE variants)
- NVIDIA Q3200-RA – 72-port managed Quantum XDR switch (oPSE variant)
- NVIDIA Q3400-RA – 144-port managed Quantum XDR switch (oPSE variant)
Together, these switches span multiple generations of InfiniBand technology and management models, enabling customers to design fabrics that balance performance, scalability, and operational simplicity. Whether deploying edge-scale clusters or building exascale-class AI infrastructures, this portfolio provides the flexibility to meet a wide range of workload and data center requirements.
Figure 1. NVIDIA QM9700 64-Port Managed Quantum NDR InfiniBand Switch
Did you know?
NVIDIA InfiniBand switches do far more than just move data—they actively accelerate application performance through in‑network computing. Using technologies such as SHARP (Scalable Hierarchical Aggregation and Reduction Protocol), these switches can perform collective operations like reductions directly in the network fabric, significantly reducing the amount of data that must traverse the cluster and offloading work from CPUs and GPUs. This capability becomes increasingly important at scale, where communication overhead can dominate workload runtime; by processing data in transit and enabling advanced routing, congestion control, and self-healing behaviors, modern Quantum, Quantum‑2, and Quantum‑3 InfiniBand switches help deliver near-linear scalability for AI training and HPC workloads.
Ordering information
The following table lists the NVIDIA InfiniBand switches orderable from Lenovo.
* PSE = Port Side Exhaust; oPSE = Opposite Port Side Exhaust
Lenovo EveryScale
Lenovo EveryScale is a framework for designing, manufacturing, integrating, and delivering data center solutions, with a focus on High Performance Computing (HPC), Technical Computing, and Artificial Intelligence (AI) environments.
With EveryScale, Lenovo pre-tests and validates a wide selection of Lenovo and third-party components, including NVIDIA InfiniBand switches, to create a "Best Recipe" of components and firmware levels that work seamlessly together as a solution. From this testing, customers can be confident that these best-practice solutions run optimally for their workloads and are tailored to their specific needs.
Lenovo EveryScale hardware is pre-integrated, pre-cabled, and pre-loaded, optionally with an OS image, and is tested at the rack level during manufacturing. This process ensures reliable delivery and minimizes installation time in the customer data center.
With EveryScale, customers can focus their efforts on maximizing business value instead of consuming valuable resources to design, optimize, install, and support the infrastructure required to meet business demands.
Learn more by reading the Lenovo EveryScale product guide.
Model summary
The NVIDIA switches offered by Lenovo are as follows:
- Mellanox QM8700 Managed HDR InfiniBand
- Mellanox QM8790 Unmanaged HDR InfiniBand
- NVIDIA QM9700 Managed NDR InfiniBand
- NVIDIA QM9790 Unmanaged NDR InfiniBand
- NVIDIA Q3200-RA Managed XDR InfiniBand
- NVIDIA Q3400-RA Managed XDR InfiniBand
Mellanox QM8700 Managed HDR InfiniBand
The Mellanox QM8700 is a 1U, 40-port managed HDR InfiniBand switch based on the NVIDIA Quantum architecture, delivering up to 200 Gb/s per port and 16 Tb/s of non-blocking throughput. Designed for top-of-rack or leaf deployments, it enables scalable AI and HPC fabrics with ultra-low latency and advanced in-network computing capabilities, including SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) acceleration to offload collective operations from CPUs. Features such as adaptive routing, congestion control, and advanced topology support help ensure efficient fabric utilization and consistent application performance.
For more information, see the NVIDIA QM8700 datasheet and QM8700 user manual.
Mellanox QM8790 Unmanaged HDR InfiniBand
The Mellanox QM8790 provides the same 40-port HDR 200 Gb/s performance and 16 Tb/s switching capacity as the QM8700, but in an externally managed (unmanaged) configuration. Built on the same Quantum switch silicon, it delivers ultra-low latency and high message rates optimized for large-scale AI and HPC clusters. With SHARP-based in-network computing, adaptive routing, and congestion management, the QM8790 ensures efficient communication across distributed workloads, making it ideal for environments where fabric management is handled externally or centrally.
For more information, see the NVIDIA QM8790 datasheet and QM8790 user manual.
NVIDIA QM9700 Managed NDR InfiniBand
The NVIDIA QM9700 is a 1U managed switch based on the Quantum-2 architecture, delivering 64 ports of NDR 400 Gb/s InfiniBand connectivity and up to 51.2 Tb/s of aggregated throughput. Designed for next-generation AI and exascale HPC environments, it integrates advanced features such as SHARPv3 in-network computing, adaptive routing, and self-healing fabric capabilities to maximize throughput and minimize latency. With high radix density and support for multiple topologies, the QM9700 enables efficient scale-out designs for large GPU clusters and data-intensive workloads.
For more information, see the NVIDIA QM9700 datasheet and QM9700 user manual.
NVIDIA QM9790 Unmanaged NDR InfiniBand
The NVIDIA QM9790 delivers the same Quantum-2 NDR 400 Gb/s performance as the QM9700, with 64 ports and 51.2 Tb/s switching capacity, but is designed for externally managed deployments. It retains all core architectural benefits, including SHARPv3 acceleration, RDMA support, adaptive routing, and congestion control, ensuring high efficiency and scalability for AI and HPC clusters. This model is well suited for environments that rely on centralized fabric management while still requiring state-of-the-art performance and scalability.
For more information, see the NVIDIA QM9790 datasheet and QM9790 user manual.
NVIDIA Q3200-RA Managed XDR InfiniBand
The NVIDIA Q3200-RA is a Quantum-3–based XDR InfiniBand switch delivering up to 800 Gb/s per port in a high-density 2U design with 72 ports (across dual internally integrated switches). It is engineered for hyperscale AI and HPC environments, offering massive bandwidth, ultra-low latency, and advanced in-network computing capabilities, along with adaptive routing and telemetry-driven congestion control. This design enables efficient scaling of next-generation GPU clusters and supports very high port density for modern AI fabrics.
Figure 10. Q3200-RA front view
For more information, see the NVIDIA Q3200-RA datasheet and Q3200-RA user manual.
NVIDIA Q3400-RA Managed XDR InfiniBand
The NVIDIA Q3400-RA is a flagship 4U Quantum-3 XDR InfiniBand switch that provides 144 ports of 800 Gb/s connectivity, delivering exceptional radix and scalability for the largest AI and HPC deployments. It incorporates next-generation in-network computing, adaptive routing, and self-healing capabilities to optimize large-scale distributed workloads. Its extremely high port density and throughput make it ideal for spine or core layers in exascale-class AI infrastructures.
Figure 12. Q3400-RA front view
For more information, see the NVIDIA Q3400-RA datasheet and Q3400-RA user manual.
Specifications
The following table shows the specifications of each switch. There are two variants for some switches:
- PSE (Port Side Exhaust): Airflow is Back-to-front - Air enters from power side and exits at port side
- oPSE (Opposite / reversed airflow): Airflow is Front-to-back - Air enters from port side and exits at power side
* QM8790 and QM9790 support external management
** See the Fabric Manager section
Transceivers
The following table lists the supported transceivers.
Networking cables
The following table lists the supported cables.
In this section:
Optical cables
Copper cables
RJ45 cables
Fabric Manager
For InfiniBand networks to operate, there must always be a Fabric Manager present on the network providing subnet management functionality. The QM8700 Managed HDR InfiniBand Switch and QM9700 Managed NDR InfiniBand Switch come with an on-board Fabric Manager that can be used for this purpose. For all other switches (and as an alternative for the QM8700 and QM9700), an external Fabric Manager will need to be installed on a server that is connected to the InfiniBand fabric using an InfiniBand adapter.
There are two software options available for the Fabric Manager:
- OpenSM
- NVIDIA Unified Fabric Manager
OpenSM is included with the NVIDIA DOCA software package. OpenSM discovers, initializes, and configures InfiniBand fabrics, managing routing paths and switch forwarding tables. It runs in the background to ensure continuous communication between servers and storage. For more information, see the NVIDA DOCA help center and search on OpenSM.
NVIDIA Unified Fabric Manager (UFM) is InfiniBand networking management software that combines enhanced, real-time network telemetry with fabric visibility and control to support scale-out InfiniBand data centers. The two UFM offerings are available from Lenovo:
- UFM Telemetry for Real-Time Monitoring
The UFM Telemetry platform provides network validation tools to monitor network performance and conditions, capturing and streaming rich real-time network telemetry information, application workload usage, and system configuration to an on-premises or cloud-based database for further analysis.
- UFM Enterprise for Fabric Visibility and Control
The UFM Enterprise platform combines the benefits of UFM Telemetry with enhanced network monitoring and management. It performs automated network discovery and provisioning, traffic monitoring, and congestion discovery. It also enables job schedule provisioning and integrates with industry-leading job schedulers and cloud and cluster managers, including Slurm and Platform Load Sharing Facility (LSF).
The following table lists the UFM subscription licenses available from Lenovo.
For more information on NVIDIA UFM, see the NVIDIA Unified Fabric Manager product page.
Rack mount kit
The following table lists the optional rack mount kit for the NVIDIA switches.
Tip: The Q3200-RA and Q3400-RA both include a rack mount kit as standard:
- Q3200-RA: Includes Rack mount kit 930-9SKIT-00L0-00C
- Q3400-RA: Includes Rack mount kit 930-9SKIT-00L0-00B
Power cables
The following table lists the power cables that can be ordered with the NVIDIA switches.
Tip: The QM8700 and QM8790 use C13 power cables; all other switches use C15 cables.
Operating environment
The NVIDIA switches have the following operating characteristics:
- Normal operating temperature range (sustained): 0°C to +40°C (32°F to +104°F)
- Shipping and storage temperature range: -40°C to +70°C (-40°F to +158°F)
- Normal operating humidity: 10% to 85%, non-condensing
- Non-operating humidity: 10% to 90%, non-condensing
Related publications and links
For more information, see these resources:
- Lenovo EveryScale product guide
https://lenovopress.lenovo.com/lp0900-lenovo-everyscale-lesi - NVIDIA Quantum InfiniBand Switches
https://www.nvidia.com/en-us/networking/infiniband-switching/ - NVIDIA QM8700 and QM8790 datasheet
https://nvdam.widen.net/s/zmbw7rdjml/infiniband-qm8700-datasheet-us-nvidia-1746790-r12-web - NVIDIA QM9700 and QM9790 datasheet
https://nvdam.widen.net/s/k8sqcr6gzb/infiniband-quantum-2-qm9700-series-datasheet-us-nvidia-1751454-r8-web - NVIDIA Q3200-RA and Q3400-RA datasheet
https://nvdam.widen.net/s/nfdzskhmnc/infiniband-datasheet-quantum-family-3231555 - NVIDIA Quantum User Manuals (select the "InfiniBand ToR/Fixed-Conf. Switches" link)
https://networking-docs.nvidia.com/switches#nvidiatab-hw-documentation - NVIDIA Unified Fabric Manager
https://www.nvidia.com/en-us/networking/infiniband/ufm/
Trademarks
Lenovo and the Lenovo logo are trademarks or registered trademarks of Lenovo in the United States, other countries, or both. A current list of Lenovo trademarks is available on the Web at https://www.lenovo.com/us/en/legal/copytrade/.
The following terms are trademarks of Lenovo in the United States, other countries, or both:
Lenovo®
ThinkAgile®
ThinkSystem®
The following terms are trademarks of other companies:
LSF® is a trademark of IBM in the United States, other countries, or both.
Other company, product, or service names may be trademarks or service marks of others.
Configure and Buy
Please select a locale
Full Change History
Course Detail
Employees Only Content
The content in this document with a is only visible to employees who are logged in. Logon using your Lenovo ITcode and password via Lenovo single-signon (SSO).
The author of the document has determined that this content is classified as Lenovo Internal and should not be normally be made available to people who are not employees or contractors. This includes partners, customers, and competitors. The reasons may vary and you should reach out to the authors of the document for clarification, if needed. Be cautious about sharing this content with others as it may contain sensitive information.
Any visitor to the Lenovo Press web site who is not logged on will not be able to see this employee-only content. This content is excluded from search engine indexes and will not appear in any search results.
For all users, including logged-in employees, this employee-only content does not appear in the PDF version of this document.
This functionality is cookie based. The web site will normally remember your login state between browser sessions, however, if you clear cookies at the end of a session or work in an Incognito/Private browser window, then you will need to log in each time.
If you have any questions about this feature of the Lenovo Press web, please email David Watts at dwatts@lenovo.com.
