Lenovo 4X67A97320 Thinksystem NVIDIA 2-WAY Bridge for H200 NVL - 4X67A97320

Lenovo 4X67A97320 ThinkSystem NVIDIA 2-Way NVLink Bridge for H200 NVL

Overview

The Lenovo 4X67A97320 is a 2-way NVLink bridge designed for NVIDIA H200 NVL GPU installations in Lenovo ThinkSystem servers. If you are provisioning a multi-GPU node and need the direct GPU-to-GPU interconnect that enables NVLink peer memory access and high-bandwidth inter-GPU traffic — without routing through the host CPU — this is the bridge component that physically makes that possible. At 10 lb, this is a substantive piece of hardware, not a passive bracket: it carries the NVLink connector stack that joins two H200 NVL GPUs into a unified compute fabric.

GPU interconnect hardware is easy to overlook in a BOM, but the wrong bridge — or a missing one — means your H200 NVL pair operates as two independent PCIe devices rather than an NVLink-bonded pair, with a measurable impact on workloads that depend on large-model weight sharing, collective communications (NCCL), or direct GPU memory access across devices.

Key Features

• 2-Way NVLink Configuration: Bridges two NVIDIA H200 NVL GPUs into a directly interconnected pair, enabling inter-GPU bandwidth and peer memory access that PCIe alone cannot deliver — critical for large-language-model inference and distributed training workloads where GPU-to-GPU data movement is on the critical path.
• H200 NVL-Specific Fit: Engineered for the H200 NVL form factor. Using a bridge intended for a different GPU generation or slot pitch risks improper seating and connector damage — this component matches the H200 NVL's physical and electrical interface.
• 10 lb Assembly: The 10 lb weight reflects the robust connector and retention hardware required to maintain reliable NVLink contact across server vibration and thermal cycling — a practical indicator of build quality versus lightweight passive adapters.
• ThinkSystem Platform Integration: Sourced and validated within the Lenovo ThinkSystem ecosystem, so it carries through Lenovo's server qualification process rather than being a generic third-party bridge.

Integration & Compatibility

The 4X67A97320 is intended for Lenovo ThinkSystem servers configured with NVIDIA H200 NVL GPUs. Verify your target server model supports a 2-way NVLink bridge topology in its GPU cage/riser configuration before ordering — not all ThinkSystem chassis that accept H200 NVL cards have the physical clearance or bracket positions for the bridge assembly. Consult the ThinkSystem product guide for your specific server SKU to confirm compatibility. Manufactured in Mexico (Country of Origin: MX).

Frequently Asked Questions

Q: What GPU does the Lenovo 4X67A97320 NVLink bridge support?

A: The 4X67A97320 is specifically designed for NVIDIA H200 NVL GPUs installed in Lenovo ThinkSystem servers. It is a 2-way bridge, meaning it interconnects two H200 NVL GPUs via NVLink.

Q: How many GPUs does the 4X67A97320 bridge connect?

A: It is a 2-way bridge, connecting two NVIDIA H200 NVL GPUs into a directly bonded NVLink pair.

Q: Is the 4X67A97320 compatible with NVIDIA H100 or other GPU generations?

A: Based on available evidence, this bridge is specified for the H200 NVL. Compatibility with other GPU generations (H100, A100, etc.) is not confirmed in the product evidence — do not substitute without consulting Lenovo's ThinkSystem configuration guides.

Q: What does the 4X67A97320 weigh?

A: The assembly weighs 10.00 lb, reflecting the robust connector retention hardware required for reliable NVLink contact in a server environment.

Q: Where is the Lenovo 4X67A97320 manufactured?

A: Country of Origin is Mexico (MX).

James Everett

When I spec a multi-GPU ThinkSystem node around NVIDIA H200 NVL cards, the 4X67A97320 is the bridge component that determines whether you actually get NVLink peer bandwidth between the two GPUs or whether they end up operating as isolated PCIe devices. That 10 lb assembly weight is telling — this is not a lightweight passive connector, it is a retention-heavy bridge built to maintain solid NVLink contact through the thermal and mechanical stresses of a running server.

Technical Highlights:

• 2-Way NVLink topology: Directly bonds two H200 NVL GPUs, enabling inter-GPU memory access without CPU involvement — the difference between NVLink-aware workloads running as intended versus hitting PCIe bandwidth bottlenecks.
• H200 NVL-specific geometry: Physical fit is generation-specific; the slot pitch and connector alignment on H200 NVL differ from prior GPU generations, so this is not a universal bridge you can reuse across GPU upgrades.
• 10 lb build weight: Heavier than generic alternatives, which in practice means more robust retention hardware — relevant in chassis that see vibration or frequent hot-aisle temperature swings.

Deployment Considerations:

• Confirm your specific ThinkSystem server SKU has the physical clearance and bracket positions for a 2-way bridge before the server ships — retrofitting a bridge in a tight GPU bay after initial build is significantly more disruptive than installing it during initial configuration.
• The bridge is H200 NVL-specific; do not assume it fits H100 SXM, H100 PCIe, or earlier A100 NVLink configurations — connector profiles and retention clips differ across generations.

This component belongs on the BOM for any ThinkSystem H200 NVL 2-GPU node where the workload depends on NVLink-scale inter-GPU bandwidth — large-model inference, distributed training, or HPC collective communications where GPU-to-GPU data movement is on the critical path.