900-9X662-0073-ST0 Nvidia Connectx-6 Card 25GBE 2Port PCIe 4.0 X8 SFP28 Network Adapter

• "900-9X662-0073-ST0"
• "Nvidia ConnectX-6 Lx"
• "25GbE SFP28 PCIe 4.0 x8"
• "dual-port 25G NIC"
• "ConnectX-6 Lx 25GbE Smart NIC"

Long-tail phrases and buyer intent snippets

• "Buy Nvidia ConnectX-6 Lx 900-9X662-0073-ST0 25GbE card"
• "ConnectX-6 Lx 25G SFP28 driver and firmware download"
• "ConnectX-6 Lx vs ConnectX-6 Dx comparison for virtualization"
• "Best SFP28 transceivers for ConnectX-6 Lx 25GbE"

Nvidia ConnectX-6 Lx Overview of the 900-9X662-0073-ST0 Category

The 900-9X662-0073-ST0 listing refers to the Nvidia/Mellanox ConnectX-6 Lx family of high-performance Ethernet SmartNICs, specifically a dual-port 25GbE SFP28 adapter in a PCIe Gen4 x8 form factor. This category page focuses on the ConnectX-6 Lx 25GbE dual-port PCIe cards — their technical capabilities, typical deployments, key compatibility notes, and buying considerations for data centers, cloud platforms, telco infrastructures, and high-performance workstations.

These adapters are designed to deliver predictable low-latency networking, hardware offloads for common network functions, and broad vendor support for modern server platforms. The product SKU 900-9X662-0073-ST0 maps to the MCX631102AS-ADAT device variant and is commonly sold as a tall-bracket internal PCIe card with two SFP28 ports that can operate at 25GbE per port or be reconfigured for a single 50GbE port in some environments.

Performance & Offloads

ConnectX-6 Lx devices emphasize a mix of raw throughput and smart offloads: kernel bypass technologies, Large Receive Offload (LRO), Large Send Offload (LSO), Receive Side Scaling (RSS), and advanced stateless/stateful offloads that reduce CPU cycles for networking. These offloads are especially valuable in virtualization, containerized microservices, storage networking (iSCSI/NVMe over TCP), and software-defined networking (SDN) where CPU cycles are better spent on application logic than packet handling.

Cost-to-Performance Sweet Spot

25GbE is widely adopted in modern leaf/spine data center fabrics because it allows a significant step up from 10GbE without the cost and port density challenges of 40/100GbE. The ConnectX-6 Lx family was specifically engineered to offer the economics of 25GbE while providing Smart NIC features that improve utilization and reduce server CPU load — making it an attractive option for cloud providers, telco edge sites, and enterprise virtualization clusters.

Compatibility & Ecosystem

ConnectX-6 cards are supported by a wide range of server vendors (OEM-qualified for many Cisco, Dell EMC, HPE, Lenovo server families), and work with standard operating systems such as mainstream Linux distributions and Windows Server with Nvidia /Mellanox drivers. PCIe Gen4 compatibility also future-proofs system upgrades where available.

Category Subsections Detailed Deep Dives

Technical Architecture (h4 variants below explain elements)

NIC ASIC & MAC

The ConnectX-6 Lx integrates a high-performance Ethernet MAC and multi-core network processor that offloads packet processing and provides telemetry. The card handles packet classification, steering, and advanced filtering in hardware when configured, which improves throughput and reliability under heavy loads.

Host Interface & DMA

Using PCIe Gen4 x8, the adapter provides wide, low-latency DMA channels between NIC buffers and host memory. This reduces copy overheads and increases achievable line rates for sustained flows — particularly useful for storage traffic and high-message-rate applications like trading, real-time analytics, or distributed databases.

SFP28 Optics & Cabling

SFP28 supports both DAC (direct attach copper) and fiber optics (SR/LR transceivers) enabling short-reach, long-reach, and passive copper connectivity. Choosing the right transceivers (SFP28 SR for short multimode fiber; LR for single mode long distances) depends on your rack/inter-rack topology and link budget.

Use Cases & Deployment Scenarios

Hyperscale & Cloud

High VM density environments benefit from SR-IOV and hardware offloads: ConnectX-6 Lx reduces per-VM CPU overhead, enabling more effective consolidation and lower TCO.

Edge & Telco

25GbE is a practical choice at the edge where 100GbE ports may be too expensive. ConnectX-6 Lx supports telecom workloads that need predictable latency, secure boot options, and sometimes host management capabilities in OCP variants.

Storage & NVMe-over-TCP

Transporting storage traffic over 25GbE is increasingly common for NVMe-over-TCP and SMB3/iSCSI; hardware acceleration for checksum and segmentation offloads helps achieve high throughput while reducing CPU usage.

High-Performance Computing (HPC)

While HPC often targets 100GbE or InfiniBand, many clusters use 25GbE for specific tiers or for control/management networks; when combined with RoCE and tuned drivers, ConnectX-6 Lx can play a role in hybrid HPC fabrics.

Buying Guide What To Check Before You Add to Cart

1. Confirm Exact OPN / PSID

SKUs in the ConnectX-6 family have subtle sub-variants (crypto enabled vs disabled, secure boot, bracket type, OCP vs HHHL). The 900-9X662-0073-ST0 identifies a particular ConnectX-6 Lx EN adapter — verify the PSID on firmware compatibility pages before bulk purchasing to ensure match with your image/firmware.

2. Server Slot & PCIe Gen Compatibility

Make sure your target server has a free PCIe x8 Gen4 slot (or a Gen3 x8 will work with reduced headroom). On motherboards where Gen4 is shared between slots, verify BIOS settings to avoid unintended bottlenecks.

3. Transceivers & Cable Choices

• Short runs: SFP28 DAC cables (passive or active copper) are cost-effective for top-of-rack to server links.
• Fiber: Select SFP28 SR/LR optics based on distance; verify vendor interoperability lists if mixing manufacturers.

4. Firmware & Driver Lifecycle

Check for the latest firmware bundle and driver compatibility matrix. Nvidia publishes release notes and maintenance bundles — staying current reduces vulnerability exposure and ensures best performance.

Performance Tuning Tips

• Set jumbo MTU where appropriate for storage or big data flows to reduce packet overhead.
• Enable Receive Side Scaling (RSS) or RPS-based steering to distribute interrupts across CPUs.
• Tune for latency or throughput depending on application profile — trading off CPU use vs latency is common.

Comparisons & Alternatives

ConnectX-6 Lx vs ConnectX-6 Dx

ConnectX-6 Dx is the higher-end Smart NIC variant that often includes more comprehensive crypto offloads and expanded feature sets aimed at security-sensitive, highly accelerated workloads (for example, TLS offload). ConnectX-6 Lx targets cost-efficient 25GbE deployments with many of the performance and offload benefits but may lack some Dx crypto features depending on SKU. Choose Dx if you need integrated crypto acceleration or specialized Smart NIC P4/runtime features.

Additional Resources & Manufacturer References

For authoritative technical details, firmware compatibility, and driver downloads consult Nvidia official ConnectX-6 Lx documentation and the product datasheets/firmware release notes. OEM qualification documents (for Cisco, Dell, Lenovo) are useful when validating support in a specific server model. These manufacturer resources include SKU tables and feature flags that directly map to the 900-9X662-0073-ST0 OPN.

Frequently Requested Data Points

It’s primarily a dual 25GbE SFP28 adapter. Some ConnectX-6 Lx cards can be configured for a single 50GbE link in particular firmware/host configurations, but verify your exact OPN and firmware documentation before relying on 50GbE as a supported mode.

Closing Technical Notes

This page is built to help purchasers, engineers, and architects evaluate the 900-9X662-0073-ST0 NvidiaConnectX-6 Lx dual-port 25GbE card for modern datacenter needs. It addresses the SKU’s key specs — dual SFP28 ports, PCIe Gen4 x8 interface, secure boot considerations, and firmware/driver lifecycle — while offering practical guidance on cabling, OS support, and deployment scenarios. For final procurement and firmware compatibility checks, always cross-reference the exact PSID and OPN against Nvidia published firmware tables and your OEM server compatibility lists.