eBay

Product Overview


This dual-card NVLink 300G high-speed interconnection adapter board is custom-engineered for SXM2 interface compute cards. Your motherboard must support PCIe bifurcation; you need to split PCIe lanes into x8/x8 via BIOS settings before operation.

Designed exclusively for NVIDIA V100, P100, PG199 and other SXM2-architecture GPUs, it enables ultra-high-bandwidth NVLink communication between two graphics cards, greatly improving efficiency for AI training, deep learning and parallel computing scheduling. It acts as the core conversion baseboard to install rack-mounted SXM2 server GPUs onto standard desktop PCs and regular PCIe hosts.


Key Specifications


  1. Physical Dimensions: 30cm length × 11.5cm width, fits standard full-length PCIe slots and matches GPU bays of mainstream computer chassis.

  3. High-Speed Interconnect: Native support for 300Gbps NVLink dual-card channels, allowing ultra-fast video memory data exchange between two GPUs and removing bandwidth bottlenecks during multi-card cluster training.

Product Design Advantages


Direct-Plug PCB Architecture

Adopts an integrated hard-connection design with no extra cables or transfer wires required. SXM2 cards plug straight into onboard slots for tighter connection, extremely low signal attenuation and outstanding stability under prolonged heavy workloads.

Sufficient Power Supply Design
Comes with four independent 8-pin GPU power connectors on the board. Generous power margin for each card fully meets the full-load power draw of high-power compute cards like V100, eliminating frequency throttling and unexpected restarts under heavy loads.

Native Thermal Control Expansion
Reserved standard fan headers support original server air cooling modules directly. It also works perfectly with split dual-GPU liquid cooling modification kits. You can freely switch between air cooling and liquid cooling modes to balance quiet operation and extreme heat dissipation performance.

Compatibility List


Compatible with all SXM2-form-factor GPUs:

  • NVIDIA V100 SXM2 (16GB / 32GB / 64GB versions all supported)
  • NVIDIA P100 SXM2
  • PG199 series SXM2 compute accelerator cards

Application Scenarios


Local large model deployment (running 32B/64B high-VRAM models via Ollama), AI image generation, deep learning model training, scientific computing, virtualized GPU passthrough, multi-GPU computing cluster building. It perfectly solves the core pain point that SXM2 server GPUs cannot be directly mounted on ordinary PC motherboards.