PNY NVIDIA NVLink Adapter

The PNY NVIDIA NVLink Bridge is a high-performance interconnect solution designed for NVIDIA Ampere architecture-based graphics boards and data center GPUs. It allows for the connection of multiple GPUs, enabling memory pooling and performance scaling. This bridge is essential for systems that require high computational power and large memory capacity, such as those used in deep learning, scientific computing, and high-performance computing (HPC) environments.

The NVLink bridge connects two compatible GPUs, effectively doubling their memory capacity. For example, when two NVIDIA RTX A6000 boards (each with 48GB of memory) are connected using an NVLink bridge, the total available memory becomes 96GB. Similarly, the NVIDIA A100 data center GPU, which has 80GB of memory per board, can achieve up to 160GB of total memory when connected with three NVLink bridges. This feature is particularly useful for applications that require large datasets or complex computations.

The NVLink bridge also provides significantly faster data transfer rates compared to traditional PCIe connections. The NVIDIA RTX graphics boards support up to 112.5 GB/s of bi-directional data transfer, while the NVIDIA A100 GPU can achieve up to 600 GB/s with three bridges per pair of boards. This high-speed interconnect enhances performance scaling by effectively doubling the number of CUDA Cores, Tensor Cores, and RT Cores available to the system.

It is important to note that application support is required for the memory pooling and performance scaling features to be enabled. Users should check with their software vendor (ISV) to determine if their specific applications are compatible with NVLink technology.


Brand: PNY
Model: NVIDIA NVLink Bridge
Specifications: Compatible with Ampere architecture-based NVIDIA RTX graphics boards and data center GPUs; requires application support for memory pooling and performance scaling
Features: Enables memory pooling, provides high-speed bi-directional data transfer, supports performance scaling by doubling computational resources