NVLink allows two GPUs to directly access each other's memory. This allows much faster data transfers than would normally be allowed by the PCIe bus. This videos discusses the NVLink architrecture from a dual computer system up to an advanced HPC 8-GPU system. The two NVIDIA GeForce RTX 3090 GPUs featured come from a loaner Exxact high performance machine learning workstation.

** System Used **

TRX40 Motherboard
Threadripper 3960x
128GB Memory (16GBx8)
2x 4TB PCIe 4.0 NVME
2x NVIDIA GeForce RTX 3090
NVLINK Bridge

2:01 HPC, Scientific, Rendering
3:25 High-End HPC NVLink
6:52 Unified Virtual Addressing (UVA)
7:14 CUDA Code Example

For more information about the machine featured in this video, please visit:
https://www.exxactcorp.com/Deep-Learn...