New Infinity Link technology offers high-speed GPU-to-GPU transfer in multi-GPU networks.
AMD recently announced that its new AMD Radeon Pro VII graphics card will be available starting in mid-June. The AMD Radeon Pro VII is optimized for engineers and researchers performing CAE simulations or developing high performance computing (HPC) applications.
The new AMD Radeon Pro workstation GPU has 16GB of High Bandwidth Memory (HBM2) and the long-anticipated high-bandwidth Peripheral Component Interconnect Express (PCIe) 4.0. The PCIe 4.0 motherboard interface doubles the available bandwidth to GPUs, Wi-Fi, Ethernet cards, hard drives and SSDs.
For engineers and product designers using large-sized 3D files, the Radeon Pro VII offers a 1TB/s memory bandwidth and full ECC capability designed to perform with low latency.
The company is packaging a new technology with the release called AMD Infinity Fabric Link, which allows users to perform high-speed GPU-to-GPU transfers (5.25x PCIe 3.0 x16 bandwidth and 168GB/s) in a peer-to-peer multi-GPU network. According to AMD, the Radeon Pro VII with its Infinity Fabric Link will be an ideal fit for the company’s Radeon Open Compute (ROCm) software ecosystem for HPC and machine learning.
Bottom Line
The wait for PCIe 4.0 is over. The increased bandwidth available for HDDs, SDDs, Wi-Fi and other components will be a big help for expediting data transfer. The Infinity Link technology will also help researchers (who all have Radeon Pro VII GPUs) perform tasks in a multi-GPU environment with greater speed.
Workstations equipped with AMD Radeon Pro VII-GPUs will be available from OEMs in the second half of 2020.