RDNA2 card for data center and cloud

AMD Radeon Pro V620 RDNA2 card for data center and cloud

With the Radeon Pro V620, AMD has announced a fast RDNA2 card with 32GB of graphics memory, which is designed for server use. Usage scenarios are, for example, VDI, machine learning, cloud-based rendering, or cloud gaming.

With the AMD Radeon Pro V620, the RDNA2 architecture is coming to the data center. A Navi-21-TX-Chip is located on the passively cooled card, which with 72 compute units, 4,608 stream processors, 72 ray-tracing accelerators and a 128MB infinite cache corresponds to the Radeon RX 6800 gaming card. XT. However, the Radeon Pro V620 is equipped with 32GB GDDR6 and the graphics memory is twice as large as in the sister gaming model.

The memory is connected through a 256-bit interface, the memory bandwidth must be 512 GB / s. The maximum clock frequency of the GPU is 2200 MHz. According to the manufacturer, the server card thus achieves an FP16 performance of 40.55 TFLOP and 20.28 TFLOP with the FP32. The full-height PCI Express 4.0 card occupies two slots on the server motherboard and the manufacturer claims that the maximum power consumption is 300 watts.

Image gallery

Application areas

As application scenarios for the Radeon Pro V620, AMD names use on VDI servers for desktop and workstation applications, machine learning, 3D rendering and cloud visualization, as well as gloud games.

With Single Root I / O Virtualization (SR-IOV), up to 12 virtualized computers can be provisioned with four GB of dedicated graphics memory per card. The card is said to be very suitable for cloud gaming thanks to its high performance and support for DirectX 12 and Vulkan 1.1. For machine learning, Radeon Pro V620 supports AMD’s own ROCm platform, which integrates machine learning frameworks such as Tensorflow and PyTorch.

There is currently no information on pricing and availability.

(ID: 47782801)