AMD unveils Radeon Pro Vega: 8-16GB HBM2, performance up to 22 TFLOPs

AMD has officially unveiled the Radeon Pro Vega, the new Vega-based graphics card that for Apple’s new iMac Pro all-in-one desktop PCs. This new Radeon Pro Vega is the third entry in the Vega family, following the professional Radeon Vega Frontier Edition and consumer Radeon RX Vega graphics cards.

The new Radeon Pro Vega comes in two variant, with 8GB and 16GB models using HBM2 providing 400GB/sec of memory bandwidth. AMD is using the Vega 10 GPU, with 256 TMUs, and 64 next-gen Vega NCUs, each with two compute engines. Each of the compute engines has two separate compute clusters each with 512 stream processors and 32 TMUs, so in total Radeon Pro Vega has 4096 stream processors and 256 TMUs.

Addition to that, there are 64 ROPs, and 16 distinct renders back end that are slapped into the 2048-bit memory bus with HBM2, in both 8GB and 16GB versions.

AMD’s next-gen Radeon Pro Vega graphics card has a new rasterizer that converts polygons into pixels much faster by breaks them into batches, which are shrunken down small enough to fit into the on-chip cache. This is a Vega exclusive feature that allows the Radeon Pro Vega graphics card to use less VRAM and require less memory bandwidth, while increasing the throughput compared to previous-gen cards and previous methods.

The optimized pixel engine on Vega also sees the shade-once technology looking at pixels earlier in the graphics pipeline so it can see which ones are being rendered behind other on-screen objects, with non-visible pixels discarded, reducing the shader workload on the Vega GPU. Not only that, but Radeon Pro Vega is capable of writing images more efficiently thanks to the new direct connection between the pixel engine and the on-chip cache.