In the new patch of LLVM, two GPU cores of AMD Vega 20 and Vega 12 were revealed.
The picture shows Vega 64
Among them, Vega 20 we have already learned that based on TSMC's 7nm process, the first product is the MI25 accelerator card, which is used to accelerate the deep learning ability of the machine in the AI scene. It will be launched before the end of the year.
The new patch of LLVM confirms that Vega 20 (GFX906) will support seven instruction sets such as fdot2, sdot2, udot2.sdot4, udot4.sdot8, and udot8.
The Vega 20 also appeared in 3DMark 11 before, showing that it has HBM2 memory with a speed up to 1250MHz and a capacity of 32GB. Other parameters include 4096 stream processors, support for PCIe 4.0, and power consumption between 150~300W.
As far as the Vega 12 core is concerned, there is no further information available to determine its hierarchy. There are arguments to replace Polaris's mainstream level products.
