Spotted originally by Phoronix.com and further examined by some sites like Computerbase.de, a couple of entries inside the driver that refer to shader engines and a relation of shader engines with compute units suggest that Vega GPU could have four shader engines and 16 GCN compute units. With these two numbers in mind, and if the Vega has the same numbers of stream processors per compute unit as Fiji GPU, it could end up with a total of 4096 stream processors.
Other driver entries also suggest that each shader engine also has two Asynchronous Compute Units, one render back-end and 16 ROPs, which means that Vega will end up with 64 ROPs.
Combined with up to 8GB of HBM2 and the previously promised 12.5 TFLOPs of compute power, the Vega GPU will finally mean that AMD will have a decent competitor for the high-end graphics card market.
Of course, these specifications, as well as the performance, are yet to be confirmed but we will probably hear more and more as we draw closer to the promised Q2 2017 launch date. As written and promise earlier by AMD, the company will probably launch its new Vega-based graphics cards at Computex 2017 show in late May.