In terms of what AMD will earn for CPU products... most of this is determined by how many wafers they can make in a period of time. Hopefully, with the 2017 holiday season in the past AMD can push tons of wafers into higher margin areas until any potential ramp in semi-custom during summer/fall of 2018.
AMD can look at the total number of wafers, wafer yields, related binning percentages, and the expected use die in various product sets. They can even look at the expected costs vs revenues in alignment with these factors.
In short, the wafer starts put a cap on the earnings to a large degree. Above and beyond that there may be IP deals that provide revenue or semi-custom efforts that help them advance their IP in response to serving a customer R&D project.
However, there is a escape hatch in the WSA that allows AMD to use alternate foundries at a price. I don't know the timeframe involved but it would seem that last round technologies (as they would have lower transfer costs than the latest nodes) to support existing OEM products and inventory could be migrated to alternate fabs.
This is assuming that GF continues to be on time -- otherwise it might make sense to migrate the next reound of products on the latest nodes to keep up the product/roadmap cadence and foster market confidence in AMD's ability to deliver as stated.
Tip for 2018... keep an eye out for AMD increasing the use of alternative foundries.
At this point, on the GPU side, the wildcard is success into server and/or professional markets. Personally, I'll class the latest APUs as CPUs and leave them in that category for revenue considerations.
Disclosure: I am/we are long AMD.
Additional disclosure: I maintain a bull thesis on AMD based on their ability to bring competitive products to markets in which they have or have had little to no penetration and their dedication to an ongoing roadmap of advancements as evidenced by their published exascale computing and related APU efforts.