Quote Originally Posted by Stukov View Post
I think the reason you are seeing such a speed boost variance is based on what limitation a CPU bottleneck is affecting performance. While you may get 10% where no bottleneck exists, when you compare a slower CPU (such as AMD's APU) vs the fastest (Intel top of the line) you are going to see a much bigger boost with the APU % wise vs that of the Intel EE. You will also see greater improvement in a game which uses loads of CPU calculations like AI, such as the Swarm demo, Starcraft, or other RTS type games.
No, the reason there is such a variance is that Starswarm was coded especially for Mantle. It's beyond a best case scenario. In 90% of the Mantle titles we see for the next few years, the standard DirectX + Windows driver will be the primary rendering tool while Mantle will be added in. Hence why there is a much lower gain in BF4, the poster child for an actual Mantle-supporting game.

Quote Originally Posted by Iconyu View Post
I don't actually trust results with Kaveri, there are too many unknown factors like HSA, the 512 shaders, and the fast memory to take into account. At the very least I would expect Mantle to push the GPU compute work over to the APU shaders to free up the GPU shaders. Which isn't something I can replicate on my Intel.
HSA isn't active right now, nor will it come into effect within games. Games actually do a pretty good job of partitioning resources (especially those that support DirectCompute, etc.) already so HSA is mostly focused on compute-focused apps through OpenCL 2.0.