Yep
(This is not directly related to mantle but I am using it to show that 45% is overly optimistic).
Performance increase was 0%. Not 10-15%.
In their defense the improvement is simulated but it clearly shows that this is not an 'up to' statement. They also said these were the results of internal predictions/simulations on gaming, digital media, and productivity, none of which showed any improvement outside OpenCL.
This is why the 45% claim is very likely over estimated.
I mean, just looking at common logic would show that BF3 and BF4 are extremely CPU light in singleplayer, using only a couple cores and getting 200+ fps per second on an i5 (hitting the engine cap). Frostbite does not have a draw call problem. Multiplayer CPU dependancy is likely as a result of other things such as networking, AI, and other stuff. There are a lot more players in MP but the disparity is larger than that.
IMO mantle is a mistep. AMD should have spent the time developing a physics engine for use on their igps, making it easier on the CPU. RTS should use it for pathfinding and AI which would probably would increase performance just as much if not more. The oxide demo is not going to run nearly so smoothly once they add in proper AI and animations. There is no pitch and yaw on the ships. There may be 1000's of units but they really aren't doing anything.
Then you watch a video like this and you really are not that impressed anymore.
http://www.youtube.com/watch?v=NRTmVr4pU8Y