They've only been at the TDP disadvantage since Intel's shrink to 32nm- their Pineview Atoms were both hotter and slower. Their Cedarview Atoms are merely slower.
Given that Ontario is on the 40nm node, it's still doing very nicely against Atom. (And that TDP will only be hit under full load; I'm not sure how power draw until normal workloads compares.)
Temash should be interesting- AMD finally get that 28nm shrink they need, they integrated the FCH for further power savings, they get GCN graphics to make GPGPU offloading work better (hopefully a full HSA design), the cores clock faster and have more execution resources. (Floating point vectors go from 64 to 128 bit, which should make a serious difference in float-vector heavy applications like games.) They just need to ship the damn thing!