Leaked specs are hardly on par with actual parts and that seems to be the case here. Based solely on the specs, it doesn't make much sense to me that H variant would actually be designed with a higher TDP than the U counterpart. Both are 4C8T at same base/boost clocks, but the U unit has 9 CUs against H's 8 CUs and still pushes Vega clock further than H equivalent. Considering both share same platform, litography, building process and architecture, I expect that this would translate in an higher TDP specs for the U parts. It is not exactly usual that both performance gains from enabling more CU and rising clocks come along with lowering TDP. I'm not sure that better boost algorithms alone can achieve it. For that to happen, I think there might occur a significant disparity between batches of the same chip to justify binning in that level. There might be something else that specs alone can't explain.