So you are mentioning bare die, I would actually take a BGA Processor with an unlocked multiplier.
Bare-die is old school though, like my old 1.4 ghz Tbird. I was sweating bullets as a PC builder noob, trying to mount a cooler on that thing. All you have to do is put the IHS in the package so people can put that guy on there if they want it (which I think a lot of people would, especially since some HSFs suck with bare die anyway, and bare die screws up a lot of mounting mechanisms).
Point is, they (AMD, Intel) save money by not having to solder or epoxy on the IHS (just provide a narrow band of epoxy as a mount guide for the IHS). The end user gets an IHS if they want it, or bare die if they don't.
As far as going BGA is concerned . . . my old AM3 board saw 3 different CPUs, and it would have been 4 if Thuban/Zosma prices weren't still so high. So I don't know if I can get behind that for normal desktop use.
(Apparently AMD APUs also use TIM now, not solder)
AMD APUs have used TIM since at least Trinity. It may be that Llano had no solder.
P.S. The processor could even be Carrizo (although I don't know if the HDL Excavator core scales well to higher TDPs)
If I were an HSA developer, I would definitely want access to commodity Carrizo FP4 BGA boards in something like MiniITX form factor. The very idea that I have to buy a whole laptop just to do proper HSA development with full iGPU context switching is laughable at best. Plus FP4 BGA Carrizo boards would be really sweet at cluster nodes for the right application/user.
How many existing FM2+ boards are overbuilt beyond the bare minimum specification for 95 watts?
If I had to guess, I'd say two: the A88x Pro and the Crossblade Ranger.
If so, by how much? 10 watts? 20 watts? 30 watts? Or more? (I know the higher end FM2+ boards have much better power delivery than the lower end FM2+ boards even though the official power rating is the same)
I have not personally found any upper limit on power delivery on my A88x Pro, though I am not running the most power-hungry APU (mine only has 384 shaders after all). The closest I ever got to power-related throttling was when I tried running Prime95 Small FFTs and Furmark on my 7700k simultaneously @ the full overclock (4.7 ghz, blah blah blah). That was with amdmsrtweaker forcing the P5 state to 4.7 ghz. At that point, the CPU reverted to the P4 state. So I forced P4 to 4.7 ghz and everything ran smoothly.
Note that I was using Small FFTs so the iGPU and CPU would not fight over memory bandwidth (they'll do that in Blend). Also, the thermal margin stayed below 80C so I wasn't getting that kind of throttling either.
Oh, and to address Shintai's assertion that only 3 AM3+ boards support the 9590: untrue. pcpartspicker lists 11 motherboards with BIOS support for that CPU. They may not support any overclocking with the chip - that is, you'd be stuck at the stock 4.7 ghz/5 ghz turbo. But they support the chip. Even the lowly ASRock 970 Performance has support for the 9370 and 9590.