They're marketed as different sockets, so I wouldn't be surprised if that does not work. E.g. because a SP3r2 motherboard doesn't route power to unpopulated dice on the module.
Past experience suggests it's very likely they will be compatible, at least at a basic level.
See eg. or Socket 'L' vs. 'F' in AMD's history (hint -- they're the same thing), or the various Socket 2011 Intel iterations across server/enthusiast markets.
You may lose out on some features (eg. RDIMM support with server CPU, overclocking support in either direction, etc.)
Untimately, I hope, the cross-compatibility of EPYC CPUs on enthusiast chipsets will be a decision for motherboard manufacturers to make, as it has been in the past. 64 PCIe lanes should be enough for... at least some of us ;)
ETA:
As parent mentioned, the situation this time is more complicated.
Threadripper in an EPYC board is likely to be problematic -- the CPU doesn't have enough PCIe/Infinty Fabric connectivity to allow CPU links and PCIe to be active simultaneously. Even in a 1P system, only half of the board's potential PCIe lanes would be available. Due to these issues, it's likely such a setup may not be supported. It would be a strange thing to try in the first place, however, especially if equivalent EPYC parts end up similarly-priced to Threadripper.
EPYC in a Threadripper board is the news I'm hoping (and expecting) to be better -- the parent talked about powering the extra dies... an interesting consideration, though I expect (and hope) the only pins which won't be broken out on a Threadripper board will be the extra PCIe lanes.
