Infrared camera shows the regulator transistors (?) to be the hottest component as I noted earlier:
I believe those are the actual output transistors. My understanding of the IRS2092 chip is that the chip itself is really just a driver of some other output device(s). Compared to a "monolithic" design, where the switching logic is combined with the output device into one chip (TI tpa311x for example).
So I think it is technically possible to make a 700W amp with the IRS2092, as the limit won't be the IRS chip, but whatever output MOSFETs are used. Proper heatsinking will of course be
very important. Not sure what kind of efficiency you get with IR2092 based designs, but 90% is not uncommon for class D amps. At 700W output, that's still 70W of heat to get rid of. That needs either a fairly large heatsink or active cooling.
FWIW, the CDA and SDS series amps from
Class D Audio are based on the IRS2092 chip.
IIRC (big
if!), the IRAUD reference design amps suffer from low input impedance. So if you're not driving them with some kind of buffer (preamp), you're likely to have a suboptimal experience. And also IIRC, that's the main upgrade between the CDA and SDS series amps from Class D Audio, the latter adds a (I think) THAT1200 (or similar) front-end.
With regards to post-filter feedback: one interesting example of this, with the IRS2092 chip, is
ChocoHolic's SystemD LiteAmp. This was long on my "want to build" list, but I never got around to it. I'd be really curious to see how this measured.
In general, I would caution against using this review as a proxy for all IRS2092-based amps. As with DAC chips, I think the performance is overwhelmingly defined by the implementation, rather than the chip itself.