I've never used the GPU at all and it's not even enabled in my kernel so I don't think I could test it. The heaviest load I managed to find is to run "openssl speed rsa2048 -multi 4". It keeps all 4 cores very busy running code that was hand-optimized to keep the ARM instruction pipes busy. I use this to validate overclocking capabilities as it's the workload I found to be the first one to hang or power down a board. Your "yes" workload does almost nothing in comparison
Regarding the DC-DC it's not really an issue, it's just that apparently I'm hitting some hard limits. What I noticed is that I can reach about 2.7A before the board suddenly powers off. And I have to unplug it and replug it to restart it. On a CS008 device using the same CPU and the same DC-DC regulators (ACT8846Q, SYR827 and SYR828), I can reach about 3.5A and could never see it cut off. One difference I'm seeing is that the inductors are slightly larger on the CS008 (240nH vs 220 on MiQi, maybe higher current rated though I'm not sure).
Another explanation could be the length of the copper tracks. On the CS008, the micro-USB connector is 1 cm away from the SYR827. On the MiQi, it's roughly 6 cm long. So maybe the voltage losses along these tracks at such high currents are enough to let the SYR827 decide to cut off due to too low a voltage on VCC_SYS. And now that I'm thinking about it, it could really be the cause because I remember that on the CS008 I noticed a long time ago that power adapters delivering 5.0V were not enough and that those delivering 5.2V were OK. I'll try to measure the voltage on C834 to see how it goes at 2.7A (before it cuts off). I have a 5.1V/30A power supply here that I can trust as it manages to drive 5 overclocked MiQis
Anyway, I want to insist on the fact that it's not a real issue, I'm driving the board at much higher current than what it's supposed to be rated for. But it's also true that if I can manage to make it run at 2.0, 2.1 or even 2.2 GHz with either sufficient cooling or aggressive thermal throttling, it would be great as it would allow me to reach higher speeds when less cores are being used.
Now I have enough boards to experiment a little bit, I can take the risk to sacrifice one for the good of the remaining ones (though I'd prefer to keep all of them intact of course).
Best regards and happy new year to you as well,