I have a USB power meter bridging the power supply and my Pi's power adapter. I wanted to measure the power consumption for Pi's GPU when it is handling various processes. Using QPULib, a library that lets you access the 12 QPUs, I ran a piece of CPP code on a number of QPUs, in parallel. As it happens, the current consumption turned out to be the same, no matter how many QPUs were used.

Now, that implies the power consumption is independent of how many active "cores" are being used off of the GPU; but, since the power consumption must change with increasing number of active QPUs, this would mean that activating even a single QPU turns on all the 12.

Or, this just means that QPULib is just a hoax. I would really prefer the former conclusion than the latter, and quite radical, one. Alas, these are the only thoughts coming up in my mind. Are any of these two correct or something else is up, herein?

  • There are so many factors affecting power consumption that IMO you're unlikely to get conclusive results this way. – Dmitry Grigoryev Feb 14 '18 at 14:24
  • You could test to see if you actually get improved performance with using more QPUs to rule out your 'hoax' theory; might be interesting if you include the power readings with (a) no QPUs in use (b) some and (c) all QPUs. – Aurora0001 Feb 14 '18 at 16:17