Yeah, you just can't do it like this. How do you know that powermetrics accounts HW decode properly to GPU or CPU and not the SoC/Other section. That's problem one. This is made evident by the fact that powermetrics reports the GPU power consumption during 4K60 VP9 decode to be
2mW.
And it certainly isn't being accounted to the P-cores or E-cores, so all you're seeing is DRAM usage. The "Other" category in the legend is supposedly package power, but it's less than the sum of other rest, so it's either inaccurate or the legend is wrong.
Now that's a problem in and of itself. The bigger issue is that, on M1 Macs, powermetrics is just unreliable. If you look at Anadtech's testing, you'll see that powermetrics sometimes overshoots wall power, and sometimes undershoots it drastically. As in, by 25%.
Not only that, but powermetrics reports an idle of 200mW package power, while the actual laptop is drawing 7.6W at idle from the wall. This is pretty ridiculous - there is nothing that's going to be drawing power outside of the SoC and the screen, and the screen is not drawing 7.4W.
Anadtech also did a wall power measurement of the M1 Mac Mini and found a 4.2W idle.
So clearly, and unequivocally, powermetrics is unreliable. It is unclear and not nearly granular enough compared to the tools used in the PC world to be able to trust it. For example, on any modern PC laptop, you will get the exact power rail where the voltage and current is measured, and not only that, you will get the point at which this power is measured - before or after which VRM, exactly?
Not only that, but you will also get the actual power measured by the battery BMS at any given instant.
Even this is still taken with a huge asterisk and people still measure actual power draws, because even this level of granularity isn't enough. Vague, filtered numbers for powermetrics just doesn't cut it. If you want to make these comparisons, you have to compare wall numbers.
> And what has sharing iGPU design got to do with Video Decoder IP? ( If you are referring to Samsung and AMD ) The point is A video decoder that uses 500mW on Laptop is still more than 150mW on Smartphone.
Because the video decoder is part of the iGPU. Is your theory that AMD made a video decoder for smartphones that is 4x more efficient and just decided not to include it in their other GPUs?
> Early Intel CPU requires 2W+ for AV1 4K60 content hardware decode.
Now you're comparing AV1 to VP9... And those CPUs have only partial hardware decode for AV1. This is an improvement compared to the M1, which has no hardware decode at all for AV1, and uses even more than 2W to playback AV1 at 4K. That's the part 2 of the article you linked.