3 ms·

All modern integrated GPUs have full accelerated video decode for both H.265 and VP9. Where did you get your 150mW figure from?

ksec 5y ago
Full accelerated video decode does not equal low power video decode. I think I need to keep repeating this until I stamp out all accelerated video decode are created equal. ( Otherwise you would have market wide AV1 decoding by now ) You could spend 1W or up to 4W decoding 4K60 on non-Mobile optimise Video Decoder IP. It doesn't block your CPU, but it is not low power. >Where did you get your 150mW figure from? It is consistent with Apple's video decoding power requirement over the years. May be slightly lower given at one point it was 250mW. You could do your own measurement on your macOS with powermetrics utility.
- sudosysgen 5y ago
  Moden laptops ALL have dedicated low-power video decode. There is no modern laptop that will use 4W decoding video on the GPU. EVERY single modern laptop has a video decode IC that is optimized for mobile usage. Hell, AMD laptops even share their iGPU desgins, loosely speaking, with some smartphones. > It is consistent with Apple's video decoding power requirement over the years. May be slightly lower given at one point it was 250mW. You could do your own measurement on your macOS with powermetrics utility. No, you can't. powermetrics on macOS is not granular enough to do that. If you figured out a way, please be explicit on how you did it.
  - ksec 5y ago
    > EVERY single modern laptop has a video decode IC that is optimized for mobile usage Mobile as in Laptop or Smartphone? Early Intel CPU requires 2W+ for AV1 4K60 content hardware decode. That is not in any shape or form "optimized for mobile usage" in my book when. A 500mW or 1W Hardware Decoding on SmartTV is not really "optimized for mobile usage" in my book. Or do you want to look at GPU Hardware accelerated Motion JPEG2000 decode? Which has even more absurd number. 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. >No, you can't. powermetrics on macOS is not granular enough to do that. If you figured out a way, please be explicit on how you did it. Here. A whole detail article on similar findings. https://singhkays.com/blog/apple-silicon-m1-video-power-consumption-pt-1/ https://singhkays.com/blog/apple-silicon-m1-video-power-cons...
    - snvzz 5y ago
      I can't help but notice they didn't include any PCs in their measurements. >Early Intel CPU requires 2W+ for AV1 4K60 content hardware decode. Whereas Apple silicon has no hw decode for AV1. Benchmarks linked just use DAV1d.
    - sudosysgen 5y ago
      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.

ksec 5y ago

Full accelerated video decode does not equal low power video decode. I think I need to keep repeating this until I stamp out all accelerated video decode are created equal. ( Otherwise you would have market wide AV1 decoding by now ) You could spend 1W or up to 4W decoding 4K60 on non-Mobile optimise Video Decoder IP. It doesn't block your CPU, but it is not low power. >Where did you get your 150mW figure from? It is consistent with Apple's video decoding power requirement over the years. May be slightly lower given at one point it was 250mW. You could do your own measurement on your macOS with powermetrics utility.

sudosysgen 5y ago
Moden laptops ALL have dedicated low-power video decode. There is no modern laptop that will use 4W decoding video on the GPU. EVERY single modern laptop has a video decode IC that is optimized for mobile usage. Hell, AMD laptops even share their iGPU desgins, loosely speaking, with some smartphones. > It is consistent with Apple's video decoding power requirement over the years. May be slightly lower given at one point it was 250mW. You could do your own measurement on your macOS with powermetrics utility. No, you can't. powermetrics on macOS is not granular enough to do that. If you figured out a way, please be explicit on how you did it.
- ksec 5y ago
  > EVERY single modern laptop has a video decode IC that is optimized for mobile usage Mobile as in Laptop or Smartphone? Early Intel CPU requires 2W+ for AV1 4K60 content hardware decode. That is not in any shape or form "optimized for mobile usage" in my book when. A 500mW or 1W Hardware Decoding on SmartTV is not really "optimized for mobile usage" in my book. Or do you want to look at GPU Hardware accelerated Motion JPEG2000 decode? Which has even more absurd number. 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. >No, you can't. powermetrics on macOS is not granular enough to do that. If you figured out a way, please be explicit on how you did it. Here. A whole detail article on similar findings. https://singhkays.com/blog/apple-silicon-m1-video-power-consumption-pt-1/ https://singhkays.com/blog/apple-silicon-m1-video-power-cons...
  - snvzz 5y ago
    I can't help but notice they didn't include any PCs in their measurements. >Early Intel CPU requires 2W+ for AV1 4K60 content hardware decode. Whereas Apple silicon has no hw decode for AV1. Benchmarks linked just use DAV1d.
  - sudosysgen 5y ago
    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.