AMD announces new Ryzen AI APUs (Strix Point) in Computex 2024

Sure,it’s not a bios nor a microcode. It’s a GPU Firmware update: you can look here

2 Likes

I believe we are talking about different issues. I am referring to this one: Power consumption for HW accelerated video decoding for Radeon iGPUs is simply outrageous (#3195) · Issues · drm / amd · GitLab TLDR: video hardware acceleration under linux is so inefficient that it makes it borderline useless. This is likely to be solved by new VPE block in gpu (basically a hardware scaler that can run without involving power hungry parts of the GPU) which is not available in Phoenix but probably is in Strix (gpu being RDNA 3.5)

1 Like

Alright thanks for pointing the exact issue :slight_smile: . True that hardware video decoding is too much hungry imho.

I am not entirely sure it’s fully a hw issue as it performs much better on windows. I’d be pretty happy with a bit under 1w over idle for 1080p 30 playback myself.

1 Like

It is not but as far as I understand the problem it is not trivial to do rendering in linux the way windows does. And even in windows older intel systems are more efficient than current AMD ones. Far off 1W you mentioned. See discussion in bug report I linked above. That may still be a bug fixable in software/firmware but the fact that AMD driver developer is participating in this discussion and nothing came out of it doesn’t give me much hope

I am also not sure it’s entirely or even mostly the rendering and presentation bit. On my old intel thinkpad I can ffmpeg hw decode 720p for less power than sw, on arm (rk3588) I can ffmpeg hw decode 720p for less power than sw (hell I can transcode, so decode h264 then encode h265 1080p at allmost 500fps for less power than it takes the framework to just decode at 30 fps), on my amd framework I can’t do that. Sure the amd chip does have stupidly efficient cores but shouldn’t hw decoding still be more efficient?