Quicksync quality per generation

[ferk]

Hello, first of all, sorry if this is not the correct way to post this or if it's answered somewhere else (I tried searching but I couldn't really find exactly what I was looking for).

Now, with my question. I'm having some doubts about buying a mini pc. I had one a few months ago which was a N100 Alder Lake. It got extremely hot so I sold it. While I had it, I was able to experiment with quicksync and hevc quality, which in my opinion was pretty similar to nvenc (I'm not sure if this is true, just my opinion).

I have now seen an offer on another mini pc, this one being a i5-8259U, older quicksync generation so I'm not sure what kind of hevc quality to expect. I know that over the years Intel has improved a few things in quicksync (av1 decode and all that), but, do you guys know if the quality (in terms of vmaf and similar) has improved and how much if that were the case?

[pandy]

My personal impression is that newer hardware offer usually higher quality of video encoding - this is natural - more complex HW is able to deliver better quality.
So if your goal is quality then or you use newest HW encoder possible (with moderate bitrate increase when compared to CPU driven encoder) or you are accepting higher bitrate and to deliver same quality on older HW encoder.
This is always tradeoff between quality and bitrate.

[ferk]

My personal impression is that newer hardware offer usually higher quality of video encoding - this is natural - more complex HW is able to deliver better quality.
So if your goal is quality then or you use newest HW encoder possible (with moderate bitrate increase when compared to CPU driven encoder) or you are accepting higher bitrate and to deliver same quality on older HW encoder.
This is always tradeoff between quality and bitrate.

Gotcha. I made some testing last night with a Skylake laptop that I have and the same video at the same bitrate achieved 92.24 vmaf with Intel and 93.9 in Nvenc (I have a RTX 3080), considering Skylake doesn't have hevc 10 bit encode, I guess the Coffee Lake would be around the middle? Better than Skylake, worse than Nvenc.

[pandy]

Gotcha. I made some testing last night with a Skylake laptop that I have and the same video at the same bitrate achieved 92.24 vmaf with Intel and 93.9 in Nvenc (I have a RTX 3080), considering Skylake doesn't have hevc 10 bit encode, I guess the Coffee Lake would be around the middle? Better than Skylake, worse than Nvenc.

To be honest i can't tell - for sure NVEnc is pure HW solution where Intel offer also hybrid mode (HW+CPU) - in theory this should lead to better quality. Recent ffmpeg introduced if i recall correctly Ultra HQ setting so it may be important to use latest possible drivers and software tools.

Based on https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video?useskin=vector seem CoffeLake using newer gen than SkyLake so there is chance that it will offer higher quality at the same bitrate.

You can try to use rigaya https://github.com/rigaya tools to squeeze absolute maximum possible from both HW encoders.
Also recent Intel GPU's (ARC series) may offer better quality (so instead replacing CPU's perhaps it could be easier to simply use Intel ARC graphic card - they offer also AV1 encoding - not so common).

[ferk]

Gotcha. I made some testing last night with a Skylake laptop that I have and the same video at the same bitrate achieved 92.24 vmaf with Intel and 93.9 in Nvenc (I have a RTX 3080), considering Skylake doesn't have hevc 10 bit encode, I guess the Coffee Lake would be around the middle? Better than Skylake, worse than Nvenc.

To be honest i can't tell - for sure NVEnc is pure HW solution where Intel offer also hybrid mode (HW+CPU) - in theory this should lead to better quality. Recent ffmpeg introduced if i recall correctly Ultra HQ setting so it may be important to use latest possible drivers and software tools.

Based on https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video?useskin=vector seem CoffeLake using newer gen than SkyLake so there is chance that it will offer higher quality at the same bitrate.

You can try to use rigaya https://github.com/rigaya tools to squeeze absolute maximum possible from both HW encoders.
Also recent Intel GPU's (ARC series) may offer better quality (so instead replacing CPU's perhaps it could be easier to simply use Intel ARC graphic card - they offer also AV1 encoding - not so common).

Thanks again for your answer. I'll check the ffmpeg thing.

Unfortunately the ARC gpu is not a choice. I'm trying to set up a home server with some multimedia capabilities. I had the older mini pc with that idea in mind, but I had to get rid of it because I didn't trust it due to very high temperatures. The laptop as home server is another idea, but Skylake is a bit behind in that regard. I've been testing it with Tumbleweed because I need docker (and Windows with docker doesn't work too well) but if I can get a mini pc on offer (the i5 8259U was) then perfect. It would have been a bit more powerful than the N100 (and the laptop), but if I'm not sure about quicksync...

[pandy]

If your goal is to transcode on the fly (i.e. some movie library to render device such as TV) then you can set highest possible bitrate - If your intention is to store (long term?) transcoded movies then i would consider using CPU transcode to get highest possible quality.

[ferk]

If your goal is to transcode on the fly (i.e. some movie library to render device such as TV) then you can set highest possible bitrate - If your intention is to store (long term?) transcoded movies then i would consider using CPU transcode to get highest possible quality.

Yeah, I do have a desktop pc with the 3080 and a Ryzen 5700x which does cpu transcoding relatively fast (10-20 fps at slow preset). I do that with the videos I care the most, but for example I have a Plex server with around 3-4TB of things, sometimes I have to transcode 20-30 videos of something and doing it in software would take A LOT of time and for some things I don't really care if the quality is slightly worse if I can do it 10 times faster.

I can always use nvenc in those cases but I would have to move the files around, plus nvenc tops at around 100-120 fps and 150w while the N100 could do 200-220 fps at much lower power consumption (my unit was bios limited to 10w in the cpu side so... around 20w total?).

I could get another N100... But I cannot know the thermals before I purchase it and I had a really bad experience with the one I had. I feel like sometimes I overcomplicate things a bit to be honest.