Comparison between custom Magix AVC QSV and NVENC renders.

Cornico schrieb am 31.10.2017 um 13:20 Uhr

Yesterday I rendered a short video taken by my GpPro in 2,7K 50p lineair with 60 Mbps.
Still looking for the fastest and best renderoption with the new Magix AVC encoder I rendered first with the default QVS and NVENC options for Internet HD 1080p 50 fps. QVS was a little faster and the video's looked the same.

I also rendered with a customized template I usually use for rendering of fixed HD camera video and screencaptures.
I used this template twice, 1 time for OSV and 1 time for NVENC

Also here was the QSV (1) a litle faster, BUT has terrible faults compared to NVENC (2)

Screenshots of the preview of the rendered video's back on the timeline of VPRO 15 shows it clear to me.
Result 1 (QSV)
Result 2 (NVENC)

Added on question about it Result 3 (Magix Mainconcept)

Kommentare

Musicvid schrieb am 31.10.2017 um 13:33 Uhr

Very interesting indeed.

Can you add a CPU-only render as a reference?

NickHope schrieb am 31.10.2017 um 13:42 Uhr

What CPU, with which version of Intel graphics? I wonder whether not all QSV AVC rendering is the same (Wikipedia). (Wild speculation: Maybe even Intel chipset driver version is involved)

mateuszszar schrieb am 31.10.2017 um 13:49 Uhr

@Cornico - next time use high profile.

mateuszszar schrieb am 31.10.2017 um 15:01 Uhr

Look at your first screen 👇 (Profile: "main" -> should be "high")

Ehemaliger User schrieb am 31.10.2017 um 17:04 Uhr

What did you decide is best for encoding?

Also what cpu do you have

thanks

john_dennis schrieb am 31.10.2017 um 17:26 Uhr

"Also what cpu do you have"

Cornico is one of the courteous individuals who shows information in his profile.

Ehemaliger User schrieb am 01.11.2017 um 05:30 Uhr

Thankyou John.

Cornico, were the birates the same for the outputed files?

Not sure why your QSV file is so low quality. You may have a faulty cpu or drivers.

 

NickHope schrieb am 01.11.2017 um 07:12 Uhr

According to Cornico's profile he has an i7-7700 and an i7-7700HQ which are Kaby Lake and have Intel HD Graphics 630. My point was that QSV on these CPUs might be different from QSV on other Intel CPUs, so it's worth bearing this info in mind in the analysis.

@Cornico If you share a sample I can see what QSV on my old laptop (with a much lower/older CPU) does with it.

Ehemaliger User schrieb am 01.11.2017 um 08:38 Uhr

He might have a faulty cpu. Maybe overheating and throttling the performance of the CPU. I think QSV is real time transcoder, if it throttles would not slow down performance but reduce the quality. I use intel QSV for screen capturing. Normally use intel HEVC, but when I used to export to my older laptop I'd encode in Intel h.264. Quality seemed identical to what i was capturing, although i did not 'pixel peep' (as the camera guys say) I would have noticed quality as bad as cornico's defective cpu.

This is the setting in Bandicam. As said. it shows HEVC as that's what I use now, but dialog and settings are same for h.264

dream schrieb am 01.11.2017 um 09:43 Uhr

@Cornico do a test between your pc vs laptop . same settings(use only qsv magix 1080p).i think pc wins,but at what difference ? i wanna know.

and i have a laptop with 4710 hq ,i use qsv magix ,it renders really fast.and also looking for buying desktop with 8700k ,will post results of that also.

Ehemaliger User schrieb am 02.11.2017 um 00:33 Uhr

Interesting. still looks terrible.

Can you upload that source clip, if long use software to clip first few hundred frames. If you don't have the software to do that do a very high quality software encode.

thankyou

Ehemaliger User schrieb am 02.11.2017 um 02:42 Uhr

I confirm your findings. QSV is worse, NVENC is also pretty bad both on high quality and default, and software encode mainconcept is by far the best quality. Only the software encode gives the quality one would release if they valued their work.

NVENC if for example you need to do hours of encoding a day or need to get something up onto the internet as fast as possible. Maybe low quality NVENC is good enough for low quality youtube but it's been tradition that you throw highest quality encode possible at it especially now that YT are encoding even everything in VP9 including 360p & 480p which gives higher qualiy

It's also possible to notice the quality purely based on the VBR output of the file sizes. It equates to the amount of detail in the end result. The following is for a 2.5second encode. I only did short duration as I originally was going to upload a frame of each, but it really isnt necessary, I don't disagree with Cornico's finding.

Intel QSV:6.22MB (20.3mb/s)

NVENC 6.96MB (22.5mb/s)

MainConcept AVC (software) 10.3MB (33.3mb/s)

 

Ehemaliger User schrieb am 02.11.2017 um 10:04 Uhr

Just don't use hardware encoding. It's low quality. At first I thought it was about efficiency. They were not efficient but give them enough bitrate and they'll look as good as a software encode. This didn't turn out to be true. Not even at 100mbit.for 1080p50

Hardware encoding should not be used by anyone that respects what they produce. Reducing bitrate of a substandard 'codec' just produces an even poorer encode. It's not a winnable situation.

Ehemaliger User schrieb am 02.11.2017 um 11:34 Uhr

You have an I7 apparently running at 4.2ghz per core. Why not use it instead of low quality hardware encoding.

I won't standby and allow you to ruin your encodes. I can't allow it.

NickHope schrieb am 03.11.2017 um 06:04 Uhr

I have 12 combinations available here on my laptop, which has Intel HD 4400 graphics, with the same standard template: 2 decoders (compoundplug.dll or so4compoundplug.dll) x 3 GPU (Intel or AMD or none) x 2 encoders (MainConcept or QSV). Having rendered to most of them, including QSV, it seems the result of every combination is slightly different, but they're all acceptable, and none are anything like as poor as Cornico's QSV screenshot.

(Side note: Personally I will continue with the safer and higher quality route of leaving GPU acceleration of video processing set to none, and rendering externally with x264.)

Ehemaliger User schrieb am 03.11.2017 um 11:00 Uhr

I have 12 combinations available here on my laptop, which has , with the same standard template: 2 decoders (compoundplug.dll or so4compoundplug.dll) x 3 GPU (Intel or AMD or none) x 2 encoders (MainConcept or QSV). Having rendered to most of them, including QSV, it seems the result of every combination is slightly different, but they're all acceptable, and none are anything like as poor as Cornico's QSV screenshot.

(Side note: Personally I will continue with the safer and higher quality route of leaving GPU acceleration of video processing set to none, and rendering externally with x264.)

I was concentrating only on the far right bottom corner where there was that long grass moving in wind slighly and in shade/dark. with QSV it did poorly, like Cornico's except not as bad but it was the same Compression artifact. Seeming low fps? and no detail, it only slightly improved with turning quality to 7(highest quality). The problem with Nvidea hardware on default and high quality setting was with very obvious noise from frame to frame and less detail than software encode.

It is interesting. Traditionally with other software Nvidia encodes were the worse.

NickHope schrieb am 03.11.2017 um 14:18 Uhr

My poor screenshot comes from a customized template, The same settings for the NVENC gave better result. Did you also try this customized template?

Oops, I missed that in your first post, but I've done it now. Here's a zip of the QSV render with Intel HD 4400 graphics, and the MAGIX AVC MainConcept mode render for comparison: https://drive.google.com/file/d/1oGjhFQCchtPXnlgMMj7ICjpIw3FedRev/view?usp=sharing

The QSV has a nasty "pulsing" thing happening and is basically unusable. Its average bitrate is low at 10.3 Mb/s, whereas the MainConcept came out at 14.4 Mb/s.

Ehemaliger User schrieb am 03.11.2017 um 14:31 Uhr

Realmedia used to do a an unusual thing whereby they would lower the fps in regions in the periphery, and I think INTEL QSV must be doing similar. Or it appears to. The grass in lower right hand corner seems to be at a lower fps as well as lack of detail. Good for compression if you don't care about the grass, but I don't think the encoder should be making that decision.

dream schrieb am 03.11.2017 um 15:37 Uhr

I have 12 combinations available here on my laptop, which has Intel HD 4400 graphics, with the same standard template: 2 decoders (compoundplug.dll or so4compoundplug.dll) x 3 GPU (Intel or AMD or none) x 2 encoders (MainConcept or QSV). Having rendered to most of them, including QSV, it seems the result of every combination is slightly different, but they're all acceptable, and none are anything like as poor as Cornico's QSV screenshot.

(Side note: Personally I will continue with the safer and higher quality route of leaving GPU acceleration of video processing set to none, and rendering externally with x264.)

you are awesome,please tell me with that 12 combinations (considering) which is good for me-

amd 8 core chip with Nvidia - render with nvenc (cant render with qsv )

or intel 6 core with nvidia - render with qsv/nvenc

p.s. quality doesnt matter,quantity matters.time should be saved.

Ehemaliger User schrieb am 03.11.2017 um 16:09 Uhr

I tried using real time Bandiscam screen capture of cornico video video with nvidia and intel QSV quck sync at quality 80 which is aprox 60mbit/s capture was at 1080p60fps h.264 and I did not see the probems with encode that I see with Vegas Pro . It makes me believe it's driver based and vegas pro's fault. It shouldn't be the case that a real time screen recorder produces better video quality than a video editor not working in real time.