NVENC AVC Rendering Quality?

j-v wrote on 9/26/2019, 12:29 PM

GPU-Enabled HEVC is still painfully slow, relatively speaking.

I recall it coming in around the same as software x264 in a single test.

In terms of quality it seems "less bad" than hardware AVC. That's jmo.

I don't know what "painfully slow, relatively speaking" means , but I'm very satisfied with VP17 and its possibilities, playing full speed on preview best full my 4K project with 2 tracks of heavy 4K HEVC 50p files of my GOPro 7 and rendering a 1 minute part of the project to Magix HEVC UHD 3840x2160 50p default template with NVenC from Nvidia. You are able to see it all here, taskmanager included.

john_dennis wrote on 9/27/2019, 3:12 AM

Using my RX480, I didn't find AMD VCE or AMD HEVC "painfully slow". At the default Magix settings, neither of them produced picture quality of my Happy Otter Script x264 render which really was "painfully slow". (I rarely watch that pot boil, so I don't much care.) Increasing the bit rate of the AMD HEVC render didn't improve the picture quality as much as one might expect. I didn't try 100 Mbps as I've had trouble with bit rate that high with hardware encoders in the past and didn't want to push my luck. Like this...

Musicvid wrote on 9/27/2019, 9:11 AM

I'll reoeat some tests on QSV HEVC -- I wasn't really paying a lot of attention.

Musicvid wrote on 9/27/2019, 11:43 AM

Using ABR to eliminate the wide disparities in CQ indices between QSV h264 and h265, I am getting 50 fps encodes from QSV h264 and 42 fps encodes from QSV HEVC, about a 20% time increase as collateral for the smaller file. That's on pretty tame source. But that's only one of several reasons I'm sticking with h264 variants for the time being.

If you are using HOS for your comparisons, depending on how it is set up, the governing factor may be Frameserver, not the hardware encoders themselves. I encountered something similar.

john_dennis wrote on 9/27/2019, 11:57 AM

I only use frameserver for HOS GOP-15 Zero Latency. The rest are native to Vegas Pro 15.

Musicvid wrote on 9/27/2019, 12:24 PM

I'd say you've pointed out another good reason to choose NVENC or AMD over QSV.

wwaag wrote on 9/27/2019, 8:56 PM

"the governing factor may be Frameserver, not the hardware encoders themselves. I encountered something similar."

The FrameServer is constant regardless of the type of render--cpu, QSV, Nvenc or VCE. It simply renders an uncompressed signpost avi-frame by frame.

Musicvid wrote on 9/28/2019, 9:00 PM

Sorry for the second editorial, but I am so over QSV.

Using VideoRedo 6 (I'm in love btw), I sanitized and encoded the first ten hours of Ken Burns' Country using QSV h264 at CQ15. The results are mushy, and the poor motion estimation makes it almost unwatchable in places, especially old movie clips. A total waste of conversion time, if one at all appreciates the incoming quality of a Burns masterwork.

Fortunately, VRD 6 has x264, and I feel safe and warm at RF20. Really, If I haven't got the "time" to encode properly in x264, I must not be that invested in the recording project. Did I say that QSV sucks?

Re-encoding the original .mts as we speak . . . Interested in hearing further reactions as to NVENC and AMD, but I'm still disappointed.

https://www.vegascreativesoftware.info/us/forum/really-disappointed-with-hevc--116394/

[End of Editorial /]