Feedback: Quick Sync and NVENC - Laptop

Maltahl

New Member
This is a Feedback based on my own experience in the recent updates of OBS 0.612b.

(UPDATED QUICK SYNC GOT FIXED BY UPDATING TO NEWEST DRIVER!)

Example of 1080p 48 fps with Quick Sync best quality with 3500 bitrate - it looks pretty damn nice with Intel HD 4600. Compared to the gains of using it. Better results may reached with Desktops, since this was 1 PC setup and it was a laptop.
http://www.twitch.tv/maltahlgaming/b/506392616

Quick Sync tests with 3 different presets using these settings using Intel HD 4600 on a laptop:
720p @60 fps with 3500 bitrate - downscaled from 1920x1080

Best Quality preset review:
Quality is great but most of the time when i stream i get skipped frames because the encoder is "too slow". this was not a problem in the earlier versions of OBS. Note i have been trying to work this out in Chat with Palana and Jim with no resolving of the problem.

Balanced preset review:
a bit worse quality than Best Quality but with same problems as Best quality.

Best Speed preset:
Feels like same quality as Balanced but handles fast motion very well.
No errors like with best quality and balanced preset.

Overall Quick Sync review: needs minor retouches but else i dont think anything more can be done unless Intel gives us more options. i honestly think that next gen of Intel i-core cpus ( Boardwell) will deliver amazing video stream quality but time will show.

NVENC tests was done on same laptop with a GTX 765m with 2 different setups:
Setup 1: 1080p 60 fps - preset high quality - 3500 bitrate.
Setup 2: 720p 60 fps - preset high quality - 3500 bitrate.


Note that other presets and sometimes the same preset cause problems with getting stream live on Twitch compared to Quick Sync.

Setup 1 review: Quality is great but cannot handle movement that well and gets worse than 480p with pixels if too much movement but this is expected.

Setup 2 review: Quality is also pretty great, and movement was handled pretty well with low/none pixelation in high motion games.

Overall NVENC Review: still needs more work since there is often problems getting stream live and getting stable bitrates to twitch on rock solid internet. Nvidia also needs to work on their encoder compared to Intel.

These problems on NVENC not getting stream live sometimes is not shown on OBS log file, same thing goes for Quick Sync too slow to encoding skipping frames error, only shown entry is something between the line of: cpu too slow.
If you want video examples i will be happy to show some.

TL:DR
Quick sync needs minor retouches but is pretty good so far, NVENC not worth your time yet.

Quick Sync is pretty neat with Haswell and should be used unless you got the spare CPU to do with normally.
 

AndehX

Member
video examples would be good. I've tried both, and the quality between the 2 seems to be the same for the most part. For high motion games, I've gone back to using CPU encoding as the quality is just terrible, even with a 3500bit rate
 

Bensam123

Member
Yeah, from what I've seen the high motion scenes are pretty terrible. MOBAs dont take as much bit rate as a FPS either. So you probably could be streaming at 720p@30fps with a 2000 bit rate and things would look fine. 3500 for 720p@60 is actually really high for a MOBA and would be on par if you're playing a FPS with a CPU encoder.

That said, LoL isn't really hard on the CPU to begin with so this probably isn't a best case scenario for hardware encoders. Something like BF4 or Planetside 2 would be a better show case as they stress the CPU and you can show meaningful FPS gains.
 

Maltahl

New Member
Did some extra Testing - Thanks Jack0r for the (nearly) lossless footage to base these comparison on.
Used the video source plugin to do this.

Remember to DOWNLOAD, since dropbox media player compress the footage down.

1080p 30 fps - Quick Sync - 3500 Bitrate / 25000 kbps buffer ( CBR enabled dont worry)
https://www.dropbox.com/s/2nhu5oarv... bitrate - Quick Sync Best Quality preset.mp4

1080p 30 fps - x264 Fast Preset - 3500 bitrate / 25000 kbps buffer ( CBR enabled)
https://www.dropbox.com/s/70negt5gsvadhtc/1080p 30 fps - 3500 bitrate - x264 Fast preset.mp4

1080p 30 fps - NVENC HQ Low latency - 3500 bitrate / 25000 kbps buffer (CBR enabled)
https://www.dropbox.com/s/hrwzaofyivl2909/1080p 30 fps - 3500 bitrate - NVENC.mp4

All of the footage in 1 zip file
https://www.dropbox.com/s/m10brpz8gfxm1 ... 20Test.zip
 

Bensam123

Member
You guys definitely should use FPS's when testing this stuff. 3500 bit rate is really high for a RTS, it's not going to stress anything at all. The car chase scene was a start, the zombie FPS didn't have enough movement in it.

Use the same content to make direct comparisons. Use high resolution, high quality source material.

If you really want to do this, I'd suggest usiung BF4 close quarters footage 1080p@60fps with buffer=0 and crf=15 in the custom parameters as a source.

Also why are you using such a huge buffer for the tests?
 

Maltahl

New Member
Cause it helped on pixelation in my case. without it there would be more pixelation.

First off i do not have BF4, secondly recording it in 1080p 60 fps is impossible for my machine, atleast to get very good quality of out it.
 

Boildown

Active Member
You should use a five minute (or longer) video with plenty of high action sequences. And then use the Video Source Plugin to use the same input video for all your tests. To make the initial input video, save to disk at a high enough bitrate (25-50Mb/s) so as to be essentially identical to the original gameplay.

Edit: Oh, you did this. I was thrown off because the hitbox.tv recordings show different video sequences. But your description says you did this, which is good. :)

In my testing, short minute or two long videos don't have consistent results, I consider five minutes to be the minimum for any kind of statistical relevance, and obviously longer than that is better.
 
Top