EEVblog #698 - GPU Video Rendering

[wraper]

For me it feels that video quality is actually dropped since 60 fps. As I'm using big 30" 2560x1600 monitors where I can see all the detail, it feels like movement became smoother but picture is somewhat blurry. Like resolution/overall quality of video seriously degraded.

[Fungus]

it feels like movement became smoother but picture is somewhat blurry. Like resolution/overall quality of video seriously degraded.

That could be Youtube doing that...

[Fungus]

PS: What about a different codec?

This one claims to be able to encode four 1080p streams in real time on a PC: http://www.videolan.org/developers/x264.html

(nb. I haven't tried it, I just saw that page...)

[EEVblog]

For me it feels that video quality is actually dropped since 60 fps. As I'm using big 30" 2560x1600 monitors where I can see all the detail, it feels like movement became smoother but picture is somewhat blurry. Like resolution/overall quality of video seriously degraded.

No one else has mentioned that at all 

[EEVblog]

PS: What about a different codec?
This one claims to be able to encode four 1080p streams in real time on a PC: http://www.videolan.org/developers/x264.html

I do use x264 (Handbrake), have done so for years.
It's not as easy as it sounds to integrate with editors directly.

[wraper]

For me it feels that video quality is actually dropped since 60 fps. As I'm using big 30" 2560x1600 monitors where I can see all the detail, it feels like movement became smoother but picture is somewhat blurry. Like resolution/overall quality of video seriously degraded.

No one else has mentioned that at all 

Was not completely sure if it just subjectively felt because of FPS or was a real deal. So to be sure now I compared mailbag 697 and 680 which are shot in similar conditions.  Now completely sure that quality really degraded. Stopped both videos so there is no movement blur, and I can compare details. In episode 680 I can more or less distinguish body hair on your hands  . In episode 697 it's just a mess. Also 60 fps smooths out the wrinkles on your forehead a bit. Don't know if it is your camera or youtube crippling details to save bandwidth. I don't have camcorder, therefore cannot say anything about them. However on my Sony SLT-A77 camera which is by no means cheap, 1080p 50 FPS setting severely degrades details compared to 25 FPS. Images are of different size because I made screenshots with a snipping tool.
ep. 680


ep. 697

[wraper]

Background is much clearer too on older video. Of course might be focus/focal ratio or different camera used but in newer videos overall background is very blurry. There is some yellowish tint in episode 697 too. Colors in episode 680 look more natural and blue ESD mat looks blue, not yellowish. Have not calibrated my monitors with calibrator, however they are factory calibrated with calibration certificate including actual data and are not too old. Therefore should be much more accurate in color reproduction than average ones in the wild.
680:


697:

[EEVblog]

Those are likely to be focus issues. I usually have auto face tracking focus on so it can hunt.
Need to do proper controlled tests with fixed focus and upload test videos to see the difference.

[EEVblog]

Here you go, controlled tests.

50fps video using my usual process:


25fps video using my old process:


Original 60fps footage direct form camera:
! Private video

[justanothercanuck]

So I got my i7 4790k + GTX980 system set up...

... and nuts.  It seems my copy of Premiere is too old to support CUDA without using some $300 plugin...  (something called rapiHD?)

I'll have to find a new copy to test. 

[EEVblog]

Videos on YT filmed outdoors often appear sharper.

Of course they do. Outside lighting is an order of magnitude brighter than indoors, even on a dark overcast day.
This is why you get every twit thinking Gopro cameras have supurb sensors and optics. Take them indoors and they become a joke.

[EEVblog]

The only other thing that occurs to me is you have quite diffuse lighting. Which is obviously important for many aspects of benchtop photography. I wonder if experimenting with a spotlight for your head shots to see if more defined shadows and highlights gives an impression of increased sharpness.

Have tried it and people didn't like it. It's also very fiddly to get right.
The last thing you want is to dick around with lighting every time you shoot a video.
Sure I could put up huge studio lights in front of me but they would be a complete PITA and just get in the way every day, believe me, I've tried.

[EEVblog]

I think the biggest improvement may be in optimising bit rate and encoding setting.

No, I'm already using high bit rate and a high constant quality encoding rate.

[gildasd]

The only other thing that occurs to me is you have quite diffuse lighting. Which is obviously important for many aspects of benchtop photography. I wonder if experimenting with a spotlight for your head shots to see if more defined shadows and highlights gives an impression of increased sharpness.

Have tried it and people didn't like it. It's also very fiddly to get right.
The last thing you want is to dick around with lighting every time you shoot a video.
Sure I could put up huge studio lights in front of me but they would be a complete PITA and just get in the way every day, believe me, I've tried.

I agree with you.
I've done stage/event lighting, so I see minor stuff that could improve a bit your quality...
But improving your setup would reduce your flexibility, increase your set-up time and cost a bunch (you would need a light mixing table and then faff around with a light-meter)
I really don't think it's worth your effort.
At most, I would put custom contrast curves and fiddle with clarity in your post process.  That would make it more pleasing to the eye, but less realistic.
Personally I come here to learn stuff, so I'd rather have somewhat realistic colours and contrast to finding Dave very pretty.
Quite frankly, for you being a confessed "non arty guy", I think your general sense of light, colour and composition is pretty good.
(for example, your skin tones are nice and you don't have a big reflection on your nose)

[Corporate666]

But at the end of the day, it goes back to cornering a Mercedes engineer and demanding to know why they are using 19" wheels when you have reams of data showing that 17" wheels are superior for performance... the guy doesn't owe you (or me) an explanation that suits our thought process.

The Mercedes engineer isn't on youtube, with a forum, that encourages debates on the mater.

Welcome to the internet.  If you can't handle someone calling you out of your complete and obvious BS, it's not the place for you.

It was clear from the video that it was a CPU horsepower problem, and we had the author of Handbrake posting here as well as Dave posting the communication from the guy @ Sony which illustrated it was not a GPU, software or driver issue.

So claiming that the solution was software, an SSD and drivers was the BS.  And when confronted with that, rather than "handle it" you lashed out at the unwashed masses for being silly enough to even want 60fps when you knew it wasn't what they really needed.

I think that pretty well matches your statement "If you can't handle someone calling you out of your complete and obvious BS, it's not the place for you" pretty well.

[Corporate666]

For me it feels that video quality is actually dropped since 60 fps. As I'm using big 30" 2560x1600 monitors where I can see all the detail, it feels like movement became smoother but picture is somewhat blurry. Like resolution/overall quality of video seriously degraded.

I'd check bandwidth... I haven't noticed a degradation in quality from 60fps, and I watch some gaming channels where people would notice if there was a quality drop when the content creators moved from 30 to 60fps.

Youtube obviously adjusts their data stream based on what you have available (as well as caching), because I've noticed drops in quality when - for example - I am downloading a huge file in the background.

[EEVblog]

Quite frankly, for you being a confessed "non arty guy", I think your general sense of light, colour and composition is pretty good.
(for example, your skin tones are nice and you don't have a big reflection on your nose)

My Sony camera is not white balanced, just auto everything.
My main Canon camera is however properly white balanced and I fiddle with manual exposure a lot for many shots, plus manual DOF.
I like to think my composition is pretty good.
Nothing ever gets touched in editing in terms of the image, it's 100% straight from the camera.

[EEVblog]

Youtube obviously adjusts their data stream based on what you have available (as well as caching), because I've noticed drops in quality when - for example - I am downloading a huge file in the background.

Youtube is a mysterious beast. they can and do change video quality on a whim it seems.
They always keep the original uploaded footage, so they can automagically re-transcode at any time for any codec or bandwidth changes they want to make this week.

[SeanB]

The latest changes from them are causing stuttering audio for me. Can be a bit annoying, and is not present on all video, just a good portion of them. Really annoying that, though it might also be due to upstream traffic shaping or jittery servers locally.

[hikariuk]

Here you go, controlled tests.

50fps video using my usual process:

25fps video using my old process:

Original 60fps footage direct form camera:

I only see differences in quality I see are the normal switching from the initial low quality stream to the high quality stream jump in crispness I get with basically every channel I follow on YouTube.  And I see that on all three of those videos.

[wraper]

25  FPS is definitely better, but it is noticeably worse than ep. 680 compared previously. Therefore, youtube definitely reduces quality to save bandwidth, however your camera might be at fault too. As I see, all videos were shot at 60 FPS originally. For proper test 25 FPS original shot would be great, of course if your previous workflow included shooting at such FPS. [EDIT: I would say that there is more difference in quality between 25 FPS Test and ep 680 rather than between all those different FPS test videos]. And they all have yellowish tint too. If your camera has function where to set white balance automatically against white object, you could set it to ideal value in just a few seconds against sheet of white paper.  Here is upscaled (magnified) comparison.

25


50


60 original

[wraper]

Difference will be more noticeable if you download them in a one folder and cycle through with Windows Photo Viewer or any other program.

[dexters_lab]

just wanted to bump this briefly, i had not seen Dave's original episode before i started a thread about the same rendering problems in Movie Studio.

We seem to have very similar setups.

I have done a little more testing and digging and it seems that simply, Movie Studio does not use CUDA at all... it's basically not supported any more, Sony stopped supporting it after the GTX 5xx series of GPUs

 

from the sony website:
"NVIDIA recommends NVIDIA Quadro for professional applications and recommends use of the latest boards based on the Fermi architecture."

the 500 series was the last to use the Fermi architecture

[Fungus]

Sony stopped supporting it after the GTX 5xx series of GPUs

 

So? Maybe they abandoned it because they figured out it wasn't actually speeding anything up.

It sounds like a good idea on paper but maybe it simply doesn't work in practice.

[dexters_lab]

Sony stopped supporting it after the GTX 5xx series of GPUs

 

So? Maybe they abandoned it because they figured out it wasn't actually speeding anything up.

It sounds like a good idea on paper but maybe it simply doesn't work in practice.

no i think it did work, the reason why i got all interested in it was i just upgraded my video card, i went from a GTX-550Ti to a GTX-750Ti so it would better support my dual monitor setup... but lost the gpu rendering in the process. The old card was significantly faster at rendering than my (slightly faster, on paper) new one.

maybe sony dropped it so they can redevelop it to use OpenCL better, which is cross platform unlike CUDA