EEVblog #698 - GPU Video Rendering

[senso]

Not wanting to be another crying poster, and I know Dave's opinion is in the mood of dont like don't watch..
But, this 50 frames crap is really stupid, but I have crap internet connection, the older 720p 25 frames played fine, but now, its buffering time, A LOT.
And the image quality is down the drain, don't know if its because of the frame-rate, stupid compression by youtube, crap camera lenses, but look at this, there is nothing in focus in that scene, in fact, almost all the last teardown video is out of focus, how is that possible for someone that claims that he sees no difference in image quality 
480 looks almost the same and at least there is no buffering and stuttering.

http://s14.postimg.org/6vbkyu9ps/quality.jpg

[EEVblog]

And the image quality is down the drain, don't know if its because of the frame-rate, stupid compression by youtube, crap camera lenses, but look at this, there is nothing in focus in that scene, in fact, almost all the last teardown video is out of focus, how is that possible for someone that claims that he sees no difference in image quality 

The latest mailbag I shot at a constant iris setting, and yes, it was slightly out of focus, nothing at all to do with the 50fps or anything else. It was an error on my part.
And the image quality is not "down the drain" and not single person has proven it is. Two are now one of two or maybe three people out of 200,000 thousand subscribers who claim this is a problem.
Yes, youtube changes quality and compression and streaming all the time on a whim, I am not in control of that. The quality of video I am uploading is generally excellent and better than I ever have.

[gildasd]

I don't want to state the obvious, but the exemples above are down to artifacts due to Bayer matrix, anti alias filter, lens alignment...
So not much Dave can do apart from getting studio quality equipment, the time needed to process it and the marriage counseling if he was stupid enough to waste so much money on said toys.
And it's silly, Dave is simply future proofing (so that his present videos don't look like his first one 3 years down the line) not trying to do a sequel to "The Hobbit"... For that he would need to be from NZ and find ewe's attractive and improve his cricketing.
Quite frankly, the arguments are down to the level of pixel peeping that remind me of audio foolery - same shit, different part of the spectrum.

[EEVblog]

Dave is simply future proofing (so that his present videos don't look like his first one 3 years down the line) not trying to do a sequel to "The Hobbit"

Correct.
I can remember when Youtube switched to HD (first 720p, then 1080p), the playback quality was awful, but eventually got better as technology progressed, and Youtube simply re-encoded the original uploaded content without anyone knowing.
So it is in every content producers interest to upload the best quality footage they can.

[codeboy2k]

Dave, you used to make the mp4 video available.  That video would be raw, untouched by youtube.

Are you still doing that?  People should be looking at that, not what youtube mashes it into.

[EEVblog]

Dave, you used to make the mp4 video available.  That video would be raw, untouched by youtube.

No, I never have, only the 640x360 podcast version.

[codeboy2k]

Dave, you used to make the mp4 video available.  That video would be raw, untouched by youtube.

No, I never have, only the 640x360 podcast version.

Ah, I was mistaken then, I thought it was the youtube version.

[george graves]

There is a good video that youtube produced in-house, that explains their encoding and archiving method.  It also goes into some detail about serving up different size streams, and how they load balance the servers across regions, popularity, and how they have to store about 10 different versions of ever video up loaded.

I spent the last 10 mins looking for it.

Never mind - found it.   ~3:00 is the fun facts.   

And it's not as informative as I remembered.  I think there was another video from google talking about load balancing.  That's magic to me.

[dexters_lab]

some more testing done today.

i dropped back in my old GTX-550Ti so see the difference between a setup using CUDA and one not.

i7 CPU rendering alone i rendered 769 frames in 60 seconds

GTX-550Ti CUDA enabled rendered 1920 frames in 60 seconds

so, unlike Dave i'm not going to build some expensive Zeon setup, i'm off to buy a GTX-580 or GTX-590

[mariush]

Drop the quality settings in the configuration and you'll have 3000 frames in 60 seconds on the i7 cpu. 

Cuda profile is more relaxed, you don't get the same quality as the cpu encoding with default preset.

[dexters_lab]

Drop the quality settings in the configuration and you'll have 3000 frames in 60 seconds on the i7 cpu. 

Cuda profile is more relaxed, you don't get the same quality as the cpu encoding with default preset.

just changed to baseline profile, 1 slice, draft video quality at 4mbps... rendered 1451 frames in 60 seconds CPU only

i dont want to decrease quality to improve rendering speed

[Fungus]

I was just evaluating Cyberlink PowerDirector for editing blog videos.

From a quick test it encodes video over six times faster than Premiere Pro when you tick the "Hardware Encoding" box.

I don't know how Premiere Pro speed compares to SONY Vegas speed but it might be worth looking into.

notes:
a) Even without the "Hardware Encoding" box checked it was still 2.5 times faster than Premiere Pro.
b) I think it's way easier to use for making this sort of video.
c) It has a killer feature - match up the volume levels of all the video clips with a single mouse click.

[dexters_lab]

thanks for the tip, i'm sticking with Movie Studio

i'm bidding on a GTX570 card right now which should give me back my GPU rendering with a decent boost over my old 550Ti, i would have had a 580 or 590 but i would also have to change my PC case to fit it in!

[dexters_lab]

update:

now with my GTX570 card i'm rendering the same test as before at 2630 frames in 60 seconds, a 37% speed increase.

[EEVblog]

UPDATE:
Assembled my Dual 2.6GHz Xeon E5-2630 v2 24 core machine and first tests show it's SLOWER than my i7 3770 machine at rendering using all the codecs I would normally use
Passmark benchmark shows it is a much faster machine as you'd expect, and all 24 cores seem to be operating 100% during rendering.
Something seems horribly wrong   

[Monkeh]

The assumption that the codecs in question scale linearly to x number of cores could easily be faulty.

It is only 12 cores.

[EEVblog]

The assumption that the codecs in question scale linearly to x number of cores could easily be faulty.
It is only 12 cores.

vs 4 cores in the i7 machine.
The info I could find indicated that the multi-core Xeon machine would do the business.
 

[Monkeh]

Well, what codecs (very, very specifically, if you please) are you using?

[dexters_lab]

Damd, thats a shame after all the effort from you and the hardware donations from viewers.

I am still happy with my reverse upgrade to the gtx-570.

The 570/580/590 cards are getting cheaper all the time so still room to upgrade even more.

[Howardlong]

It appears little has changed in this respect in the intervening 5/6 years since I last battled to get transcoding performing in a predictable manner.

It would be useful if the codecs provided some indication of requirements for their various algorithm choices depending on processor, cores, cache and GPU rather than being a black box.

Back then, I must've blown a fair few quid on hardware in a largely fruitless attempt to improve performance, it turned into an expensive game of pinning the tail on the donkey.

[gildasd]

The assumption that the codecs in question scale linearly to x number of cores could easily be faulty.
It is only 12 cores.

vs 4 cores in the i7 machine.
The info I could find indicated that the multi-core Xeon machine would do the business.
 

Is your codec XEON optimised?
In totally other uses (3D rendering) I've seen major differences in versions of render engines or simply finding the "good" set-up in the vast amount of options that 3D max and the optional 3rd party plug-ins.
- I would redo a complete check of all CPU related options in your software, there might be a "slap on the forehead" misplaced thing to click somewhere.
- Use your Xeon machine as a slave. This is worthless for small efforts, but I've had success using it to render long "flythrough" 3D sequences that would crash either my first or my secondary PC used singly.
- Yell at the intern to make it work.

[Fungus]

- I would redo a complete check of all CPU related options in your software, there might be a "slap on the forehead" misplaced thing to click somewhere.

He's got 3 times as much CPU power (at least!)

If all CPUs are busy and the codec is running slower then it's probably not because of a hidden check box.


I believe the i7 has QuickSync video but that Xeon doesn't (correct me if I'm wrong...)

http://en.wikipedia.org/wiki/Intel_Quick_Sync_Video#Hardware_decoding_and_encoding

Maybe the codecs have been using that without telling anybody.

(It would also explain why using the GPU for encoding didn't make much difference)

[dexters_lab]

- I would redo a complete check of all CPU related options in your software, there might be a "slap on the forehead" misplaced thing to click somewhere.

He's got 3 times as much CPU power (at least!)

If all CPUs are busy and the codec is running slower then it's probably not because of a hidden check box.


I believe the i7 has QuickSync video but that Xeon doesn't (correct me if I'm wrong...)

http://en.wikipedia.org/wiki/Intel_Quick_Sync_Video#Hardware_decoding_and_encoding

Maybe the codecs have been using that without telling anybody.

(It would also explain why using the GPU for encoding didn't make much difference)

No, gpu rendering made a massive difference to me, 37% faster than an i7. you just have to use specific cards for it to work

[Fungus]

(It would also explain why using the GPU for encoding didn't make much difference)

No, gpu rendering made a massive difference to me, 37% faster than an i7. you just have to use specific cards for it to work

OK, I should have said: "GPU rendering with Dave's graphics card". My bad.

PS: The important point was that an i7 might be much better at H264 video than a Xeon.

[Monkeh]

If all CPUs are busy and the codec is running slower then it's probably not because of a hidden check box.

Not if he's stuck with a single threaded decoder or filter bottlenecking things.