EEVblog #698 - GPU Video Rendering

[edy]

I think the point Dave is trying to make is that even a high end GPU card that seems to promise faster rendering does little to improve things. Seems the GPU is one-way... faster decoding for playback but broken (as far as drivers, support, etc.) for encoding. Just not delivering what it promises. The solution is unlikely to be in just using faster hardware to brute-force the video editing and rendering. The solution is in working with "smarter" software and/or an intermediate format that allows quick editing (without doing so much encoding) and just applying the encoding on the final edited video cut... which will be best on a very fast CPU and software made to support the full power of it... not really anything to do with GPU.

Running poorly designed software on a faster machine will be less likely to improve the situation. What you need is smart software and settings to reduce the work needed.

[mariush]

Dave, you can render the video and audio separately (in two steps) and then merge the video and audio using MKVToolsnix ( https://www.bunkus.org/videotools/mkvtoolnix/ ) if Handbrake doesn't have that functionality already.  That should take care of the sync-ing issues you noticed.

Just throwing it out there....  another way to improve the encoding speed would be to use several computers.  (since you have those computers from trash)

* Render the video in Vegas and save it to a intermediary format that's supported by x264 codec. AVI with lossless codecs (as long as that codec is installed on all computers x264 runs on) works great.
* Make the video available to several computers (put it on Windows share or NAS with good network card).. or just copy that big file to each computer so that it's available locally before starting to encode... with gigabit network cards copying at 100 MB/s it only takes minutes to copy tens of GB to computers.
* start several computers and run x264 on each computer to encode a part of the big video file
* merge the segments together at the end.

An example of command line x264 to encode a video that's 90000 frames long ( 60 minutes @ 25fps) :

pc1  : x264.exe  --profile high --tune film --crf 20 --seek 1 --frames 45000 --output segment0.264  c:\path\to\input.avi   
pc2  : x264.exe  --profile high --tune film --crf 20 --seek 45001 --frames 45000 --output segment1.264  network_drive:\\path\to\input.avi

--seek and --frames are the magic parameters. seek means start from frame x , --frames means how many frames to encode.  --crf tells the quality preset, same meaning as the one in Handbrake.

The number after --frames is just informative, it's the maximum frames to encode, if there's only 15000 frames to encode for example x264 will encode that many and won't complain or stop with error messages, so you can basically script everything (write a .bat file on each computer or something like that) and everything will work fine.
 
note: first frame may be 0 actually, not sure, anyway it doesn't matter.

So one part can be run on one computer, the other part can be run on another computer, then put together the two (or several) segments.

Separately, you can render the audio to an ac3 or aac file directly in sony software or render to lossless WAV file and use Handbrake to encode to ac3 or aac..

With elementary streams (the .264 extension), it's possible to just copy one file to the end of another and basically you have one continuous .264 file...

With Total Commander ( windows explorer replacement) I simply select the first video segment (segment0.264) and then go in File > Combine files...  and Total commander automatically finds segment1.264 and segment2.264 and so on and combines all segments into a single .264 file.

Then you can use MKVToolnix or maybe Handbrake to mux the .264 file and the ac3/aac file together into a mp4 or mkv video file.

[Razor512]

Wanted to add a few comments that will show the benefits of 60FPS.

When you record video, to maintain motion quality, you use a shutter speed that is double your output frame rate, so if you do 24 FPS, then you have a shutter speed of 1/48 second, and that remains constant and exposure is controlled using ISO and ND filters (aperture can also be used but that will negatively impact your cinematography.

Due to the shutter speeds used, at 24FPS, you get a lot of blur, and thus with a detail oriented show, if something moves, you will lose a significant amount of detail while it is in motion.
While you can avoid that by boosting the shutter speed, but if you do so, the the video will look choppy, as at less than 60FPS, your eyes can clearly see every single individual frame, and your brain will register every single individual frame (if one frame is not blending into the next), which is why many gamers fine 60FPS to be the minimum for a good gaming experience.

Once you get past the point where your eyes and brain can easily notice and examine individually every single frame, then you no longer have to worry about the shutter speed rules for motion quality. you can do 60FPS at 1/1000s shutter speed, and it will look as smooth as 1/120s shutter speed. With this, depending on the available light, you can use higher shutter speeds which will preserve the full 1080p detail of the object, even while it is motion. If there is little to no movement then is will not matter much, but if the camera is moving, even if it is just do to some vibration, then the higher shutter speeds help keep the detail.

While it is possible to tell the difference between 60FPS and 120FPS on a 120Hz display, you have to look very closely and have to have a side by side view of the 2 frame rates, or have a game repeatedly jump between 60 and 120FPS, e.g., enabling vertical sync, and then using an overclocking tool that allows for remote control of the overclock settings, repeatedly adjust the clock speed enough to cause the card to jump from 60FPS and 120FPS.

You can try this using msi afterburner, and a easy to run game, (I recommend left 4 dead 2, or team fortress 2, or half life 2 deathmatch, as you can easily hit 200-300FPS, and thus can easily use vertical sync and clock speed adjustments to get the game to jump between 60 and 120 FPS, even on an entry level gaming PC.

[Richard Crowley]

Wanted to add a few comments that will show the benefits of 60FPS.

While that is all true, it fails to address the question: what practical benefit is high frame-rate for a talking-head informal video blog?

[gildasd]

Wanted to add a few comments that will show the benefits of 60FPS.

While that is all true, it fails to address the question: what practical benefit is high frame-rate for a talking-head informal video blog?

Because it gives Dave his nerd-on. You want to take away Dave's nerd-on? You won't like Dave if he can't get his nerd-on...
It's his choice, it's his vlog. Much the same way that if he so chooses to wear a tutu and asks us about the best way to dry clean it, he not asking about the validity of wearing a tutu.
Let Dave wear his bloody tutu.

[Richard Crowley]

Because it gives Dave his nerd-on. You want to take away Dave's nerd-on? You won't like Dave if he can't get his nerd-on...
It's his choice, it's his vlog. Much the same way that if he so chooses to wear a tutu and asks us about the best way to dry clean it, he not asking about the validity of wearing a tutu.
Let Dave wear his bloody tutu.

Interesting you should mention that.  I discovered Fran Blanche via the cross-promotion here on EEVblog. She has some interesting videos. But I was rather put off by discussions of corsets, etc.   

[rizzy]

Hey Dave,

it looks as if Nvidia cards are not the best option for Sony Movie Studio.
Here is a link to a benchmark done by Sony http://www.anandtech.com/show/7481/the-amd-radeon-r9-290-review/14

The benchmark shows that even the (early 2010) Radeon 5870  beats the GTX Titan in XDCAM EX rendering and it's a fact that Nvidia cripples its "consumer" cards like the GTX 970 in respect of GPU computing performance.

In my opinion all the gpu acceleration thing is a matter of compatibility and optimization of software and hardware and even though AMD cards seem better supported they won't work (well) with Catalyst 11.3 but will do with Catalyst 11.2 according to the release notes of Movie Studio 12 (see "Known issues"): http://dspcdn.sonycreativesoftware.com/releasenotes/moviestudiope12_readme_enu.htm

I'd write Sony an email and ask for advice for your specific use (1080p-60i) and maybe give a Radeon card a try. Would be really good press for Sony if they could help you with this one.

[thmjpr]

Did you guys did watch the video, he used an ATI HD7850.
That anandtech benchmark does not seem useful in this case as it does not compare to CPU only rendering. It is also likely using the sample clip with heavy special effects.

If you check this one out: http://www.tweaktown.com/articles/6489/exploring-sony-vegas-pro-13-rendering-with-a-visiontek-radeon-r9-290/index5.html

For their custom video (without special effects, ignore the other ones), the difference between onboard video and a 290 is 20%. So the difference between an HD7850 and a 290 should be minimal.

[3roomlab]

Did you guys did watch the video, he used an ATI HD7850.
That anandtech benchmark does not seem useful in this case as it does not compare to CPU only rendering. It is also likely using the sample clip with heavy special effects.

If you check this one out: http://www.tweaktown.com/articles/6489/exploring-sony-vegas-pro-13-rendering-with-a-visiontek-radeon-r9-290/index5.html

For their custom video (without special effects, ignore the other ones), the difference between onboard video and a 290 is 20%. So the difference between an HD7850 and a 290 should be minimal.

well it seems there is more (or better) optimizations in vegas v13 + ATI ... but i suppose it would also be codec specific ... then again, it does show that the v13 is biasing toward GPU handling, so if the GPU driver is somewhat incomplete (likely in CUDA case. or even ATI depending on which codec) ... then biasing encoding toward GPU resources would be wasted. (v13 being optimized for GPU, is now slower in CPU only mode  )

**edit ... aha ... they are rendering some string of sony specific efx (a sony benchmarking file vp11.zip ... a 2.5Gb file) etc ... but what dave is doing is mostly re-encoding of video codecs ... quite likely a different comparison no? in that aspect of purposely encoding, it is likely more streamlined specifics is required in the GPU driver (specifically for x264 yes?no?), just like how major game titles with some unique special GFX features require even GPU driver updates (or so i think)

possibly for backward compatibility, that test file 2.5gB seems to be a vegas 11 test file

[Edgar Amalyan]

I don't think GPU acceleration has ever lowered encoding times. I remember when I had two gtx 690s, and the 6000 cuda cores slowed down an i7. It only works through their own implementations like shadowplay, but the quality is trash compared to x264. I'm fairly sure GPUs will actually do something only if the project has a lot of GPU related tasks, which is near zero in Dave's case.

Anyway, can you output a lossless but compressed file from Sony? I've never tried it so I don't know, but lossless compressed files encode much faster as no work is needed to be done other than compression. Lagarith codec for example. You will get a larger file obviously, and might run into HDD bottlenecks but the quality will be untouched going into handbrake, and will probably be faster.

[Corporate666]

Of course. But it's pretty linear.

Sure, but how much of a performance increase are you looking for?  Or, a better question is... how critical is getting more performance?  Is it just to minimize the time of a tedious task, or is it that you're not producing content you otherwise would without more speed?   There is no magic bullet like a video card or piece of software that is going to turn a 3 hour task into a 1 hour task.  A dual processor 18-core Xeon setup will do that, but it will cost thousands.  None of us know your financial situation so nobody can tell you what the best solution is... but since you're close to the top of the consumer CPU speed rankings, the options are squeeze the most out of what you have (overclock), invest in higher power gear (Xeons), or outsource it (cloud computing).  I don't think there's any other way.

Of course it will, it's simply a matter of cost.
.
.
.

Not an option due to bandwidth caps and upload/download speeds. Pointless to even consider, will always be faster to render locally.

And based on this, your only option is to suck it up and pay the cost of a Xeon machine, or deal with the consumer i7 chips, then.

I personally like 60fps content and disagree with folks that it's pointless.  One could argue that HD isn't needed, or >720p isn't needed... but being at the bleeding edge is always valuable, especially when people can choose to watch it in that format or not.

Another channel I watch is Linus Tech Tips.  They did a funds drive to contribute to them upgrading to a new office.  Maybe you could do that for a new PC to let you produce more 60fps 1080p content?  I would be happy to chip in for that, I am sure hundreds of other EEVBloggers would be too.

Then you can do an old-school 1 or 2 hour video on the build (in 60fps) and it will be money very well spent for the users, and you get a machine that will chew up and spit out your videos in record time.

[EEVblog]

Wanted to add a few comments that will show the benefits of 60FPS.

While that is all true, it fails to address the question: what practical benefit is high frame-rate for a talking-head informal video blog?

Because it looks better to many people. It's smoother and gives a more detailed image.
Many people have said they instantly knew it was a higher frame rate without even knowing I had switched to 60fps.
The difference is real and tangible.

[EEVblog]

but since you're close to the top of the consumer CPU speed rankings, the options are squeeze the most out of what you have (overclock), invest in higher power gear (Xeons), or outsource it (cloud computing).  I don't think there's any other way.

Correct, there is no other way than a faster CPU. cloud computing is not an option.

And based on this, your only option is to suck it up and pay the cost of a Xeon machine, or deal with the consumer i7 chips, then.

The current top end consumer i7 should double my speed.

Another channel I watch is Linus Tech Tips.  They did a funds drive to contribute to them upgrading to a new office.  Maybe you could do that for a new PC to let you produce more 60fps 1080p content?  I would be happy to chip in for that, I am sure hundreds of other EEVBloggers would be too.

Only problem with that is I would have hundreds of people invested in it telling me I'm doing it wrong 

[Howardlong]

Wanted to add a few comments that will show the benefits of 60FPS.

While that is all true, it fails to address the question: what practical benefit is high frame-rate for a talking-head informal video blog?

Because it gives Dave his nerd-on. You want to take away Dave's nerd-on? You won't like Dave if he can't get his nerd-on...
It's his choice, it's his vlog. Much the same way that if he so chooses to wear a tutu and asks us about the best way to dry clean it, he not asking about the validity of wearing a tutu.
Let Dave wear his bloody tutu.

Hallelujah, well now I understand. How could I ever have doubted the value of the nerd-on? Thank you!

Dave, wear that tutu as much as you want, dear. But be careful of getting your twisted pair stuck in the elastic.

[codeboy2k]

Correct, there is no other way than a faster CPU. cloud computing is not an option.

Two faster CPUs.  That's a cloud of two. 

You will definitely get the most bang for your buck with a faster CPU.  After that, add more hosts and farm it out, but it's a local farm. I think that's what people mean when they say cloud computing.  You certainly don't want to send it to Amazon AWS or Google App Engine or Microsoft Azure, and I think your reply indicates you know that already.

[DanielS]

Correct, there is no other way than a faster CPU. cloud computing is not an option.

Two faster CPUs.  That's a cloud of two. 

Or just an i7-5960X - eight cores, sixteen threads in a single CPU with quad-channel memory controller. Overclocked to a conservative 4GHz, that should be about twice as fast as the 3770K is ever going to be. More expensive than two PCs but far more convenient.

[pickle9000]

The current top end consumer i7 should double my speed.

Another channel I watch is Linus Tech Tips.  They did a funds drive to contribute to them upgrading to a new office.  Maybe you could do that for a new PC to let you produce more 60fps 1080p content?  I would be happy to chip in for that, I am sure hundreds of other EEVBloggers would be too.

Only problem with that is I would have hundreds of people invested in it telling me I'm doing it wrong 

To which you reply dickheads.

In this case you end up with two pc's, great for screw ups and downtime. Work on one while rendering on another or having a helper editing at the same time you are.

[tru]

A problem that I see with overclocking is the immense wattage for peak CPU usage vs non-clocked, for example:

http://www.bit-tech.net/hardware/cpus/2012/05/01/intel-core-i5-3570k-cpu-review/7

i7, 3930k peak usage:
stock = 252watts
overclocked = a whopping 525 watts!!

[rizzy]

Did you guys did watch the video, he used an ATI HD7850.
That anandtech benchmark does not seem useful in this case as it does not compare to CPU only rendering. It is also likely using the sample clip with heavy special effects.

If you check this one out: http://www.tweaktown.com/articles/6489/exploring-sony-vegas-pro-13-rendering-with-a-visiontek-radeon-r9-290/index5.html

For their custom video (without special effects, ignore the other ones), the difference between onboard video and a 290 is 20%. So the difference between an HD7850 and a 290 should be minimal.

Yes of course I did watch the video but I thought a 290 is a factor of eight more performant (DP FP) than a 7850 or a GTX 970. To be honest I don't know how video encoding works and what type of calculations are needed but as you can see from the benchmark you should get a significant performance improvement with OpenCL, either with the Intel HD graphics or with a dedicated ATI (although Dave did not).

What I wanted to say is, that there might be a way to speed things up, just by using the right codec with the right driver and maybe with a different piece of hardware. Just ask Sony what they can recommend or if they have any more benchmarks.

[EEVblog]

FYI someone has kindly offer two Xeon E5 2630 v2 2011 processors, so along with a new dual CPU motherboard that should work very nicely!
And it seems the R9 290 card can offer some decent GPU advantages on Sony Vegas and has been proven to work with Sony it seems, if it comes to that.

So I'd need one of these dual CPU motherboards to support the CPU's :
http://www.asus.com/au/Motherboards/Z9PED8_WS/
And presumably an "EEB" case?

[EEVblog]

For their custom video (without special effects, ignore the other ones), the difference between onboard video and a 290 is 20%. So the difference between an HD7850 and a 290 should be minimal.

Well that's the thing, you just don't know until you whack down $600 and try it on your codec with your raw image files, your edits, your transcoder etc..
GPU is just so very pot luck.
CPU on the other hand just works

[DanielS]

A problem that I see with overclocking is the immense wattage for peak CPU usage vs non-clocked, for example:

i7, 3930k peak usage:
stock = 252watts
overclocked = a whopping 525 watts!!

The CPU itself is not going to use anywhere near 500W. That figure is for the whole system including any accessories drawing power from it and the PSU's own losses as well.

An i7-5960X overclocked to 4.5GHz uses less than 200W but total power over the EPS12V cable(s) may have instantaneous peaks close to 300W due to VRM and copper losses.
http://www.tomshardware.com/reviews/intel-core-i7-5960x-haswell-e-cpu,3918-11.html

[justanothercanuck]

So I was looking around and it does seem that Nvidia removed "nvcuenc" from their drivers...  but I've also heard that they replaced it with "nvenc".  Maybe Sony needs to update their program?  I've heard you can unpack an older driver and put the file into the windows directory somewhere in the meantime...  Might be something worth looking into.

Edit: I have a 980 sitting in a box at home, waiting on more parts to put the machine together.  Once I get it set up, I'll fire up premiere and see what it'll give me in terms of encoding options.

[tru]

A problem that I see with overclocking is the immense wattage for peak CPU usage vs non-clocked, for example:

i7, 3930k peak usage:
stock = 252watts
overclocked = a whopping 525 watts!!

The CPU itself is not going to use anywhere near 500W. That figure is for the whole system including any accessories drawing power from it and the PSU's own losses as well.

An i7-5960X overclocked to 4.5GHz uses less than 200W but total power over the EPS12V cable(s) may have instantaneous peaks close to 300W due to VRM and copper losses.
http://www.tomshardware.com/reviews/intel-core-i7-5960x-haswell-e-cpu,3918-11.html

Well, yes but the difference (delta) in my example would be:
i7, 3930k @4.7GHz peak usage:
525 - 252 = 273 watts  (extra wattage due to cpu overclock vs stock)

If you look at the url again:
http://www.bit-tech.net/hardware/cpus/2012/05/01/intel-core-i5-3570k-cpu-review/7

You can see the extreme edition seems to be the exception:
i7, 3960x @4.7GHz
210 - 179 = 31 watts

Perhaps the CPU voltage settings also impact the results, but trying to keep stock voltage with overclocking won't always work - I remember that it is pot luck, and most require an increase to keep it stable which means increase in wattage in peak usage.

And you've cheated by simply showing us the result of the extreme editions, but those cost $$$.

[Corporate666]

The current top end consumer i7 should double my speed.

EDIT:

Just saw that someone has offered up a pair of E5 Xeon's.  That is the way to go moreso than any i7 chip.  Will be very interested in the results.