SONY VEGAS PRO 11 and new pc, very slow render times with Neatvideo

[christodoulidesd]

Dear community,


I've been having some off problems lately, and i want your opinions:


Here's my setup:


Sony vegas pro 11 x64
i7 2600k asus-stock-overclocked at 3.4ghz, 8core
Windows 7 ultimate x64
8gb ram gamers edition 1600mhz
msi nvidia gts450 1gb ram ddr5
normal sata-2 hard disk drives at 7200rpm, western digital


I have shot a live band playing with my Sony HVR-HD1000E at 1080-50i
The project is 2 hours long and has the following video fx:


neat video for noise reduction
sony color curves for brightness and contrast
sont hsl correction for desaturating the colors
sony color correction


My problem is that when i render at MPEG2-Blu-ray 1080p it needs 25 hours to export the project!


During render, Cpu load is average 30-40% which is nothing and ram is 8gb-ram.


What am i doing wrong ?


I have gpu acceleration (using the latest option in vegas pro 11 and cuda cores) enabled but even if i turn it off, it doesn't change a thing. Nor does exporting to interlaced makes it go any faster.


Take a look at this Vegas Pro 11 GPU acceleration http://www.sonycreativesoftware.com/vegaspro/gpuacceleration


Since my cpu and ram aren't being stressed at all, i imagine it's either a bad setting in the project / software or the gpu being relatively old and slow, so If i invest 660 euros to get the asus nvidia gtx680 2gb ram which is a beast, ASUS - Graphics Cards- ASUS GTX680-2GD5 http://www.asus.com/Graphics_Cards/NVIDIA_Series/GTX6802GD5/ will i see any worthy difference or will it be money thrown away? Would it be better strictly for video applications (i am not dealing with graphics or animation) to opt for quadro? 'Cause sony shows otherwise in their gpu benchmark site which i linked above.


Any experience with neatvideo?


I would deeply appreciate your inputs.


Thank you!
Demetris

[poisondeathray]

Neat video is very likely the bottleneck - that's why your cpu usage isn't higher - vegas's encoder is waiting for the processed frames so you have wasted idle CPU cycles. Temporal denoising is very slow, even with GPU accerleration.

[christodoulidesd]

Thanks, i suspected 'cause without it, it will do the project in less than 30 minutes . In this case would a far more expensive gpu do the trick or is it unrelated to this matter? Thank you very much for your help!

[poisondeathray]

More expensive, faster GPU will help, but some people think it will be suddenly lightning fast - it won't be. Instead of 1-2 fps you might get 3-4. That's still a huge improvement percentage wise, but still many times slower for encoding than realtime

You can have a look at their website forum for benchmarks on various cards. Your results will differ , because of the settings you use. The higher the temporal radius, the slower, etc...

e.g.

http://www.neatvideo.com/nvforum/viewtopic.php?p=2988&sid=3681d468b7e8c2d77ceba815f2abe30f

Here are some known figures with 1920x1080p, radius=1 (GPU alone):
- GTX 590 13.5 fps
- GTX 580 10.1 fps
- GTX 470 6.4 fps
- GTX 560 Ti 5.8 fps

GTX 570 would be about 8 fps and so 2x GTX 570 will probably give around 15 fps, which is a bit faster than one GTX 590.

All these cards have enough memory to let Neat Video process 1920x1080 frames.

Note these results were not in Vegas, so you will get different results, also the mix of CPU/GPU and different settings plays a role, but hopefully it can give you of some rough idea

Also, if you are doing 2pass encodes, what some people do is encode to a lossless intermediate first with the slow filters applied. That way you endure the slow bottlenecking filter once instead of twice

[dLee]

During render, Cpu load is average 30-40% which is nothing and ram is 8gb-ram.

In Vegas Pro 10, I had to change / decrease the number of Threads to get close to 100% CPU utilization.

Does this Thread / Slow Rendering / Low CPU utilization issue affect Vegas Pro 11?

[poisondeathray]

During render, Cpu load is average 30-40% which is nothing and ram is 8gb-ram.

In Vegas Pro 10, I had to change / decrease the number of Threads to get close to 100% CPU utilization.

Does this Thread / Slow Rendering / Low CPU utilization issue affect Vegas Pro 11?

It does, but it depends on what task you have, and how the project is set up.

The bottleneck in the OP's case is a really slow filter - so it won' t matter in his case how the threads are set up. Did you see that it was ~50x faster without neat video?

My problem is that when i render at MPEG2-Blu-ray 1080p it needs 25 hours to export the project!

Thanks, i suspected 'cause without it, it will do the project in less than 30 minutes

[dLee]

I did see that, I was speculating that the Thread issue might be making things even slower.

Neat Video's rep is: very effective, but slow.

When I researched my problem, I never found a good technical explanation of the Thread / Low CPU utilization issue. I assumed that it was a one time fix, having to do with the number of Threads vs. CPU cache vs. # of CPU cores.

Could the total # of Threads needed by Neat Video and Vegas combined be responsible for at least some of the bottleneck / low CPU utilization?

I can see that the Vegas encoder could indeed be waiting for Neat Video, but why aren't the two of them maxing out the CPU?

Do you have a good link to an explanation of the Vegas Thread issue and how to manage it?

[poisondeathray]

I don't have a good link for the Vegas threads issue, but there are some discussions on the Vegas forum.

The thread issue is actually very complex and there are way too many variables to account for between systems to make broad generalizations - it depends on OS scheduler, compiler differences, different architectures (e.g. AMD will run differently than Intel), Logical (e.g hyperthreading) vs Physical cores, programming differences - some filters are better threaded than others, whether or not GPU is used, GPU settings & speed, and probably a dozen more factors

The best way is to experiment with a given set of conditions then see if it's faster or slower. Higher CPU usage doesn't always equate to faster render times. Sometimes the process scheduler can have conflicts. The end measurement should always be export times

There might be some thread setting that will make his set of conditions marginally faster or slower, but I really doubt it will make a big difference with neat video in the mix

[dLee]

The Vegas forum discussions, at least when I looked at them, were not definitive - there were a couple of fixes that don't work.

The 30 - 40% CPU utilization during filtering / rendering would bother me. I can buy "Higher CPU usage doesn't always equate to faster render times", but I would work hard to get those numbers closer to 100%.

[poisondeathray]

It's not definitive, because there are too many variables to make broad based generalizations. What may work for an intel non HT system might be completely reversed if you toggle HT on. You might decrease performance for other folks - so it's better to error on the side of caution

In this specific case, you get those numbers closer to 100% by eliminating the bottleneck (neat video), or use faster settings

I use it for vdub and AE, but not vegas - if you've ever used neat video in other appliations, it's the same way. Changing threads or memory allocation has no effect. The bottleneck is the primary issue. If NV cannot utilize all resources, you need NV to "fix" or improve that - it's not the host applications' fault

I'd be very surprised if playing with the threads in vegas that he could gain even 1% difference in speed (regardless of what CPU usage says)

[dLee]

So are you saying that CPU utilization is higher when NV is pared with vdub or After Effects?

Or that NV is not particularly good at utilizing the full potential of current CPUs?

Or is this all "apples and oranges" - nonequivalent comparisons?

[poisondeathray]

So are you saying that CPU utilization is higher when NV is pared with vdub or After Effects?

Or that NV is not particularly good at utilizing the full potential of current CPUs?

Or is this all "apples and oranges" - nonequivalent comparisons?

I'm saying it really depends on the variables listed above. You really have to test with a very specific set of conditions and settings. And the results might be completely different for another set of variables. It sounds like a wishy washy reply, but I've done dozens of tests confirming this .

You cannot compare them directly, because even slight differences can yield different resutls. For example the VFW decoder you use in vdub might be different than the on in AE, or vegas. I've seen huge differences between decoder speeds (some are 100% faster!) when connected to a null renderer , but if the bottleneck is still NV

You can do systematic tests on a given project, vary the thread count 1 by 1, vary the memory allocation, and do step wise tests, but the results will flip flop on another project . There are too many variables to account for to make broad generalizations. I cannot stress this enough

For example , even between versions of software , there are differences. In earlier AE versions, the optimal thread count for most projects was without HT (so only physical cores) , but newer versions seem to work better with logical count of cores.

In fact, the ONLY consistent thing is that NV is slow, and any changes you fiddle with will make only minor speed improvements in the grand scheme