PPro CS5 - CUDA quality vs intermediate codecs

[John Farnsworth]

Am I correct in assuming that although CUDA accelerates the editing and display of H.264, it does nothing to reduce its lossiness? If so, it would appear that you still need an intermediate codec such as Cineform's to maximize the quality of the final, edited video. True?

[David Newman]

We believe so, and even editing native H.264 has it own quality implications. See my blog entry on the subject: CineForm Insider: Why use an intermediate for DSLR video?

[John Farnsworth]

Thanks for the info.

By the way, wouldn't editing H.264 without an intermediate codec create the exact same file whether you use a CUDA capable card or do it "natively" (slowly) on the CPU? I'm under the impression that CUDA does exactly what the CPU would normally do, just much faster since the algorithm is in hardware. As an analogy, a hardware based square root calculator gives precisely the same answers as software running on the system CPU.

[Pete Bauer]

If the same algorithms are used, that's true. But this blog article discusses how they're taking advantage of the speed of parallel GPU processing power to use more calculation-intensive but better algorithms. So I would guess that the benefits of CUDA and an intermediate codec would be more or less independent and additive.

[John Farnsworth]

Thanks for the excellent link.

As I read it, IF you set PPro (without CUDA) to its Maximum Quality setting, its algorithm is the same as CUDA's algorithm. Other quality settings, of course, use algorithms different from CUDA's. Unfortunately, the article 1. limits the discussion to scaling, and 2. doesn't mention what codec is being edited. I'm specifically interested in editing H.264 (or an intermediate codec) since I use a Canon 5DII.

At this point it still appears to me that CUDA provides the same exact results that PPro can provide at its Max Quality setting. And this points me towards using an intermediate codec, such as Neoscene, for what I understand to be its higher quality final output. I still don't feel like I have a complete handle on the situation, though. Still trying to figure things out.

[Peter Manojlovic]

Hey John....

You've got the opportunity to do your own testing...
I believe Cineform still has free trials...

There's good and bad in both choices...But i'd put my money down on a workflow that provides stability. Above and beyond anything else...
To do this, play with nested sequences and numerous effects...Colour correction, opacities and other filters that tax the system would be a good start.
Spit out your final exports with AME (or whatever you choose), and see if there's any hiccups between native files using MPE and Cineform intermediate..

Good luck!!!

[Robert Young]

Originally Posted by John Farnsworth

Am I correct in assuming that although CUDA accelerates the editing and display of H.264, it does nothing to reduce its lossiness? If so, it would appear that you still need an intermediate codec such as Cineform's to maximize the quality of the final, edited video. True?

A case can be made for the notion that if you stay within PPro for all of the editing and only render final delivery formats (BR, Flash, etc.) from the edited timeline, that native AVCHD editing will produce a very high quality delivery product.
On the other hand, some advantages of using Cineform DI:
1) The CF codec is 4:2:2, 10/12 bit, and is very "lossless"
2) It is less demanding of system resources for editing/previewing/rendering/etc.
3) If you need to render part, or all of the timeline to a CF master .avi, the image quality will be preserved.
4) The master .avi can be exported to other apps for further treatment, to apply global changes, or transcoded by third party apps.
It really depends on what sort of projects you are doing and what kind of workflow you will be using as to which approach would be best.
I've used CF since the early HDV days and still find it to be pretty much bulletproof, but lots of people are editing in native formats in CS5 and are happy with the workflow and results

[Brian Parker]

I would agree with Robert's points. It's impossible to increase quality of files just by converting to an intermediate format. You can't get better than the orignal file. So creating intermediates for a straight, all PPro edit would seem to be an unnecessary step.

If however, you need to export a file to send it out to another program, and then need to export from that program to bring back into PPro, then Cineform would be the best non-lossless solution.

Or, if you want to do color correction on the files before the edit using Cineform's first light, then you would be saving lots of cpu during the playback on the PPro timeline compared to editing native files and using PPro's color correction tools.

Finally, when it comes time to export, you will likely be going down a step in quality anyway. Most likely you will be reducing quality to create a DVD or web format for delivery. However, as an archival format to use as a master for future exports, or re-edits, Cineform once again shows it's usefulness as a high quality relatively small file size solution.

The caveat in all this being of course that the After Effects exporter, and Premiere Pro implementations of Cineform aren't finished yet (for the mac at least). So editing native files with the "Maximum quality" checkbox on, or using CUDA would be the best solution if your computer is good enough.

[David Newman]

Brian,

Please review my blog post here (CineForm Insider: Why use an intermediate for DSLR video?) on why converting to an intermediate does increase the quality. While this does seem defy logic, that are valid technical reasons for this. CineForm has always prefiltered the image to correct/minimize artifacts or errors that the MPEG/H.264 deocoders do when presenting the data. NLEs assume the decoder ouput is correct, when it rarely is (certainly not the way the NLEs use it -- they don't handle native 4:2:0 well for example, covered in the blog.) By up converting the image then storing it an immediate at 4:2:2 or 4:4:4 with significantly lighter compression, you will get an image that is more correctable than the source, it can be pushed more before the artifacts of the source compression are visible.

[Brian Parker]

Hi David,

I read your post but I still thought that editing natively with the "Maximum quality" checkbox applied in PPro would have given equal editing experience to editing cineform since it too would be going through a 10bit stage in the encoding/decoding. I freely admit that my knowledge comes mostly from what I read on these forums, and blogs, and defer to your expertise obviously if you say that cineform offers better chroma processing when scaling video or applying effects etc.

[David Newman]

True you get deeper pixel processing options in CS now, that is cool, but if you core chroma processing is lacking (as it is with all NLEs), extra bit-depth will not fix that. Notice I didn't need to mention bit depth once in my blog entry. Images in the blog should speak for themselves -- although my job in to make image look good, so what I see as a glaring difference can be too subtle for some. :) I was only pointing that your assumption that it is "impossible to increase quality of files ...", when our customers daily experience is otherwise.

[Robert Young]

I realize that this is a very non-technical observation, and maybe it just speaks to my limited experience with native format editing, but...
When I start a big project with CF, it's like heading out on a cold rainy day in a Toyota Landcruiser.
I know it will start, and I know it will get me wherever I need to go, no matter what the terrain.
File conversion times are so quick on my current system, and after that... priceless :)