So when you put color bars into an empty project and render a) with 8 bit settings and b) with 32 bit settings using Sony AVC (1920 x 1080 50i) you get significant level changes like I do? And only with Sony AVC?

I'm particularly interested if there is a problem with my Sony AVC codec (like the too short video event).

In the end I want to render to a 10 bit master (source is 10 bit) and I want to be able to "preview" some aspects of that rendering to an mp4 because I can't see that in vegas ((de-)interlacing, etc.). I would alsways have to change project settings. I'm sure that worked the last time I used that workflow. And this inconsistent behaviour (only Sony AVC??) does only indirectly correlate to the 8 vs 32 bit question.

Also I see a benefit from using 32 bit: I'm using many effects on event and track base. When generating clipping with one effect (maybe slightly, without noticing) I can get those areas back with another effect. I just tried it: doesn't work in 8 bit mode and in the end I don't want to pay attention to small errors in the plugin chain. So grading for 8 bit output has different results with 8 vs 32 bit.

OK, I just did a clean install on another PC: same problem. So that seems to have persisted a bit longer... ;-)

But why does this difference (8 vs 32 bit) only happen with this codec?

It's called a red herring.

Other than screwing up the output and adding an inordinate amount of time to the render, the tests I ran clearly show that grading and compositing in 32 bit float space have no effect with 8 bit source and output. The little dot in the center of the vectorscope images says, "No Difference."

Don't believe the internet hype.

However, the OP says they are using a 10 bit source, and want to export to 10 bit... which your best bet for doing that would be to set Vegas to HDR mode and use an HDR external monitor so you can actually see that 10 bit advantage, then export to HDR HEVC from there.

If you want an 8 bit source export, 8 bit is indeed the best way to go about it. Otherwise, make sure your color spaces are set correctly for your media as well as for your project.

As for why you only notice it with Sony AVC... not sure, but there's other AVC options to use in Vegas, so why stick with that old codec? You can use Magix AVC, happy otter scripts or voukoder to render to AVC using ffmpeg, etc.

However, the OP says they are using a 10 bit source,

I didn't see that in his second post until you pointed it out. There may be a theoretical difference in 8 bit output banding, but negated by Vegas' pattern dither back down to 8 bit, as illustrated below. No free lunch at this diner.

Of course, 10 -> 10 bit begs a 32 bit float grading space.

You should update your VP15.

I used to have the same issue and it was fixed in VP15 build 361.

P.S. The same people bullyed me with the 32 bit vs 8 bit as well :-P

@fifonik Thanks, I'll try that. (But I remember that some newer build had issues with green frames in P2 footage...)

Preserving clipped levels inside a filter chain is for me the main reason to use 32 bit for finishing projects (even with 8 bit sources) - exactly as it's discussed in your thread. (I also prepared an example veg file but couldn't upload it. But I think it's been completely explained in your thread.)

@max8 @fifonik

You are two of several colleagues (including wwjd) who have brought up using the float processing space to trap unrestrained (out-of-gamut) levels from artistic fx and processing filters, as a kind of a levels surrogate, something I learned to do on-the-fly at the timeline level using the output Levels filter.

Rest easy, I will tackle this "off label" use of grading space with direct comparisons for difference, gamma, levels, noise, and rendering time in a future test. I can see a convenience factor to your approach, if not staged in the ideal structure.

Your adaptation is certainly more grounded in plausibility than the "water to wine" theories being spread under the same general umbrella on the internet.

I've got to give this one to you for creativity. For integrating artistic effects outside the normal colorspace boundaries, the ability to apply the compositing layer at linear gamma in float space gives an added tool to your belt. I would be hard pressed to refer to it as grading, however. This is probably not quite what you are doing, so I'd like to see your examples as well. Best.

In another thread I already gave my example and explained my reason for using the 32 bit pixel format. I do not want to continue the discussion about 8 vs 32 with you as it is pointless.