Is it just me and the HLG (Rec.2020) MP4 footage I opted for, or is the Warp Stabilzer much slower in V.23.0?
A 506 frame clip is taking 90 minutes to analyse in proxy display!!! Yes, 'Fast Analysis' is ticked.
FAO Mr & Mrs Adobe; has there been a change to the algorithm or is the HLG Rec.2020 likely to slow it down? I usually shoot in 10bit V-Log (Panasonic GH5S) but this was filming 4x 40minute sets with 4 cameras, so I opted for Rec.2020 as the file size is better compressed.
I've just had a look at task manager as it is doing this analysis, Premiere is only using around 13% of my CPU and 0% of GPU. Can't we get this effect driven by the GPU, or let it use more of my CPU?
Yea this is ABSURD. In 2019 or 2020 I remember the Warp Stabilzier being lightning fast....Davinci Resolve does it in SECONDS. Adobe NEEDS to fix their performance issues. I have an 18 core 10980xe with 128gb of ram and a 2080ti. It BARELY uses the CPU at around 4-10% unless I'm playing back HEVC footage. It certainly gobbles up my ram. The 2080ti is barely used in normal circumstances, but IMO ALL effects should be GPU accelerated.
I've been on Premier since 2003, professionally since 2008. Stop adding features and make your software work right Adobe!
That's bad performance no question. I haven't done any Warp w HLG, so I haven't done what you're doing. Can't say what my rig does as a comparison data point.
One thing that's pretty easy to misunderstand about the process, is that there are two separate parts: analysis first, then actually doing the processing.
The first part, the analysis stage, is CPU only. But after the analysis finishes, the main work is done involving a lot of GPU.
So yea, during that first bit, you don't see the GPU involved because that's just figuring out *what* to do. Not doing anything.
I don't misunderstand the process Neil, and I'm a little tired of hearing that. It's like when Apple said we held our phones wrong when they designed the iPhone4 with the antenna in the side of the casing. My cpu should be as close to fully utilized as possible when using Premiere. I have an $8000 professionally built machine that just chugs in Premiere. That's absurd and we should stop making excuses for Adobe and demand them to fix their peformance issues.
Again, not mad at you, but you are a big name in the forums and I would love it if people like you would stand with us to demand performance! If a 100K of us boycotted Adobe for a month I'm sure they'd feel that multi-million dollar loss.
Not defending them. And don't worry about my feelings, that's silly on a forum as far as I can tell.
Just saying performance at this time is all over the place. And it isn't just Premiere. I work daily in Resolve, and work with a lot of colorists. With massive machines and their performance in Resolve is at times a puzzlement also. It's probably a bit more even. But not by a lot.
There's plenty of R users still on 17 of course. And 18.1 has some cool new stuff but is broken on some rigs.
I think trying to be all things for all users across all media and workflows has some inherent problems. And both Pr and R are having some weird issues.
Of course I hope the devs for both get their own crud sorted. I'm just observing.
Oh ... and with several million daily users, Pr has a massive user base.
I'll take that complaint back - in part. I was stabilizing a simple nest of video and adjustment layer in that project. In a new one, I applied it to bare V-Log footage and the speed was back to normal.
The algorithm just doesn't perform well on nests. If it was a nest of multiple motion videos, this would be understandable. But this case was just one MP4 with Lumertri adjustment layer, and it 'put the brakes on'.
Nesting used to work well in older versions in general across the multitude of computers I've worked on in different offices over the years, but nowadays with some footage types, even ProRes sadly, I''ve had to render a clip out with whatever effets I have on it like a speed ramp etc, then import it back in, then adding stabilizer seems to help the stabilzier process speed up quite a lot. Sometimes cutting it down from 60 minutes to 5 and I'm on a very beefy system. In the past year I've seen this work and then randomly on another clip that's nested it'll work great without that workaround. It'll even work great in one project and not in another. The performance is baffling.
"The performance is baffling."
Sadly, that's the understatement of the year. That is my biggest gripe at the moment with this app and really, with Resolve: they're totally unpredictable across machines and workflows.
When even nesting can cause Warp to go into hyper-sleep speed on a good machine, yowza. That was disturbing to read. I've not nested prior to Warp recently, used to do it every once in a while.
The last couple years using my previous rig, it was aging. So I got into the habit of Warp, render & replace, then do other stuff, simply because it could barely handle Warp. So I suppose if your machine is croaking at nesting then Warp, well ... do what works.
But, even if wise, on a good machine, that shouldn't be necessary.
Speed-ramps are another thing that can be a complete screwup on some systems, not a problem on others. What the .... ?
And how the heck do subtitles blow performance on some rigs as I've seen posted here? Graphics too ... and audio ... and ...
Yea, that total variation in performance is simply ... astoundingly ... baffling.