Blackmagic Forum

Is the new stabilizer supposed to be much slower than in 12.5? I have an i7 and 1060 GPU... but analyzing a 1:45 clip is taking far far longer than I would have ever imagined.

According to van Hurkman's What's New in Resolve 14, it is somewhat slower.

In the v14 Tracker menu's Stabilizer module, there is a pulldown menu in the lower right where you can decide what aspects you want to stabilize. Also, when in the Stabilizer module, you can select Classic Stabilizer from the options menu, and that'll make it the way it was in v12.

I was excited about the improved stabilizer in 14, but have found it so incredibly slow (analyzing) that it's virtually unusable. I've got an i7 and a 1060 6GB GPU, yet when I've just upgraded from Resolve 12.5 to 14b5, the stabilizer is SLOWWW to analyze a 1m45s clip. In 12.5, although it was different, it seemed to analyze it quickly enough. This chugs along 1% at a time...

Just did analysis on a 17s clip. Am on Win10
- Averaged 20% CPU usage
- Average 1.6GB RAM usage
- Took 3m46s

With the classic stabilizer, it took 16s to complete the tracking. Surely it's not THAT much slower is it? That's a surefire way to really kill my productivity

Hmm, it is considerably faster than Premiere's Warp Stabilizer for me (on a Mac).

And about as good…

But I admit that it's more than 3 times slower than the classic stabilizer.

Kaitlyn,

Have you tried running GPU Shark to check to see how your GPU is being used?

PeterMoretti wrote:Kaitlyn,

Have you tried running GPU Shark to check to see how your GPU is being used?

nope hadn't even heard of it. Just download/open it and look for... what when it's doing the analysis?

Yup, that's it. It's *very* lightweight and will tell you how much of your GPU is being utilized.

http://www.ozone3d.net/gpushark/

PeterMoretti wrote:Yup, that's it. It's *very* lightweight and will tell you how much of your GPU is being utilized.

http://www.ozone3d.net/gpushark/

This was taken when it reached 2%:
http://imgur.com/a/9lq1G

What I suggest you do is open GPU Shark and the Windows Task Manager Performance tab.

Run each stabilizer and see how the GPU, CPU and Memory usage differs.

BTW, what does look odd from what you posted is that the fan speed is zero but the temp is at the maximum. That does at least *seem* odd.

PeterMoretti wrote:What I suggest you do is open GPU Shark and the Windows Task Manager Performance tab.

Run each stabilizer and see how the GPU, CPU and Memory usage differs.

BTW, what does look odd from what you posted is that the fan speed is zero but the temp is at the maximum. That does at least *seem* odd.

Here is the result of the new stabilizer around 15% (a quick search online seems to indicate the fan will only come on around 60 degrees):


and the classic stabilizer about halfway through the clip.

That's really odd, the new stabilizer which is slower is using less resources than the old one which is faster.

Maybe click on Open Resource Monitor at the bottom left of the Task Manager, and see how the cores are being used, but I have to believe they are all being hit essentially equally.

BTW, what codec, format, frame size and rate is the clip?

PeterMoretti wrote:That's really odd, the new stabilizer which is slower is using less resources than the old one which is faster.

Maybe click on Open Resource Monitor at the bottom left of the Task Manager, and see how the cores are being used, but I have to believe they are all being hit essentially equally.

BTW, what codec, format, frame size and rate is the clip?

It was gopro 4k 30fps, h.264 mp4

Are you using the Studio version of Resolve?

PeterMoretti wrote:Are you using the Studio version of Resolve?

Nope just the free beta. Thinking about picking up studio once the dongle isn't necessary. I have heard the Studio handles h264 better, but does that apply to even anlayzing footage for stabilization?

I'm testing the stabilizer in 14 and I think it's also much slower than I remember from 12.5.

20 seconds of 4k mov from Canon 1DX Mark II takes 3 minutes to stabilize.
This is on a 2x14 3.1GHz Xeon with 2xTitan X cards Windows 10.
When I use HWMonitor it reports that only 6% of the CPU is used and 0% of both GPUs.

Are these normal numbers or am I doing something wrong?
Edit. I tried the old stabilizer and it's much faster, around 5x. I can also see that it's using at least 10% of both the GPU cores. I also get better results in the clips I tried, but maybe Im not using the new one correctly, I dont find so many settings.

/Andreas

Kaitlyn McLachlan wrote:

PeterMoretti wrote:Are you using the Studio version of Resolve?

Nope just the free beta. Thinking about picking up studio once the dongle isn't necessary. I have heard the Studio handles h264 better, but does that apply to even anlayzing footage for stabilization?

Kaitlyn, if you want to dropbox a file, I'd gladly take a look at it and test it on my machine.

The stabilizer in Resolve 14 is single threaded. The single thread that's working is pegging one CPU core.

So, at the moment, single core performance is all that matters to Resolve's new stabilizer. Hope you weren't holding out for a shiny new i9-7980XE or Thread Ripper to make that faster.

Jason, how did you came to that conclusion? I'm looking at the Performance Monitor in Win 10, and I see all CPU cores being used.

PeterMoretti wrote:Jason, how did you came to that conclusion? I'm looking at the Performance Monitor in Win 10, and I see all CPU cores being used.

I used Process Explorer (from Sysinternals) and looked at the threads. The thread whose stack showed it was in functions whose names indicated stabilization was busy, but there was only one such thread. There were other threads involved in decoding and other peripheral work but I just saw stabilization being done in one thread.

For whatever reason, Windows liberally reschedules a given thread across available logical processors so just looking at perfmon or other tools that give a plot per core view can be misleading. The telltale sign of this is when no single core is pegged during CPU intensive tasks. (I don't exactly understand why Windows does this as I'd have thought processor affinity would be better for CPU cache efficiency, but it definitely does this. Maybe it's for thermal management.)

So I was a bit misleading in my earlier post. It's not pegging one single core as such (because Windows' scheduler will bounce the thread around) but it is constrained by single core performance due to stabilization being done in only one thread.

Has there been a solution to this? BMD should quickly address this in the next update.

I can confirm: only 20% CPU usage and GPU is nowhere near being exhausted...

Is the 14 Stabilizer much better than the 12.5 one (I haven't much experience with it), i.e. is v14 worth the huge extra time?

Cheers

Myself Im just using the old stabiliser. from what I can see the new one also lacked some functionality that I use. For example I will often create key frames in the stabilisation and switch what point Im using as reference points. Dont think you can do that with the new stabiliser.