Dropped Frames on new M2 MacBook Pro Max - not using processor power

Thanks a lot for your help. Just hat a call from apple support.. and I was shocked how few a "final cut expert" there knows about editing. The discussion was one hour and he was constantly going back to: "whats your problem? FCPx recommends you to edit with proxies, do so an everything is fine"... Thats what you buy an M2 Max for?

joema wrote:

Also, please restate your codec specifics. Initially you said XAVC-HS (HEVC) @50p for (I think) 25i delivery, now you're saying XAVC-S (H264). Nothing wrong with that, since the FX3 won't shoot 25p in HEVC, but it will in XAVC-S.

I just tested Sony 4k/25p XAVC-S 10-bit 4:2:2 and 4k/50p XAVC-HS 10-bit 4:2:2 on a 1080/25i timeline on an M1 Ultra Mac Studio running FCP 10.6.6 and Ventura 13.4.1. With one LUT applied, it seemed to run OK with no playback lag. That was without background rendering. I had the drop frame indicator enabled, and it never tripped.

Sorry for being not precise enough.. I am always shooting in UHD 50p if possible because I have to output 50i. To preview both the footage and the timeline without stuttering I am editing in 50p.
The footage here is now in XAVC-S 10 bit 422 UHD 50p, so its H264, I have also XAVC-S 8 bit 420 UHD 25p from a Sony a7 III which runs MUCH faster. This feels like editing with proxies, skimming over the footage or the timeline updates instantly.
Do you see a difference when you edit with UHD XAVC-S codec or with proxies?? Is it as snappy?

Right now I will continue to edit without a anything connected.. Just my external SSD and it seems a little faster, still not as snappy though...
Maybe the cause could be that two of my software-monitor
Maybe I just expected too much from this "monster" m2 max.. I am still wondering why It uses only a few percent of processor power while everything is lagging.

If you turn on the dropped frames indicator while normal editing, does it warn you sometimes? Or never at all?

I found another cause for some of the dropped frames warnings: When you hit nudge or delete a part of the timeline on the left side of the playhead the whole timeline shifts while playing. This causes FCPx to "think" there must have been skipped frames and I get the warning.. Strange I didnt see this in former versions, or I always had the indicator deactivated.. Still the slow clip skimming in the browser remains.. but I guess I have to cope with in 10 bit it as the 8bit footage is much faster.

4k/50 10-bit 4:2:2 XAVC-S is a difficult codec, but it should run OK on an M2 Max MacBook Pro without proxies. On most Intel machines it is very difficult to edit.

Skimming clips in the Event Browser should not be slow or sluggish on an M2 Max. Can you take an iPhone video of this and post it somewhere we can see?

I re-tested the above codec on my M1 Ultra Mac Studio and it ran basically OK in a 4k/50 timeline. There is a little difference if the viewer is in "Better Quality" vs "Better Performance". You could try "Better Performance".

If I use the JKL keys to rapidly switch from 8x forward speed to 8k reverse, it can have a little lag. I also tried it in Resolve Studio 18.5 beta 3, and it seemed a little smoother.

Re why do monitor tools like iStat Menus or Activity Monitor not show 100% CPU or GPU, that is because it's likely bottlenecked on hardware accelerator decoding or even I/O. Of those two, decoding is more likely. There are no monitors for hardware acceleration utilization.

Even though the hardware acceleration for difficult codecs is much better than Intel machines, it's still not as smooth as an All-Intra codec. That is why everything my team shoots is either XAVC-I from the FX6 or some other All-Intra codec. The MXF container format of the FX6 also has much better metadata support for timecode, color and various camera parameters.

When I had the iMac Pro, I had to transcode all 4k XAVC-S content (whether 8-bit or 10-bit) to proxies. Now on the M1 Ultra, I can edit that directly without proxies, but we rarely shoot that anymore.

Thanks! heres a link:
Its just the clips with a lut on a 50p HD timeline:
wetransfer.com/downloads/91c9cc96da79874...0230628165658/ca35f4
On the macbook alone without monitors, the difference is not huge.

It runs okay.. its just not perfectly snappy like proxies are, but I guess thats not to be expected.. I was mistaken by reviews that claimed "simultaneous playback of 7 XAVC-HS UHD 10 bit files" But as the processor is not used too much I think it will hold the performance till the end of the project, even if i apply color correction and some effects

And I think in the first stage where I have to skim through hours of footage for a documentary I still prefer proxies.. Its just so much faster..

Its about the same lagging if I just scrub through the original media in the Quicktime Player (copied to internal SSD). Footage from an Iphone 12 is blazing fast like proxies in quicktime and in fcpx although its HEVC UHD 10 bit as well (although only 420 no 422) - seems like apples own codec is highly optimized or its a bandwidth problem - even on the internal SSD.

FinalCutdown wrote:

Its about the same lagging if I just scrub through the original media in the Quicktime Player (copied to internal SSD). Footage from an Iphone 12 is blazing fast like proxies in quicktime and in fcpx although its HEVC UHD 10 bit as well (although only 420 no 422) - seems like apples own codec is highly optimized or its a bandwidth problem - even on the internal SSD.

FCP and macOS are both optimized to use ProRes or Apple-flavored H.264/H.265 codecs. That's why iPhone footage plays back well. Third-party camera codecs are often different, even though they may use some form of H.264/H.265 codec.

I've never seen a LUT affect performance... I believe this is mentioned in Ripple Training's course on using LUTs in FCP as why it is an advantage in comparison to using color grading/correction tools in FCP (if the original footage has good exposure, white balance, etc.), i.e., to use in lieu of grading for fast-turnaround projects, etc.

And, I do wonder about the external monitors/displays with the MBP, and if they are affecting performance (by using up a lot of the shared sytem RAM, or other factors/issues). In the past, dropped frames often meant drive I/O wasn't fast enough...

FinalCutdown wrote:

Its about the same lagging if I just scrub through the original media in the Quicktime Player (copied to internal SSD). Footage from an Iphone 12 is blazing fast like proxies in quicktime and in fcpx although its HEVC UHD 10 bit as well (although only 420 no 422) - seems like apples own codec is highly optimized or its a bandwidth problem - even on the internal SSD.

I got your iPhone video showing the M2 Max screen playback. Thanks for showing proxy mode, better quality and better performance.

IMO that is normal. It is roughly similar to my M1 Ultra. 

Every codec is different. Even two 4k/50p 10-bit 4:2:2 H.264 codecs can be drastically different internally. E.g, the GOP length can differ, or the number of reference frames can differ. Certain Long-GOP codecs use "Open GOPs", which means each GOP references data in another GOP, which means hardware accelerators cannot process those in parallel. It is inherently sequential.

Other codecs use "Closed GOPs", where each one is stand-alone and can be decoded without referencing other GOPs.

Those things can have major ramifications about decode performance. 

Within a GOP, the decode/encode algorithm is inherently sequential. It cannot be parallelized or improved by GPU processing. It can only be improved by "fixed function" logic (IOW hardwired circuitry) which runs parts of the sequential algorithm faster. A video accelerator cannot be programmed with a different algorithm like a GPU can. When the logic is designed it can only work for certain types of certain codecs. There is no such thing as an accelerator that works equally well for all the hundreds or thousands of H264 variants.

If you took five different 4k/50 10-bit 4:2:2 H264 codecs from different cameras, there would likely be a significant performance difference in decode/encode performance on a given NLE and hardware platform. The codec designers trade-off many different things.

A hardware video accelerator is a full-custom ASIC. It takes years to design one. If a camera company adds a new codec or tweaks an existing codec, and if they do not communicate their design details to Apple, Intel, etc, years in advance, people buy the new camera then find out some codec doesn't work, or is slow. 

Before Apple Silicon this was exacerbated by Intel's sluggish support for hardware video acceleration. E.g, almost no Xeon CPU has QuickSync, which is why the 2013 Mac Pro was so slow on H264 and H265. The 2017 iMac Pro used a Xeon CPU, which forced Apple in a stop-gap measure to cram a video accelerator on the T2 chip, but it didn't work that well.

Your camera will shoot XAVC-SI which is All-Intra. You could try that. It is similar to the XAVC-I our FX6 cameras use. 

joema wrote:

FinalCutdown wrote:

Its about the same lagging if I just scrub through the original media in the Quicktime Player (copied to internal SSD). Footage from an Iphone 12 is blazing fast like proxies in quicktime and in fcpx although its HEVC UHD 10 bit as well (although only 420 no 422) - seems like apples own codec is highly optimized or its a bandwidth problem - even on the internal SSD.

I got your iPhone video showing the M2 Max screen playback. Thanks for showing proxy mode, better quality and better performance.

IMO that is normal. It is roughly similar to my M1 Ultra.

Every codec is different. Even two 4k/50p 10-bit 4:2:2 H.264 codecs can be drastically different internally. E.g, the GOP length can differ, or the number of reference frames can differ. Certain Long-GOP codecs use "Open GOPs", which means each GOP references data in another GOP, which means hardware accelerators cannot process those in parallel. It is inherently sequential.

Other codecs use "Closed GOPs", where each one is stand-alone and can be decoded without referencing other GOPs.

. . .

That's a wonderful explanation of various "flavors" of codecs used by some cameras. Thanks Joe!

A related point: the FCP workflow is designed to spend much of your initial time skimming clips in the Event Browser and marking ranges from keywords, favorites, and rejects. You then apply filters or create smart collections and are left with a smaller, distilled version of the material.

The software code path for skimming in the browser seems more optimized than the timeline. Used this way, it is very fast, and even borderline sluggish codecs seem smoother.

You don't want to dump a bunch of stuff on the timeline and start sorting it out. FCP is not designed to work that way.

Thanks a lot for all the details and for watching the video. So I will stay with proxies first and later on go to originals for finalizing. It works somehow ok. In premiere the performance is even worse, if quicktime scrubbing is choppy, I guess every LNE will struggle.
I think this is the reason why in mosts reviews M2 Max and M2 Pro have identical performance in video playback, the only difference is export and rendering. I waited for the m2 for so long because they promised improved video playback acceleration for more codecs and resolutions..
My Macbook mid 2012 had an intel CPU with quicksync, thats why I could keep it that long and edit Sonys XAVC-S 50p HD 8 bit and XAVC-S 25p UHD 8 bit natively. But 10 bit codecs were an absolute no-go there.

Thanks again and again!
Thats exactly the way I work - most of the time in the browser. Even there its a little slow..

btw: Premiere doesnt even play back one clip in real time..

I just received an os update for pro video codecs including XAVC. I am not sure but I think scrolling those videos feels a little more fluid now. does anyone have the same experience?

By the way. No matter if I play back the timeline, render or transcode files or export my project the M2 never uses more than 10% of its processor power. Seems to my that still FCPx is not fully optimized for apple silicon.

What version are you using? There are widely reported playback problems in 10.6.7 that were not seen previous versions.

Currently 10.6.6 - so I should wait with an update?

Yes. Don't go to 10.6.7.

Thanks, I will stay with 10.6.6 for some time now. By the way: Sometimes I was using the "downtime" function of apple to keep me away from distracting apps. I think it causes FCPx to slow down or hang as well.

You mean the Focus function in Control Center? How did you set it up?