[Joe Marler]

[Herb Sevush] “[Tom Sefton] “Besides that, it’s easy to learn!”

Only if your a newbie, according to some. Apparently experienced editor’s brains are too full of other, older stuff to take advantage of it’s ease. “

It is commonly said that novices learn FCPX easily, implying that experienced editors have ingrained, difficult-to-change habits. While there is an element of truth to that, overall the novice experience is misleading.

More likely novices learn to do simple things in FCPX easily and don’t encounter the difficult parts — else they wouldn’t be novices. By contrast an experienced editor by necessity will immediately tackle J and L cuts, media management, multicam, etc. I’d like to see the novice who can figure out how to stabilize edited multicam clips on the timeline in FCPX.

[Joe Marler]

[David Sikes] (quoting TheVerge.com article): “macOS is, in many ways, legacy software just waiting for the right moment to be deprecated”

The article makes a big deal about the iPhone 7 single-threaded CPU performance being faster than a one core of a 12-core nMP. So what? My 2015 MacBook Pro also has faster single-threaded Geekbench numbers than 1/12th of a nMP. Does that mean the MBP will replace the Mac Pro?

Re his statement of “Adobe’s Photoshop…being ported to iOS in almost their full functionality”, is he talking about Photoshop Express for iOS? Or some future full-featured Photoshop for iOS?

If that’s what he means, this was mentioned in a May 2015 CNET article, saying it would became available later that year (2015): https://www.cnet.com/news/adobe-discontinues-photoshop-touch-app-previews-its-next-generation/

I don’t recollect seeing a full-featured version of Photoshop for iOS or Android. I further don’t see how this is possible from a UI standpoint. Professional apps like Photoshop, Premiere, etc. have hundreds of menu options — pull down, fly-out, pick screens, etc. In the background the apps are doing heavy multithreaded and memory allocation work involving specific APIs for thread synchronization, data sharing, exception handling, etc.

Such complex apps can be ported between OS X and Windows since those are similarly robust operating systems from both an internal standpoint and UI sophistication. It is unclear if iOS has the back-end sophistication to run a high-end app like these, but from a pure UI standpoint there is no place to put those UI design elements. An app like Photoshop or Premiere Pro cannot be “ported” to iOS and retain anything approaching full functionality. It would have to be totally redesigned.

There was discussion of this point when Microsoft developed their “Metro” tile-based UI. While promoting that as the preferred new platform for apps, the designers themselves admitted such an OS and UI system could not support complex full-featured apps like Photoshop. The “paradigm mismatch” between the two differing UI systems was too great.

If it’s possible to port a full-featured complex app like Photoshop, Premiere, Excel, etc. to a mobile OS and UI, the proponents are free to demonstrate this. Despite all the talk I have never seen this achieved. In Microsoft’s case, the recent Windows versions of Excel and Word were given some “flat” stylistic UI elements, but this didn’t port the apps to a mobile platform at full functionality. Basically the mobile and full-featured Windows versions were whitewashed to have superficial aesthetic similarity, but the mobile version does not remotely approach the desktop feature set.

The writer of the Verge article built an entire viewpoint based on one misleading benchmark program. While it’s inevitable that mobile devices will gradually absorb new workloads as they continue to grow in power and sophistication, I don’t see myself transcoding and editing hundreds of gigabytes of 4k video on a tablet anytime soon.

[Joe Marler]

[Tim Wilson] “do we in fact need another Mac from Apple? Why? Sure, faster, but no matter what they do, somebody’s ALWAYS want faster…Of course, nobody who cares about peak performance is using a Mac anyway, and Apple hasn’t even pretended to aspire to peak status since those hilarious Super Computer ads,”

That 1999 ad dated from the RISC vs CISC era when having a non-Intel CPU was perceived as faster, better, etc. The PowerMac G4 had a theoretical Linpack peak of about 1 gigaflop, which the 1999 ad mentions (“one billion calculations per second”). That was probably at least 2x the performance of the then-fastest Pentium III, but that is floating point which is less often used in most workloads.

Considering integer workloads which are more typical of many commercial apps, even historically Intel wasn’t that far behind. This was because RISC had lower code density which caused the knock-on effect of higher bus bandwidth to achieve the same real-world performance as CISC. It also caused lower hit ratios in the various CPU and data caches, so larger more expensive caches were required to produce the same instruction hit ratio.

Then as decreasing feature size inexorably increased fabrication cost (now close to $10 billion per plant), it became more about capital investment and fabrication prowess than CPU architecture.

But the G4 and G5 were definitely fast in real-world apps of the day, as can be seen from these old Bearfeats benchmarks: https://barefeats.com/pentium4.html

It only made sense for Apple to capitalize on that in their 1999 ads. But that wasn’t because Apple somehow cared more about performance then vs today, rather they had (for better or worse) wedded themselves to a RISC CPU architecture which temporarily delivered better performance than similar Intel CPUs. When that advantage dwindled they transitioned (yet again) their CPU architecture to Intel and are fortunate to have survived that.

Today essentially everyone is using the same CPU families in desktop machines. Those are vastly faster than the so-called G4 supercomputer which did about 1 gigaflop. The Intel Xeon E5 2687W v4 does about 1078 Linpack gigaflops, or over 1,000 times faster. The biggest Xeon has 22 cores. Maybe Apple will something similar in the updated nMP.

However even with such extreme performance available, you can’t economically go any faster using the available parts. IOW even though the Power8 CPU exists, there is realistically no alternative to Intel on the high-end desktop. That said, most people who have edited a lot of 4k could use a lot more performance than the current nMP provides. But it’s less about who is fastest than providing adequate performance for increasingly challenging workloads.

People on Premiere can always go to Windows. For FCPX they are stuck with Apple, so in that sense an updated nMP is vitally important. A top-spec iMac 27 is just not comfortably fast when editing and applying effects to large amounts of 4k, even using FCPX and even using proxy. If an updated nMP does not materialize that will drive users off the Mac which means away from FCPX. You could argue it’s not worth it to Apple financially but that’s like arguing the Corvette is not worth it to GM based on annual sales. There is a halo effect beyond the actual unit and dollar sales.

[Joe Marler]

There is also a separate command-line Neat video benchmark which can be run without installing the product. Like the GUI tool within Neat Video, it runs all combinations of CPU cores and GPU, showing the frame rate achieved by each. Unfortunately it seems hard-coded to 1080p but at least that provides a comparison number.

On my 2015 iMac 27 the best number was 5 CPU cores + the M395X GPU, giving 11.2 frames/sec on 1080p

On my 2015 MBP, the best number was 5 CPU cores + the M370X GPU, giving 8.0 frames/sec on 1080p

https://www.neatvideo.com/nvforum/viewtopic.php?t=1123

[Joe Marler]

[Noam Osband] “I looked in preferences and apparently, as it says, “No computation-capable GPU devices found.” I’m using a late 2013 MPB fwiw.

Any other way to make it faster or just get used to being patient?”

The 2013 MBP had an optional GT-750M discrete GPU. Does yours have that option? If so then apparently Neat Video just doesn’t support it. If your MBP does not have the GT-750M then you have no discrete GPU to use.

However my tests on my 2015 MBP with AMD R9 M370X did not show a major improvement when using Neat Video’s GPU rendering. Therefore even if Neat Video supported the discrete GPU on your 2013 MBP, it would probably not make a big difference.

Neat Video allows tweaking a few performance-related parameters and testing the performance without actually running a pass on your clip. This is under Tools>Preferences>Performance>Check Performance. It gives you the number of frames/sec your selected parameters achieved.

At that same place under “Optimize Settings” it allows running a test which will recommend the best configuration of CPU cores and (if available) GPU. You just press the “Start” and it runs the test, giving you the option of accepting the results.

For many effects using proxy will improve performance except for the final export. However for some reason it appears Neat Video always applies the effect to the original resolution material, even in proxy mode. You can also reduce the number of bits per color channel and resolution under “Optimize Settings” but I don’t know how that affects the results, performance or quality-wise.

Neat Video is just very compute intensive and slow to run. This gets much worse on 4k which has 4x the number of pixels as 1080p, hence the run time is about 4x as long. Besides the above items your only other option is getting a faster computer. This is the kind of thing a Mac Pro with the highest GPU option would help with.

[Joe Marler]

This is a common situation, and there is no easy answer which works in all cases. Sometimes creating a multicam clip may work better, even if you only have one audio and one video source: https://www.fcpworks.com/sync-or-multicam-clips/

Another useful workaround is to simply delete (or don’t import) the short clips that don’t contain value. The presence of these makes sync failure more likely.

By using a clapper board, smart slate app (which usually emit a series of beeps) or even just clapping your hands before the take, this creates a unique sound which makes audio sync more reliable. If done on camera it also facilitates the old standby of syncing it yourself manually. You can sometimes help FCPX by using timeline markers on each track, then let it fine-tune the sync. However this is often impractical for many discontinuous A/V clips with long gaps.

Ultimately, the FCPX ability to reliably sync clips via audio (while useful) is limited when compared to a dedicated sync tool like Plural Eyes: https://www.redgiant.com/products/pluraleyes/

Another issue is when FCPX syncs a continuous audio track to an intermittent A/V track, it spaces out the A/V clips. This is logical but often you don’t want the non-sync’d intervals.

Yet another issue is when FCPX fails to properly sync a clip, it just gives its best guess and places it somewhere with no warning. When Premiere CC or Plural Eyes fail to sync, they raise an error.

There are relatively low-cost timecode-based sync devices called Tentacle, but a couple of these still cost more than Plural Eyes and they add complexity: https://www.tentaclesync.com/

IMO using Plural Eyes is the best approach for cases where FCPX will not sync reliably.

[Joe Marler]

[Jeremy Garchow] “The latest version takes full advantage of the GPU”

Neat Video can use some GPUs, but it did not support the M295X or M395X GPU used in iMacs until (I think) the recent 4.20 version. In the Neat Video plugin interface within FCPX, selecting Tools>Preferences>Performance allows selecting either CPU or GPU rendering or both, and informs you whether GPU rendering is available for your particular GPU.

On my top-spec 2015 iMac 27 with 4Ghz i7-6700K and M395X, using CPU rendering took about 5.1 minutes to process 8.8 *seconds* of 4k material. Using GPU rendering improved this to 3.3 minutes. Using CPU+GPU rendering it took 2.9 minutes. This is using the latest 4.20 version of Neat Video.

So the main lessons are:

(1) Update to the latest version of Neat Video. Within the Neat Video interface, you can check for the latest version with Tools>Preferences>Check Now.
(2) Likewise use Tools>Preferences>Performance to select GPU or CPU+GPU rendering.

We can see from the above that some effects are very compute intensive. While GPU rendering can improve things, on 4k material it will not become rapid. I have also tested this on my 2015 top-spec MacBook Pro which has a 2.8Ghz i7 and a M370X GPU. In that case GPU rendering only slightly improved Neat Video performance. The M370X is not a fast GPU but it’s only competing against a 2.8Ghz CPU so that’s a pretty low bar.

It’s a similar situation with Digital Anarchy’s Flicker Free, which can miraculously remove flicker, such as from fluorescent lights or exposure stepping. However it is extremely slow to run, taking 4.6 minutes for 8.8 seconds of 4k material. It does not use GPU acceleration but if Neat Video is any guide doing so might only improve things by 2x, which would still be very slow. So the goal is get the video right in the field and minimize use of complex, time-intensive fixes in post: https://www.digitalanarchy.com/Flicker/main.html

I have tried Photon Pro, which is much faster and cheaper than Neat Video but the results are not nearly as good: https://itunes.apple.com/us/app/photon-pro/id912539173?mt=12

[Joe Marler]

[Noam Osband] “I’m going through the selects…have highlights, and now I want to go through the hours of highlights…I would love to have everything laid out in the project chronologically…”

FCPX is optimally designed to do this in the Event Browser — before adding to a timeline. That is where you make your selects via either favorites or keywords. This is a nice feature since the selects can be ranges, not just clips. Then you can sort those by date (gear icon>sort by content created). Then you add those to the timeline.

If you have already added them to the timeline, you can simply filter on Used Media then select/copy/paste those clip ranges to a new project, and they will retain the sorted chronological order.

If you ever find the clip creation time is in error (e.g, camera clock wasn’t set), this can be adjusted within FCPX by selecting the clips in the Event Browser in List View, then Modify>Adjust clip date and time. It can optionally also adjust the creation date/time of the file itself.

[Joe Marler]

[Matt Whittemore] “I need to get a second iMac and was looking at the refurbished ones (the 5k ones with the Fusion drive). They’re all i5’s….I’m wondering if I should just get a new (non-refurbished) iMac with an i7 or if the i5 (3.2GHz, 3.5, 3.4) will be good enough…
I’m running FCPX, mostly 1080p but will probably need to edit 4k in the near future. I do lots of multicam with footage from C100’s, 5D III’s, GoPro…”

For 1080p almost anything will work. For 4k almost nothing is fast enough. Multicam 1080p H264 may require transcoding to proxy for smooth performance except on the highest-end machines. Multicam 4k definitely requires proxy, which in turn means your storage requirements are much larger since proxy is roughly doubles the storage requirement vs camera-native H264, but that’s better than optimized media which is about 8x the size.

Re i5 vs i7, in narrow cases the i7 hyperthreading feature can improve performance. I used the 3rd-party CPUSetter utility to disable/enable hyperthreading, and this made about 30% difference in FCPX export speed to H264. In other workloads such as Lightroom import and preview generation, it made no difference.

However the i7 on current top-spec iMacs is 4Ghz, so that is 21% faster (excluding turbo boost which is unpredictable) for all CPU tasks than the 3.3Ghz i5 version — independent from any hyperthreading benefit.

Nice as the retina screen is (and I wouldn’t want anything else) there is some difficult-to-quantify performance penalty from the high resolution. In side-by-side testing of my 2013 top-spec iMac 27 (non-retina) and 2015 top-spec retina iMac 27 in both Lightroom and FCPX the retina version can be a little sluggish in some tasks. This might be the timing controller (TCON) or bit-block-transfer overhead. However this is a moot point since getting the top iMac 27 requires a retina screen. But if you can get a good deal on a 2013 iMac 27 with i7, those work very well and are affordable. But the retina screen is dramatically superior, especially for text.

Re Fusion Drive vs SSD, I’ve tested that extensively and don’t see much difference in real world workloads, whether it is application launch speed, boot speed, or smaller FCPX projects that would fit on the system drive. With multicam 1080p and especially 4k, you generally have your content on an external drive so the system drive isn’t doing much. IOW the high I/O rate of SSD doesn’t help much since you aren’t pushing it that hard anyway. I’ve seen lots of people get carried away with specs, get a too-small SSD system drive, then hit problems because they ran out of space.

The main issue is 4k. Really this should not be a shock since it’s 4x the data and would require a computer 4x as fast to provide the same perceptual performance on 4k vs 1080p. A fast GPU helps but a lot of the code paths (ie H264 encode/decode) are not amenable to GPU-style parallelization.

This creates a situation where we’re knocked back years in time and again have to transcode everything for adequate performance. This hurts after getting used to editing camera-native files at DV, 720p and 1080p. It also implies you’ll need plenty of fast I/O, not some puny USB bus-powered 5400 rpm drive.

[Joe Marler]

[Craig Seeman] “There is this from Apple
Final_Cut_Pro_X_for_Final_Cut_Pro_7_Editors”

That’s the kind of thing I mean. It is mostly vague references to a “completely new trackless editing paradigm designed for speed and flexibility”. Nothing specific about optimal workflow using the new paradigm, no background info on why the change to such a radically different system, etc. They don’t even describe what the skimmer is, how it’s qualitatively different, or what new workflow it enables.

Imagine in 1984 if Apple merely described the groundbreaking MacOS by summarizing what each menu choice did, vs elaborating on *why* it’s different, what led up to the new approach, and how investing in the new approach is worthwhile.