Transcode to optimized media to improve performance of effects?

I commonly see recommendations that transcoding to optimized media is vital to achieving good performance on effects. For example on this page Larry Jordan says:

"For all editing where you are adding effects, color grading....optimize all camera native media." (italics mine)

However in my tests, transcoding to optimized media does not generally produce better effect performance than camera native H264 media. These tests used UHD 4K H264 from a Panasonic GH4. Tested effects: stabilization (analysis and render), Gaussian blur, Neat video NR, Aged film effect, and retiming with optical flow (analysis and render).

In no case did optimized media produce faster effect rendering than just using camera native H264 media.

Rather than optimized media, transcoding to proxy media did produce improvements in some cases but it was generally less than you might expect.

Is there any basis to this commonly-repeated recommendation that using optimized media somehow improves performance of effects? The problem is it takes about 8x the space of H264, and if people are blindly transcoding everything on input because of these recommendations, they are wasting a lot of space and time.

Config: 2015 top-spec iMac 27, OS X 10.11.2, FCPX 10.2.2

Hi Joe,

I don't know where Larry got this information, but it certainly isn't true for all editing. Wether you find any benefits in optimizing your media entirely depends on your system. If you have powerful enough processors and a good GPU, your system won't have any trouble decompressing H.264 on the fly and many effects will play smoothly without any need for rendering. Everone needs to test this for his own system and media. If the system plays media and effects without any issues, you don't need to optimize or to work in proxy mode.

However, in certain workflows it is a common procedure to always work with a standard "house" format, which in many cases is ProRes422. In these workflows, assistants ingest the native footage into the Libraries onto the central media server and they optimize all footage to ProRes422 while ingesting. As soon as the media are on the server, the editors or assistant editors only need to import with Leave in Place and they don't care one bit about the original format of the media, nor about where the native media are stored. They just want to edit.

- Ronny

ronny courtens wrote: ...Wether you find any benefits in optimizing your media entirely depends on your system. If you have powerful enough processors and a good GPU, your system won't have any trouble decompressing H.264 on the fly and many effects will play smoothly without any need for rendering. Everone needs to test this for his own system and media. If the system plays media and effects without any issues, you don't need to optimize or to work in proxy mode...

Ronny, there are several aspects to performance of effects which could possibly be impacted by whether the clip is H264 or optimized:

(1) During application of effect, performance of analysis phase (e.g, optical flow, stabilization)
(2) If effect is rendered in timeline, performance of render phase
(3) Editor responsiveness to various commands, e.g, JKL, trim, blade, etc.
(4) Viewer update rate when playing or fast forwarding in the timeline
(5) Export performance

Since using optimized media reduces CPU load of decoding H264, in theory this might free up additional CPU cycles to smoothly play back the effect. However in my testing on a 2015 iMac 27, I don't see any significant difference -- in ANY of the above. I also got similar results on a 2013 iMac 27.

I applied Gaussian blur, aged film, color correction, and 1/2 speed with optical flow. Using optimized media vs H264 resulted in no significant performance difference in ANY of the above -- even in timeline and viewer update responsiveness to JKL commands.

There is no question that transcoding to proxy makes a difference. It is smoother and faster -- whether effects are used or not. However the issue is the alleged necessity of using *optimized* (not proxy) media if using effects to obtain decent performance.

On the hardware, effects and material I've tested, I don't see a significant difference. Certainly not enough to warrant the 8x size increase and time to transcode everything.

I have also done similar tests on Premiere Pro CC on the same hardware. FCPX is *vastly* faster. In some case the viewer update rate is 20x faster on FCPX. The CPU load Premiere imposes for fast forwarding in the timeline is much higher than FCPX. JKL editing is almost unusable on multicam H264 4K due to severe lag. That is with zero effects applied. In fact I don't consider Premiere usable on multicam H264 4K unless the material is transcoded -- and that's on a top-spec 2015 iMac 27. That's a problem considering there's no native support in Premiere for this.

It is possible the "always transcode if using effects" advice is inertia from older versions of FCPX, old hardware, or the current state of Premiere Pro on H264 4K.

Yes I understand some groups have a workflow policy of always transcoding to ProRes or recording in ProRes via an external recorder. That has some advantages but the issue here is the recommendation to always transcode to optimized (not proxy) media to obtain good FCPX performance on effects. My testing does not show any basis for that.

hobbyist talking here, anyhow ...

very interesting observations you're actual sharing, Joe, this one and the '3000' one, thanks for taking the time testing!

It is possible the "always transcode if using effects" advice is inertia from older versions of FCPX, old hardware, or the current state of Premiere Pro on H264 4K.

There's some silly phrasing here "saying always is always wrong" ... :lol:
I guess on the high-end/pro side, proRes is a standard format, I read top-end cameras or recorders use this codec - so proRes is obligatory.

In mixed enviroments, a 'standard' is obligatory too… and if the weakest part in the workchain can not handle 'native' format, prores is probably a good choice...

Speaking from the far lower end of using FCPX, indeed: all sort-of 'live'-effects don't need proRes - even on my [strike]Tupperware-box[/strike] macMini. Different story with my homewbrewn M5 effects containing motion-blur ( I' LOVE motion blur!), that results in stutter - on my underpowered machine. Will test with 'pre rendered prores'.

So, that 'always' advice is probably just good for someways ...

Thanks again for sharing!

I completely agree with your assessment, Joe. Technology is evolving very fast and advice that might have been correct one or two years ago may not apply anymore today. And it's not only a matter of hardware, as you say. Yosemite and El Capitan have provided significant performance boosts for FCP X, also on older hardware. That's one of the advantages of having an application that is tied deep into the operating system. And this advantage will become even more significant with the next updates.

We have seen this only a few weeks ago when doing a big install at Swiss national television. Their entire post-production runs exclusively on FCP X. But their performance was lagging a bit because they were using an old X-SAN and they could not update their workstations beyond Mavericks and 10.1.3. As they work with very heavy and long multicams against tight deadlines, they had made a habit of optimizing all their footage overnight, even creating proxies to work faster.

We installed a new Sharestation Enterprise shared storage system from Lumaforge and their performance was enhanced tremendously. Next we showed them how their media would run if they also updated their 2012 MacPro stations to El Capitan and 10.2.2. And again this made a huge extra difference. During the acceptance tests we were cutting multiple heavy and long multicams in native formats (AVC-Intra 100, ProRes 422 HQ and 4444, XDCAM HD 422@50Mb/s and DVCPRO HD) simultaneously in multiple edit bays while we ingested multiple steams from EVS servers in the studios and vans over FTP and SMB at speeds that exceeded 800 MB/s. Editing was perfectly smooth without any beachballs or latency on unrendered an non-optimized timelines with multicams of 6 and 9 angles playing at Best Quality. They had never seen such a performance before, and the new system and workflows will save them tons of time. Another very happy client (-:

This clearly shows that FCP X is capable of tapping into any resource you throw at it, wether the resources come from improved hardware or a newer OS. And this will be the most significant advantage FCP X has over other NLEs, as technology further evolves.

- Ronny