A story about 4K XAVC-S, Premiere and transcoding

[joema]

On 5/19/2016 at 7:14 AM, s0ny said:

...Surprised that scrubbing is choppy on native XAVCS files, even with your specifications. Will having a GTX980 (or 1070, upcoming) remove the choppiness while scrubbing?....Is there a GPU-accelerated transcoder from XAVCS 4k 100mbps to Cineform 10 bit? I checked my GPU load (660Ti, 2Gb) and it was only at 1%, while my CPU is on full load all the way when transcoding 1h+ of 20 clips...What is the current workflow for swopping out proxy files with the originals? 

What Don said is correct. a GTX980 or 1070 will probably only help scrubbing if the stuttering is caused by effects. H264 decoding is mostly a CPU task and there are only two ways to meaningfully accelerate that (1) Quick Sync, or (2) Proprietary decode hardware such as nVidia's NVENC or AMD's VCE. In either case the software must use those APIs. Currently Premiere CC does not use Quick Sync, although Adobe made some ambiguous statements at NAB about Iris Pro graphics which might imply this is planned for a future version, but they also said Windows only for now. Note that most Xeon CPUs do not have Quick Sync hardware. 

I don't know if Premiere CC uses NVENC or to what extent. There are several versions of NVENC, each with varying capabilities. Except for Quick Sync, NVENC or VCE, I don't think GPU-accelerated XAVCS (which is H264) transcoding is possible. Some utilities may advertise this but the fine print usually says "only for effects". So if your timeline has lots of unrendered effects and you transcode that to an output file, the GPU can accelerate the effects rendering but not the encoding.

Note NVENC/VCE are bundled into GPU cards but they are architecturally not part of the GPU. They are a separate block of ASIC logic (Application-Specific Integrated Circuit): https://en.wikipedia.org/wiki/Nvidia_NVENC

The good news for Premiere users is Adobe knows this is a problem area and they will be adding significant performance improvements to Premiere, including integrated proxy media, Metal API for better effects performance (Mac only), and possibly Quick Sync (Windows only for now). 

Until those improvements are available your best best for smooth H264 4K editing is using a manual proxy workflow. Tony Northrup describes the procedure here:

http://www.rangefinderonline.com/features/how-to/Getting-Acquainted-with-Offline-Video-Editing-to-Ease-You-Into-4K-8988.shtml

Re smooth scrubbing in Premiere CC on 4K XAVCS using a 6700k or 5820, my 2015 iMac has a 4Ghz 6700K, and it is definitely not smooth, nor is my 4Ghz Windows PC, but that has an older i7-875K and GTX-660. I will be testing a GTX-1070 pretty soon and will report any differences. Unless Adobe uses NVENC I really don't see how the GTX-1070/1080 by itself will help sluggish 4K H264 scrubbing since that is an decode problem not an effects problem. However most edited content has some effects so a faster GPU is useful for that.

As bad as these issues are for H264, it will get even worse in the future if H265 or VP9 become more widely used, since those are even more CPU intensive. The Quick Sync on Skylake and later CPUs can accelerate H265/VP9 and I think nVidia's most recent version of NVENC can do this. The issue is software must take advantage of these features.

[s0ny]

Nice. Thanks for the detailed reply. 

 

Off-topic for a little - when you're done with your XAVCS 100mbps files, do you guys usually render them as 100mbps h264 files again or? 

Put it another way - is high bitrate (100mbps) used to facilitate editing only, or does it have a real effect on final output (for archival)? For now, I'm using VBR 2 Pass, 50mbps Avg and 95mbps Max to save space. Can't see much of a difference without zooming in at least.

Need a second opinion here.

[s0ny]

Hi Joema,

 

Have you tested the 1070 yet? How's the result? 

Still trying my luck to find a gpu-accelerated transcoder for XAVCS to Cineform..

[joema]

3 hours ago, s0ny said:

Hi Joema,

 

Have you tested the 1070 yet? How's the result? 

Still trying my luck to find a gpu-accelerated transcoder for XAVCS to Cineform..

I have deferred this since I do almost no editing on my Windows machine anymore. I edit on my 2015 iMac 27 using either Premiere CC or (mostly) FCPX. I would still like to test the GTX-1070 on Premiere CC because the results would be interesting, but I have no practical need for it. 

Premiere CC is quite sluggish when scrubbing H264 4k XAVCS content on both my 4Ghz PC with GTX-660 and also my 2015 top-spec iMac 27. If that is why you are considering transcoding to Cineform, Adobe will be shipping a Premiere update soon which has proxy capability so it can transcode by itself then use the smaller proxies for improved performance while keeping them in sync with the original files. I don't know what codec options they will support but it will be a big improvement.

Adobe has also mentioned vague statements about Intel "Iris graphics" improvements which might be code-speak for Quick Sync. However they say it will be Windows-only for now. That would be another big improvement. It is probably one reason FCPX is so much faster on H264 since it uses Quick Sync and Premiere does not.

If I were doing any significant editing on Premiere and had a GTX-660-series GPU, I would upgrade to the 1070 or 1080 sight unseen, without any tests. Those cards are way faster than the 660, and Premiere really needs that GPU performance -- regardless of whether GPU-assisted transcoding is possible.

A GPU upgrade plus the upcoming Premiere update will probably help you a lot.

[s0ny]

On 6/11/2016 at 8:07 PM, joema said:

I have deferred this since I do almost no editing on my Windows machine anymore. I edit on my 2015 iMac 27 using either Premiere CC or (mostly) FCPX. I would still like to test the GTX-1070 on Premiere CC because the results would be interesting, but I have no practical need for it. 

Premiere CC is quite sluggish when scrubbing H264 4k XAVCS content on both my 4Ghz PC with GTX-660 and also my 2015 top-spec iMac 27. If that is why you are considering transcoding to Cineform, Adobe will be shipping a Premiere update soon which has proxy capability so it can transcode by itself then use the smaller proxies for improved performance while keeping them in sync with the original files. I don't know what codec options they will support but it will be a big improvement.

Adobe has also mentioned vague statements about Intel "Iris graphics" improvements which might be code-speak for Quick Sync. However they say it will be Windows-only for now. That would be another big improvement. It is probably one reason FCPX is so much faster on H264 since it uses Quick Sync and Premiere does not.

If I were doing any significant editing on Premiere and had a GTX-660-series GPU, I would upgrade to the 1070 or 1080 sight unseen, without any tests. Those cards are way faster than the 660, and Premiere really needs that GPU performance -- regardless of whether GPU-assisted transcoding is possible.

A GPU upgrade plus the upcoming Premiere update will probably help you a lot.

 

If they support Quick Sync, then wouldn't the GPU matter less? 

Also curious, will Quick Sync speed up XAVCS h264 to h265 High profile transcoding too? 

[joema]

On 6/11/2016 at 2:49 AM, s0ny said:

If they support Quick Sync, then wouldn't the GPU matter less? 

Also curious, will Quick Sync speed up XAVCS h264 to h265 High profile transcoding too? 

Quick Sync only helps in encode/decode of certain "long GOP" codecs like H264 and H265, and only if the software uses the APIs. It does not eliminate or even lessen the importance of the GPU which is used for effects.

There are different versions of Quick Sync. Only the latest Skylake CPUs support H265. In theory the Skylake version of Quick Sync could greatly accelerate H264 to H265 transcoding since that could be hardware accelerated. So far Adobe has only said the upcoming Premiere update will use "Intel Iris Pro" graphics for H265. That is probably code-speak for Quick Sync.

There is no architectural relationship between a GPU and H264 or H265 encoding, and a GPU cannot normally meaningfully accelerate this. However Intel's Quick Sync circuitry is dependent on their on-chip GPU, so Quick Sync as currently designed cannot be a separate feature. For that the H264/H265 acceleration feature is often mentioned in the context of the on-chip GPU, even though it's not really the GPU doing that work.

Whether Quick Sync is useful for a specific H264 or H265 encoding style depends on the version of Quick Sync. I don't know if the Skylake Quick Sync works for "high profile". The prior versions only worked for single-pass H264, but (as implemented by FCPX) produced very good results which I cannot visually distinguish from multi-pass encoding.

Likewise nVidia's NVENC and AMD's VCE is bundled on their GPU boards but it is architecturally distinct and separate from the GPU. Similar to Quick Sync, if the software uses these hardware-accelerated encode/decode features, those tasks run a lot faster. Unfortunately NVENC and VCE are proprietary to nVidia and AMD. 

[s0ny]

Alright, Premiere Pro CC 2015.3 is released and supports h264 decoding through Iris. 

Has anyone tried it on their Skylake CPUs? Does it enable real-time, full-res 4K playback without lag? That will save me tons of time in transcoding to Cineform.

[tellure]

I'm super interested in the answer to this too.  If Premiere Pro has finally caught up to a 5-year old Macbook w/ FCP in terms of h264 playback I may have to upgrade to a Skylake CPU.

[fuzzynormal]

I'll give it a shot later in the week.

I've been attempting to settle on a good post production work flow as I try to get into Resolve... but if Premiere will let me efficiently edit h.264 4K source files on my i7 4790k / Gtx1080 PC, it's certainly worth looking into. 

Sadly, haven't had much luck yet with Resolve and it's Optimized-Media-work-around.  And it absolutely chokes on source h.264 files. 

Not to mention that Resolve can only display on two monitors and I'm using three.

Hmmm, Adobe is suddenly looking more interesting.

https://blogs.adobe.com/creativecloud/now-live-the-latest-update-of-premiere-pro-cc/

[Curtis Ross]

Ive never had a problem with my i7 and Cuda. 4k playsback smoothly in every instance in Premier CC.

 

 

[tellure]

It plays back smoothly but you can't scrub smoothly or quickly move the play head to a new position and have it start playing back instantly.

[Brian Petersen]

On 11/6/2015 at 2:23 AM, Andrew Reid said:

Thanks Don nice post.

Meanwhile on a Mac I recommend to simply use EditReady, transcode to ProRes 422 LT and you will be able to playback full quality full res 4K on a laptop. There's no quality loss from XAVC-S.

The only way you can lose quality from XAVC-S is to go backwards significantly from H.264 100Mbit/s and that is quite hard to do

Thanks for this thread.  New to Premiere, coming from FCPX.  Is this EditReady transcode process the way most are doing it on a Mac?  And are the ProRes422LT files your new masters when going to finish?  I'm assuming this method is not a proxy method where you would re-link back to original media at the end.

My main concern is final image quality.  Shot a short on A7sii in S-Log3 at high iso and the quality is great, but I would be willing to sacrifice ease of edit and storage space if it meant maintaining any sort or advantage on the image in the final product.  I'm just scared of transcoding to something and treating those files as my new masters only to find out at the end I would have had less macroblocking if I just edited natively.

Thanks for any advice!

[Brian Petersen]

I noticed on another thread the recommendation to switch the Renderer to Mercury - OpenCL (from CUDA). I now can playback native great on my Mac!

My plan now is to edit natively and then export a media managed sequence with clips with handles in ProRes422. (I'm assuming I can do that in Premiere) And use those ProRes clips for FX and color. 

 

Unless someone sees a flaw in my workflow I'm not seeing. 

 

 

[Sackboydad]

Don,

I am also struggling (badly) with PPro (currently the latest version 2015.3) and 4K XAVC-S files from my A7s-II.

I get stuttering during playback on PPro timeline (or even just playing a clip off the timeline), even without effects.  But it's not constant.  It'll play then just start stuttering.  If I stop then start the timeline, it'll smooth out for a bit then start stuttering again.  If I add Lumetri then it will slightly stutter all the time.

I have a GTX970 4GB GPU.  I know that's a decent video card.   My CPU is a little old... it's an Intel I7 3770K.  I have 32GB of RAM.  PPro and all it's mediacache files reside on two different SSDs.  The XAVC files are on a WD Black 6TB drives (but I tried putting all the XAVC stuff on SSD and the stuttering wasn't any better). 

I am to the point where I HATE editing.  It's so frustrating that I have to keep stopping and restarting playback.  Might sound flaky, but it interrupts my creative thinking when I can't watch the whole thing at one time!

I also do not want to transcode my XAVC into another format.  If I open task manager I can see my CPU is pegged at 100% when I playback.

I want to enjoy editing again! :-)   I was thinking of buying a new motherboard with the latest 6950X CPU and getting 64GB of RAM.  I think the video card is still good and I've heard that going from a 970 to a 1080 only yields about a 10% improvement. 

Any help or advice would be sincerely appreciated!

Thanks!

Stacy

[Don Kotlos]

2 hours ago, Sackboydad said:

 If I open task manager I can see my CPU is pegged at 100% when I playback.

Yep that's your problem. 

Quote

I want to enjoy editing again! :-)   I was thinking of buying a new motherboard with the latest 6950X CPU and getting 64GB of RAM.

That is a hell of a CPU for rendering but I doubt you need inot order to just improve playback. Fewer faster cores should yield better results for playback. As you can see in my tests I was never able to utilize all 8 cores and the scrubbing kept being choppy. 

Additionally there are reports that the new premiere supports h264 decoding with IRIS which would theoretically give you the best playback performance. So I would ask around if that is the case and if yes go with the latest i7. 

Quote

  I think the video card is still good and I've heard that going from a 970 to a 1080 only yields about a 10% improvement. 

Yeah 970 is quite powerful and I wouldn't change it unless you have many many effects that you want to render in real time at full resolution. XAVC-S decoding will not be improved with the 1080. 

[Sackboydad]

Don.... thanks for getting back to me!

So, just to make sure I'm understanding correctly....  Playback or scrubbing (or are you saying they're the same thing?) is completely CPU based and the faster the clock speed the better?   My older 3770K runs at 3.5GHz.... even the top-of-the-line 6950X runs only at 3.0GHz.   I really am not a moron (well, don't ask my wife!) but I would assume the new chip would be better.

I'm sure it's totally different, but windows 10 can play the 4K files at full speed and resolution with absolutely no problems at all -- it's only premiere that chokes.... and not just on a timeline... it'll stutter if I'm just playing the clip.

I'm just worried about buying a new CPU/Motherboard and finding I'm still stuck with stuttering on plain and simple 4K playback.  CPU's are generally not returnable.  I don't mind spending $2700 on new gear but I want to be sure it's going to solve my problem before I do.

[fuzzynormal]

3 hours ago, Don Kotlos said:

there are reports that the new premiere supports h264 decoding with IRIS which would theoretically give you the best playback performance. So I would ask around if that is the case and if yes go with the latest i7. 

I'm now getting back into trying Premiere after years of being away.  Just installed Win10 and Adobe apps on my latest PC this week.

Anyway, I have a GTX1080 and an i7 3770, 32GB of RAM.  CUDA is enabled within Premiere's preferences settings.  I get surprisingly acceptable scrubbing at this point with camera native 4K .h264 files.  I haven't really pushed the system, just been doing small edits so far, but rolling forward and backward, while not as "crisp" as working with prores422 or HNxHD, is rather painless so far.

GPU-z is telling me that the 1080 is working at 15-25% load during playback.  And the CPU pretty much maxes out.  When I add a few effects to clips, the GPU load jumps a bit.

I can tell you this, Resolve on my system doesn't handle these exact same files worth a squat when scrubbing.  I haven't kept up with all the hardware/API development, all I know is that Premiere is accessing something that Resolve sure doesn't.  Will it choke when there's a large documentary loaded on the system?  Can't say yet.  I'm sure hoping it doesn't!

[Don Kotlos]

3 hours ago, Sackboydad said:

Don.... thanks for getting back to me!

So, just to make sure I'm understanding correctly....  Playback or scrubbing (or are you saying they're the same thing?) is completely CPU based and the faster the clock speed the better?   My older 3770K runs at 3.5GHz.... even the top-of-the-line 6950X runs only at 3.0GHz.   I really am not a moron (well, don't ask my wife!) but I would assume the new chip would be better.

Probably because of the newer generation but not necessarily. Playback or scrubbing is mostly CPU dependent and not multithreaded well. So if you look at my tests, while not all the cores are used performance still lacks. 

3 hours ago, Sackboydad said:

I'm sure it's totally different, but windows 10 can play the 4K files at full speed and resolution with absolutely no problems at all -- it's only premiere that chokes.... and not just on a timeline... it'll stutter if I'm just playing the clip

I just checked and 3770k supports quicksync that might be used with the native program but not premiere. 

Check this solution here:

https://forums.adobe.com/thread/2177025

It might work with your system. 

3 hours ago, Sackboydad said:

I'm just worried about buying a new CPU/Motherboard and finding I'm still stuck with stuttering on plain and simple 4K playback.  CPU's are generally not returnable.  I don't mind spending $2700 on new gear but I want to be sure it's going to solve my problem before I do.

You can get smooth playback with any modern i7 CPU so don't worry that much. 

But still I believe that 6950x is not your best choice for playback/scrubbing performance. 6700k should be better if quicksync is used. If you want the extra cores for rendering then go with 6900k or 68x0k and overclock.

[fuzzynormal]

Regarding my previous post: Brainfart and the wrong info.  I'm using an i7-6700 CPU on my PC.

[joema]

On 7/16/2016 at 10:02 AM, Sackboydad said:

....struggling (badly) with PPro (currently the latest version 2015.3) and 4K XAVC-S files from my A7s-II.

I get stuttering during playback on PPro timeline (or even just playing a clip off the timeline), even without effects.  But it's not constant.  It'll play then just start stuttering.  If I stop then start the timeline, it'll smooth out for a bit then start stuttering again...I have a GTX970 4GB GPU.  I know that's a decent video card.   My CPU is a little old... it's an Intel I7 3770K.  I have 32GB of RAM.  PPro and all it's mediacache files reside on two different SSDs.  The XAVC files are on a WD Black 6TB drives (but I tried putting all the XAVC stuff on SSD and the stuttering wasn't any better)...I am to the point where I HATE editing....I also do not want to transcode my XAVC into another format.  If I open task manager I can see my CPU is pegged at 100% when I playback.

Editing 4k H264 is often sluggish with virtually any editing software on almost any computer. It is just inherently hard -- it's 4x the data per frame of HD, it's stored in a compressed "long GOP" format which must be decoded on playback. It is generally a CPU-bound task, not I/O or GPU limited.

FCPX is much faster than Premiere at this but even FCPX can struggle with 4k H264 multicam. Even on a high-end machine I would never edit 4k H264 multicam without transcoding to proxy -- using any editor.

On my 2015 top-spec iMac 27 (4Ghz i7-6700K, 32GB, 1TB SSD, M395X, 16TB Thunderbolt RAID5), Premiere CC 2015.3 is borderline usable on a single stream of 4k H264. I have never seen *any* problem with pure 1x playback (at 1/4 res) of 4k H264 on Premiere CC on my iMacs or a several-year-old Windows PC with a 4Ghz i7-875K CPU and GTX-660. The lag happens when scrubbing the timeline or using JKL commands to rapidly change from FF to REW -- not during 1x playback. If your system can't even do 1x playback there may be something wrong, either in the configuration or hardware. Make sure your Source and Program monitors are set to 1/4 resolution.

I could see editing small single-cam 4k projects without transcoding. For 4k H264 multicam, transcoding to proxy is essential whether using FCPX or Premiere.

Fortunately Premiere 2015.3 has added proxy support which greatly speeds up H264 editing. I have tested this on 4k XAVC-S content from my A7RII and similar content from a Panasonic AG-DVX200. The downside is you must transcode, but at least Premiere now supports that internally. It can be done during or after ingest. In limited testing I've done, Premiere is about twice as slow as FCPX at transcoding to proxy, but it gets the job done.

If you want fast, fluid 4k H264 editing on Premiere, this requires either a custom-built machine (or equivalent), or transcoding to proxy. Even using FCPX on a top-spec 2015 iMac, you have to transcode to proxy for fastest, smoothest editing performance on 4k H264.