NO WAY!? iMac 5K Late 2015 weaker FCPX performance than MacBook Pro 15" 2013?

[AppleGuy1993]

Hi,

So I've been testing a really heavy duty 4K project - lots of 2D and 3D titles, gaussian blur - on a Mac Pro 12 core, D700, iMac 5K late 2015 395X max out + flash and my old 15" MBP 750M.

Mac Pro - 17+ effects in real time
iMac 5K - 4-5 effects lots of dropped frames
MacBook Pro 15" Late 2013 - 8-10 effects in real time

I had a look at the CPU and GPU usage and:
iMac 5K - 25% CPU usage & 5% GPU usage with an average of 5FPS when playing the project (used iStats to measure this)
MacBook Pro 15" Late 2013 - 155% CPU usage (so one core and 1/2) and 80% GPU usage @49-54FPS.

Not sure why but FCP seems to not fully utilize the CPU and GPU of the 5K iMac.

Tried deleting all preferences, made a new account - same results.

What do you guys think? Either FCP hasn't been updated to fully utilize the 5K iMac's hardware - really strange since it's been out since November or...idk? Everything else uses my GPU normally and my CPU so really have no idea here

Any thoughts?

 

[accentaudio]

I have one on the way. I will test it when it gets here. I too find that hard to believe.

 

[joema2]

So I've been testing a really heavy duty 4K project - lots of 2D and 3D titles, gaussian blur - on a Mac Pro 12 core, D700, iMac 5K late 2015 395X max out + flash and my old 15" MBP 750M.

Mac Pro - 17+ effects in real time
iMac 5K - 4-5 effects lots of dropped frames
MacBook Pro 15" Late 2013 - 8-10 effects in real time...

I have the same iMac as you and have run many tests comparing it to my 2013 iMac 27 and my 2015 MacBook Pro. In general the 2015 iMac is faster, however there were a couple of FCPX tests where the 2013 was faster. That is the way software is -- a complex multi-dimensional surface which harnesses the lower driver and hardware with varying efficiency: https://joema.smugmug.com/Computers/IMac27BenchmarksVsOthers/n-hR5wtF/i-3mL6T5r/A

There is no reason I can think why your iMac 5K would be *consistently* that slow -- provided the test conditions were the same. There are some effects such as Neat Video which can use GPU acceleration on some GPUs and not others.

However my main suspicion would be FCPX on your iMac is configured differently. This could include one or more of these:

- Viewer is set to Better Quality on the iMac but Better Performance on the others
- You are using proxy/optimized media and did not switch the iMac to proxy in the viewer after transcoding
- Your project characteristics are not the same, e.g, 4K on the iMac and 1080p on the others. You can use 4K media in both project types. This can easily happen because the default FCPX behavior is auto-select project resolution based on first added clip. If you have some 1080p content anywhere in your project, and one 1080p clip was by chance added to the MP and MBP but not the iMac, then they could be using different resolutions, which impacts performance. Without clicking on the project and using Inspector to examine what's in use, you can't be sure.
- You have background rendering disabled on the iMac and enabled on the others. After the timeline auto-renders it will be more responsive
- You are not running the same version of FCPX on all machines. The newer versions are more optimized.
- You are using 3rd party effects and have an older, less efficient version of some effect on the iMac but not the others.

 

[AppleGuy1993]

Thanks for the feedback!

I do have the exact same settings on both - Performance Playback on both, both playing from the internal Flash Storage (700 MB/s on the MBP and 1.8GB/s on the iMac)

I did do some render tests, including GPU heavy ones, and the iMac just tops the MacBook Pro by miles, so exporting doesn't seem to be an issue - the GPU is fully utilized there. Playback does however.

 

[joema2]

Are you using optimized or proxy media? What camera and codec? Can you do a playback test with just 4k material and no effects? I edit lots of 4k H264 material and don't generally have a problem. On multicam 4k I usually transcode to proxy for better performance.

If you already know the settings are the same then the next step is determine whether it happens without effects. Unfortunately there is no FCPX command to remove all effects, but you could make a test project with a single clip and try that on all three machines. If that still exhibits the difference I can test on my iMac and MBP.

 

[AppleGuy1993]

Are you using optimized or proxy media? What camera and codec? Can you do a playback test with just 4k material and no effects? I edit lots of 4k H264 material and don't generally have a problem. On multicam 4k I usually transcode to proxy for better performance.

If you already know the settings are the same then the next step is determine whether it happens without effects. Unfortunately there is no FCPX command to remove all effects, but you could make a test project with a single clip and try that on all three machines. If that still exhibits the difference I can test on my iMac and MBP.

I'm using the Original footage from my GH4. H.264 .mov.
Editing with no effects does indeed utilise the GPU, it's only when I add a lot of effects that the GPU usage completely drops. Optimizing the media solves all the issue - but then again, I'm using original on my MBP and it's buttery smooth.

Sure! That would help a lot if you could try it out on your machines! I've included the XML + the video file here. Also, if you have iStats you can monitor the CPU and GPU usage live - mine never goes above 5% (iMac 5K).

https://drive.google.com/open?id=0B5gDE8QfzQkIRDZFQVZvcVlZRWc

 

[joema2]

OK I downloaded it and will examine it tomorrow.

 

[AppleGuy1993]

OK I downloaded it and will examine it tomorrow.

Thanks! Would be interesting to see how it performs on a Late 2015 15" rMBP with an AMD GPU since mine has the nvidia 750M.

Just as an update - I've installed the OS on a new partition - tried it with different projects - same results - GPU barely used. 4-5FPS vs 40 on the 750M

On the other hand I also did some heavy FCP X 3D project renders which exported much quicker and did use the GPU on the 5K iMac. So exporting seems fine, just the playback seems affected.

 

[joema2]

Thanks! Would be interesting to see how it performs on a Late 2015 15" rMBP with an AMD GPU since mine has the nvidia 750M.

Just as an update - I've installed the OS on a new partition - tried it with different projects - same results - GPU barely used. 4-5FPS vs 40 on the 750M....

OK I imported and ran your material inc'l XML on both my 2015 iMac 27 and 2015 MBP. I imported the content with "leave files in place" and did not use either proxy or optimized media.

It ran fine on both iMac and MBP with no dropped frames at 1x playback or fast forward. It made no difference whether viewer was set to best quality or best performance. As is typical with 4k H264 the playhead was a little jerky but the viewer frame rate was very good.

I then rendered the timeline (CTRL-A, CTRL-R) and the orange bar went away and playback was a little smoother but it was OK before. Usually I have background rendering turned off. On your content the only effects I see are the titles on the 2nd clip.

How are you measuring the 4-5 FPS? Is that a visual approximation or what? I'll be happy to do any other testing you want.

System configs:

2015 iMac 27, 4 Ghz i7, 32GB RAM, 1TB SSD, M395X GPU
OS X 10.11.4, FCPX 10.2.3

2015 MacBook Pro 15", 2.8Ghz i7, 16GB RAM, 1TB SSD, M370X GPU
OS X 10.11.4, FCPX 10.2.3

 

[AppleGuy1993]

Hi joema,

Thanks a lot for testing this! Are you sure auto transcode was turned off? If you click on the video file - inspector - info - is "original" the only one green?

That's how it is for me. I've included a screenshot of how it should look in the viewer - 4K file + 2 4K picture in picture clips + 4 2D titles and one 3D title. It's completely unplayable on my iMac 5K, even on "Performance". Used iStats to measure the CPU and GPU usage - on my iMac the GPU usage never goes above 5%. On my MBP it's almost 100% reason why it plays much smoother - 40-50FPS from 5 which is what I'm getting on the iMac. If you could try iStats (it's free for 14 days) that would be perfect! Under CPU usage it also tells you the GPU usage live - so just however over that, while the project is playing. Your GPU usage should be almost max on both machines.

 

[joema2]

OK I see the behavior, incl 5 FPS and dropped frames. For some reason when I opened your XML, some of the connected title clips were disabled (toggles with V key). I enabled them and now see the same thing as you, inc'l about 5% GPU utilization. On my MBP basically the same thing but it doesn't get quite as slow -- about 10 FPS and GPU is at least 60%.

Unfortunately I have redeployed my 2013 iMac 27 with GTX-780m so I can't test on that.

It might be some kind of OpenGL bug or a deficiency in how the titles are handled by the AMD GPU. As you can see from the benchmarks I previously posted, in general the M395X is faster than the 780m and for some things it's way faster. It was 2.3x faster on the BruceX benchmark and 2.2x faster on Photon Pro noise reduction, which are both heavy GPU tests in FCPX. But there is no question it's slow on this.

If you transcode to proxy it is very fast for your scenario -- I tested that. If I am working with H264 4k (esp. multicam or many effects) I usually transcode to proxy anyway. I would suggest that as a workaround.

Apple apparently did some interesting FCPX demonstrations at NAB but they are under NDA. They have not publically announced anything about FCPX improvements but Adobe has done several demonstrations showing vast After Effects performance improvement on OS X by using Apple's Metal API. They haven't shipped that yet but probably will soon:

There is no way Apple will not use their own API to get similar performance improvements in effects-heavy tasks in FCPX. This behavior could conceivably be resolved within weeks if Apple ships an FCPX update with those. But I am only speculating.

If your iMac is new and within the return period and you can wait until 4Q, you have the option to return it. It seems likely the 2016 iMac will have a substantially upgraded GPU. If you cannot do that I would live with it and transcode to proxy which will help everything else you do in 4k editing as well. Overall I am happy with my 2015 iMac 27 but if they use Metal to improve graphics performance that would be nice.

Edit/add: remember when you transcode to proxy, you must set the viewer to proxy. Then when you are ready to export to a file, you must set the *viewer* back to optimized/original, else the output file will only be in proxy resolution.

 

[AppleGuy1993]

Hi joema,

Thanks a lot for testing this! Are you sure auto transcode was turned off? If you click on the video file - inspector - info - is "original" the only one green?

That's how it is for me. I've included a screenshot of how it should look in the viewer - 4K file + 2 4K picture in picture clips + 4 2D titles and one 3D title. It's completely unplayable on my iMac 5K, even on "Performance". Used iStats to measure the CPU and GPU usage - on my iMac the GPU usage never goes above 5%. On my MBP it's almost 100% reason why it plays much smoother - 40-50FPS from 5 which is what I'm getting on the iMac. If you could try iStats (it's free for 14 days) that would be perfect! Under CPU usage it also tells you the GPU usage live - so just however over that, while the project is playing. Your GPU usage should be almost max on both machines.

Thanks a lot for testing this in-depth and for the detailed notes!

Yes - I agree, Metal + FCP will make a huge difference. Adobe reported 8X improvements if I remember correctly. However, AMD working on their new 14nm Polaris GPU's which the new iMacs will most likely use - so Metal is no longer needed if we get those high end gpu-s or those high end gpus are no longer needed if we get metal - hopefully we get both

Taking all these tests into consideration - this is definitely a bug with Final Cut Pro X. I've spoken to Apple and submitted a developer bug report as well to let them know of this issue. Hopefully a fix will be released soon. In the meantime they advised me to get as many people to submit a FCP feedback report - apple.com/feedback so that it gets prioritised and fixed soon.

So if everything following this post could submit a bug report via apple.com/feedback - that would help prioritise this issue and hopefully Apple will fix it soon!

And yes - in the meantime - I'll probably optimize all the complex clips I'm working with! Thanks again for testing this joema2!

 

[Koatsey]

I'm using the Original footage from my GH4. H.264 .mov.
Editing with no effects does indeed utilise the GPU, it's only when I add a lot of effects that the GPU usage completely drops. Optimizing the media solves all the issue - but then again, I'm using original on my MBP and it's buttery smooth.

Sure! That would help a lot if you could try it out on your machines! I've included the XML + the video file here. Also, if you have iStats you can monitor the CPU and GPU usage live - mine never goes above 5% (iMac 5K).

https://drive.google.com/open?id=0B5gDE8QfzQkIRDZFQVZvcVlZRWc

[doublepost=1463547236][/doublepost]Just a thought but do you have a codec on you MacBook Pro that you don't have on the iMac? QT Pro for example. Also try another bit of software that is intensive and see how it compares. If the iMac loses and you have done this with other programs then you might be having hardware issues/failures. Stabbing in the dark but just some late night thoughts.