Testing Project for RX5700 XT / RX5700 / VEGA 64 with FCP X and macOS
 
Notifications
Clear all

Testing Project for RX5700 XT / RX5700 / VEGA 64 with FCP X and macOS  

 of  2
  RSS

mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 3 years ago
 
Posted by: @wimpzilla

My point, and looking at your answer i spotted it right, was about evolving if you know you can have something better elsewhere.
I really believe you when saying that FCP was improved and it is great, but again looking at the OP bench's, one can clearly notice a rendering performance roller coaster.
Sure thing if the GUI and user friendliness are your main focus and FCP propose the best why not, but if rendering time is you wait to go not sure FCP exceed in this field.

You are not sure about FCP performance so I suggest taking a look at some comparisons. There are many interesting results, some of which I mentioned previously. Check out barefeats for eGPU-specific benchmarks (you’ll see eGPU making a big difference and sometimes being worse - interesting results). eGPU load is not that simple.

Not sure what you looked at in my answer which clearly says:

Posted by: @mac_editor

I definitely keep up to speed with what companies are doing on other platforms. I love the PC community as well - I am thinking of building one for gaming too. I make a conscious decision to stick to FCP (because its better).

The following is only guesswork, is simply not true with respect to eGPU, and also does not explain why FCP would perform better:

Posted by: @wimpzilla

Also as far as i know, on window platform, software editors are less kind to throttle eGPU performances and often provide more stable native eGPU acceleration as flexibility feature.

eGPU compute is a complicated topic because we have no idea how (in this case) Thunderbolt works. A lot of pros on this forum focus on macOS issues/are macOS users, so you are bound to hear about macOS more - which does not imply Windows side does not have issues. I just look at all this at face value.

Posted by: @wimpzilla

You are right i'm absolutely no expert in video editing, i can't give an advised opinion about which software is better and for what.

Yet you make the following suggestion:

Posted by: @wimpzilla

That's why i will still advice you to go check elsewhere to be sure that the tool you use are the best of the current field and evolve is needed.
If FCP is the best for you now, as you said it's fine, but for sure the eGPU rendering performances are maybe not you main focus point as it can be for someone else.

I know the advantages/disadvantages of the tools I use extremely well (I am as curious as a cat haha). I was just sharing my insight on the matter of overall performance of FCP vs. elsewhere. This thread does not have any Premiere Pro or any other editor’s benchmarks, yet by looking at Final Cut Pro performance variance you have drawn a conclusion that FCP is throttled (blatantly incorrect terminology) and that elsewhere there is better performance (which you are guessing).

Finally, this sums up our difference in opinion as well:

Posted by: @wimpzilla

At last, if windows broke overall eGPU support as it did, not sure that software editors are to blame anyway, to be honest i have no clue why M$crosoft broke eGPU support like that.

Ya the video editing software isn’t to blame, but it’s inherently unusable over eGPU if the platform (which is as important as the software itself) fails to support it. Video editors using AMD eGPUs are adversely affected. Of course, on the forum we have workarounds (disable updates) and have learnt from @itsage that the XConnect team is aware of the issue - which is awesome. I can edit on macOS with a piece of mind - again, not saying that macOS is perfect - I’d be the first to highlight problems and issues - which there are as this benchmark shows, and you and I both agree that this variance is weird. That’s why I said this above:

Posted by: @mac_editor

Over the last few releases of Final Cut Pro, export times over eGPU have declined. Final Cut Pro 10.3 was faster for export over eGPU vs internal GPU. What FCP needs to do is consider multi-GPU systems incorporating eGPUs - currently they support multi-GPU scenarios where all GPUs are internal (2019 Mac Pro ).

Besides making a performance argument, I also wanted to convey the importance of the “entire package” when it comes to recommending something “better” - from a video editing standpoint. I hope you found our chat interesting, because I certainly (always) appreciate your contributions/guidance but hope that it can be more informed in the future.

One more example:

See 7:24 for chart where he adds Final Cut Pro to the chart. On the same hardware, Final Cut is more than 10 times faster vs. Premiere for 1min 4K HEVC.

And another example - this shows how other video editing software have improved as well plus cost-effectiveness of PC choice (changed now with 16-inch MBP):

purge-wranglertbt-flashpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance

Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro

 
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]  


ReplyQuote
wimpzilla
(@wimpzilla)
Honorable Member
Joined: 3 years ago
 

@mac_editor

But in any shape or form you really take seriously the bench's posted by OP in your explanation.
Neither explain why the tool have such a rendering variance, tho being anyway better than the concurrence.
That why i felt your first answer was a bit off, rather than finding this bench behavior strange and interesting you just acknowledged it become better.

Also please do not call blatant thing that actually happen, i think you are taking this too personally, that why i said i spotted your answer right on!
Apple have not the best consumer practices, it is a fact, that reflect in each product that apple deliver, so please keep your "blatantly" for yourself!
I'm just trying to share some awareness about what OP numbers shown, but it seems you are clearly defending a product as it is yours.
This is the kind of awareness i would like to make you think about, from the beginning of this discussion, but i suppose i failed!

I have really no interest in mac stuff but still try to get you think about the product most of the forum users are working on.
So again, i believe you when saying that FCP is a good tool and was made better, but maybe rendering time performance is not your main focus as it is for some others!
So be kind enough to explain me if these rendering difference posted by the OP are normal and fit with what you said about FCP.
Because you pointed out multi-gpu support, but on apple side i don't remember having the possibility to get Crossfire/SLI working, so TB3 would be what we are speaking about.
Obviously taking aside the desktops MacPro line that got maybe full multi-gpu support for FCP??

 

This post was modified 9 months ago

2012 13-inch Dell Latitude E6320 + R9 [email protected] (EXP GDC 8.4) + Win10
E=Mc²

 
2012 15" Lenovo Thinkpad T530 [2nd,4C,Q] + R9 270X @ 4Gbps-mPCIe2 (EXP GDC 8.4) + Win10 [build link]  


ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 3 years ago
 
Posted by: @wimpzilla

But in any shape or form you really take seriously the bench's posted by OP in your explanation.
Neither explain why the tool have such a rendering variance, tho being anyway better than the concurrence.

Which is why I called the variance in performance “weird” in my post above, in agreement. You also used the term throttling for GPU which is not an accurate explanation of the issue. As mentioned in my explanation, performance over thunderbolt has a variety of factors (which you and I don’t know). You are making claims rather than guessing, and not considering more data points.

Posted by: @wimpzilla

I have really no interest in mac stuff but still try to get you think about the product most of the forum users are working on.

So why recommend there’s something better elsewhere with unfounded basis (I’ve explained what better implies already). If you think these benchmarks are sufficient evidence alone (while ignoring other benchmarks conveniently) then let’s stop here.

Posted by: @wimpzilla

Because you pointed out multi-gpu support, but on apple side i don't remember having the possibility to get Crossfire/SLI working, so TB3 would be what we are speaking about.
Obviously taking aside the desktops MacPro line that got maybe full multi-gpu support for FCP??

They introduced multi-GPU support for FCP for Mac Pro. You need neither Crossfire nor SLI for multi GPU compute. I responded with (and clearly):

Posted by: @mac_editor

What FCP needs to do is consider multi-GPU systems incorporating eGPUs - currently they support multi-GPU scenarios where all GPUs are internal (2019 Mac Pro ).

Meaning they should consider multi-eGPU scenarios, because it is not a thing yet.

Posted by: @wimpzilla

So again, i believe you when saying that FCP is a good tool and was made better, but maybe rendering time performance is not your main focus as it is for some others!

Why would I even be arguing in favor of FCP performance if I didn’t care about it? I’ve already explained FCP performance. I’ve already said the variance is weird. Again, see other benchmarks with how the new update has improved performance. I respond to you because I care about your feedback/suggestions and hope to address your points, but at the same time, I wish to ensure accuracy of information on this forum. This thread doesn’t boil down to: FCP eGPU performance is varied so Apple does not have best consumer practices - that’s just funny. If only everything was that simple to explain.

Posted by: @wimpzilla

Apple have not the best consumer practices, it is a fact, that reflect in each product that apple deliver, so please keep your "blatantly" for yourself!

I never claimed they did - I dislike Apple for their current repair situation, where consumables such as batteries in macs are expensive to replace (besides the other reasons I have for criticizing them). I just look at FCP performance (and other things) more objectively instead - you seem to not as this comment indicates so there is no merit in discussing further.

purge-wranglertbt-flashpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance

Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro

 
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]  


ReplyQuote
wimpzilla
(@wimpzilla)
Honorable Member
Joined: 3 years ago
 

@mac_editor

I agree with you, if you repute that there is no merit to argue further, no problem i will not answer back since if you don't care, i care even less!

But we should have stopped to be critics and stopped to argue also when major eGPU industry milestones have been reach through this forum.
And as you obviously know, my lack of objectivity made my discussion useless all theses years being on this forum!
Like half baked bandwidth, half baked eGPU acceleration, half baked eGPU hardware, without citing how much other eGPU stuff evolved.
Because you know better than me that the industry is solely dedicated to its users, features missing are only a little miss nothing else, as i already heard  here somewhere!

Concluding, i will take out your own words and experience with FCPs, OP result are only a weird behavior that do not need any further check or investigation!
FCP have been updated and it is better now, even if the eGPU acceleration is not complete and used only for its timeline benefits!

 

This post was modified 9 months ago

2012 13-inch Dell Latitude E6320 + R9 [email protected] (EXP GDC 8.4) + Win10
E=Mc²

 
2012 15" Lenovo Thinkpad T530 [2nd,4C,Q] + R9 270X @ 4Gbps-mPCIe2 (EXP GDC 8.4) + Win10 [build link]  


ReplyQuote
Massimo Franzese
(@massimo_franzese)
Trusted Member
Joined: 8 months ago
 

I have been reading this thread with interest. As a heavy user of final cut pro I do not spend time with a stopwatch just to have fun however I have to say that this test does not actually take into account the real use of the software.

Let's start with deleting rendering files. Why would you do that? I do not do that unless you have finished with a project and you are archiving it.

Final Cut Pro renders in ProRes 422 by default and of course will only render if you are applying an effect or doing some changes. If you do real video editing you always have some change so your clip is always fully rendered not just in transitions but also in playback because you do grading.

Now when it comes to sharing Final Cut Pro options are totally shite excuse my french and compressor is even worse in fact I managed to get a refund from apple. Anyway let's assume you are into masochism and want to use Apple awful and buggy codecs because you like to watch all sorts of artefacts in your clip

Starting point is project fully rendered or almost fully rendered. Here an eGPU is useful as it keeps rendering while you are busy with live effects. 

Now with your starting point ProRes 422 you can do one of three things

1. Export with the same ProRes 422

2. Go up and export for example in ProRes 422 HQ -- This is what I do

3. Go downwards and export in h264 and hevc because as I said before you like to watch some serious garbage compression

In scenario 1 the sharing will just copy the render files there will be no encoding and no need for any eGPU

In scenario 2 the sharing will have to re-render everything to 422 HQ this will be done by the eGPU and you will see a utilisation spike at the beginning then eGPU will plateau and will be CPU for most

In scenario 3 the compression will start from render files therefore GPU is almost not used and FCPX will crack on with CPU work

The largest majority of Apple Users will go for scenario 1 and 3 and someone who is OCD will go with scenario 2 maybe because they acquired in ProRes 422 HQ 

Nobody in real life will delete all render files and then use compressor to encode HEVC this is just not the way it works and it is unlikely that apple looks at a similar scenario when they optimise their product.

In terms of internal GPU vs external as many tasks are data transfers and it is more expensive to use an EGPU because thunderbolt is much slower the PCI bus it is normal that Final Cut will try and use a discrete GPU if all is doing is data transfers which is what happens at encoding (but not at rendering) there are some detailed articles on apple developer network of which I am a member by the way

I spend majority of my time looking at my footage over and over again cutting adjusting or grading that I have to wait even the whole night for encoding I do not really care that much. In practical terms however I use final cut pro to export prores only because there are issues with the other encoders and then use handbrake that has full control of settings. Handbrake uses zero gpu is only computation and therefore I am very happy with my mac mini 2018

I think those tests are interesting but don't reflect how a professional user uses the application and therefore are of academic but not real interest.

Mac Mini 2018 3.2 Ghz 6 cores
Razer Core X enclosure with Sapphire Vega 64 Nitro
Benq PD2720U


ReplyQuote
Massimo Franzese
(@massimo_franzese)
Trusted Member
Joined: 8 months ago
 

I would also add that last year I needed a new unit for 4K work and the new iMac was not there. I also work on the road with a 13" form factor so for me the choice mac mini top specs + macbook pro 13 + eGPU is great and makes much sense

If you are a desktop guy that does not do work on the road (I use proxies and use the laptop for photography for most) as of december 2019 your best bet is an iMac that does not need any eGPU a good PCI card will trash the thunderbolt interface miles away even your mediocre 2014 did that data transfers are more important than processing for many tasks.

Likewise if you have a macbook pro 16" there is a bit more value in a eGPU as you can run it with the lid closed a bit cooler but somewhat limited performance wise if you have a top spec. The units that get the most value of eGPU are those without a discrete GPU like the mini and 13 mbp which is my use case.

Your testing confirms something that was already known and well documented here https://developer.apple.com/documentation/metal/gpu_selection_in_macos/understanding_gpu_bandwidth

The resource storage modes that pass data to the eGPU are the least efficient so it is possible that as Apple is rewriting new version of FCPs those drive more process over a discrete gpu but the situation of an integrated gpu and external monitor connected to the eGPU is still preserved not in virtue of eGPU acceleration but because that GPU is driving the display. Then if you have a beast GPU it will have enough horse power but still not be as fast of a mac with a discrete card and PCI bus that will always be faster

You also have to take into account that encoding is a CPU job not a GPU as there is no render you are just decoding flushing data and compressing 

Mac Mini 2018 3.2 Ghz 6 cores
Razer Core X enclosure with Sapphire Vega 64 Nitro
Benq PD2720U


ReplyQuote
 of  2