Testing Project2: Is External GPU faster than Internal GPU while exporting video...
 

Testing Project2: Is External GPU faster than Internal GPU while exporting video in FCP X?  

  RSS

iPhone4TW
(@iphone4tw)
New Member
Joined: 2 months ago
 

Hi, Guys,

Last time I sharing this article with you about" Testing Project for RX5700 XT / RX5700 / VEGA 64 with FCP X and macOS "

I got 4 AMD graphic cards for eGPU testing.

The result shows eGPU does not have much benefit with exporting video in Final Cut Pro X

It cost more than 1.5 hour or 2 hours to finish the job.

When I using eGPU , I need to use macOS 10.15.1 and FCP X 10.4.7 to get best result.

But it still take more than half hour than 2014 iMac27 build-in graphic card AMD R9 M295X, with macOS 10.14.6 and FCP X 10.4.6.

I am so sad.

At last, I borrow a 2018 Macmini to make sure, it's not my fault to use TB2 for eGPU. ( I am using Purge-Wrangler )

It need to work with macOS 10.15.1 and FCPX 10.4.6 to get the best result.

I am so confused.

@mac_editor said that :

"Final Cut Pro has excellent optimization for internal GPUs. eGPU is very useful to improve timeline performance. "

Maybe this is why build-in graphic card works much faster.

OK!

The bottleneck is eGPU, right ?

Let me build up a hackintosh for the test.

Let's mount each graphic card internally to compare the result.

INTO THE DARK SIDE

I am using ASRock Phantom Z390 Gaming with Intel core i5-8500 for testing.

MAQ helps me to get into the dark side.

I got AMD Radeon VIIRX5700 XT、VEGA 64RX570 for this test.

We will find out which one is the bottleneck for exporting video with FCP X.

MAQ vs iMac27 2014

Before start, we need to know how much difference between MAQ (hackintosh) and my old iMac27 2014.

With Geekbench4, cpu testing result is as above.

MAQ is a little faster, but it only have 16GB RAM

However,

iMac27 2014 won the OpenCL testing.

iMac27 2014 won the Metal testing, too.

Wait!

Did we just say that internal is much faster than external ?

I am so confused.

AMD Radeon RX VEGA 64

Let's start from AMD Radeon RX VEGA 64 8GB .

Beside of macOS version and FCP X version.

You guys can also watch CPU usage and GPU usage.

GPU usage in FCP X10.4.6 is around 50%.

GPU usage in FCP X10.4.7 is much lower than 50%.

AMD Radeon RX5700 XT

AMD Radeon RX5700 XT needs macOS 10.15.1 or above.

GPU usage is a little higher than 50%.

GPU usage in FCP X10.4.7 is lower than 50%.

How about the result?

In this result, all test is finish around 1hour 30 mins to 1 hour 40 mins.

The best result is VEGA 64 with macOS 10.14.6 and FCP X 10.4.6.

No matter with graphic card is, FCP X 10.4.6 get the best result.

But...

It's still a big gap between iMac27 2014 build-in graphic card, it takes only 1 hour and 4mins to finish the job.

And almost the same result while I am using eGPU with iMac27 2014.

Damn, I am so confused.

Mac Pro 2019

As you know, in these few days.

The news about Mac Pro 2019 are everywhere on social media.

If I have that money, I want to have one.

My friend said to me : Hey, Jimmy, Let's do some trick with the model name, ok?

Then we modified the string about Model name from Mac Pro 2013 (Macpro6,1) to Mac Pro 2019 (Macpro7,1).

"Hello! Are you serious ? " I said.

It's the same hardware, and work with exactly the same macOS .

Macpro7,1

What The F......!!!!!!!!!!

No matter with FCP X 10.4.6 or 10.4.7, it only takes 1 hour to finish the job.

It's 30mins faster than before.

And it's exactly the same hardware, the same operation system.

We just modified the model name from MacPro6,1 to MacPro7,1.

How about the RX5700 XT,it works, too.

WTF!

Last time we have two variables, one is macOS version, the other one is FCP X version.

Today, we got a new variable : Model Name !!!

People said that AMD Radeon VII is the best of best, let's test the myth.

It takes 56mins and 41 sec with FCP X 10.4.6.

The New Result is ...

Let's review the result about MacPro6,1

Basically, it need 1 hour and 30 mins to finish the job.

After that we modified the Model Name without change any hardware equipment or setting.

Every test is reducing 30 mins to finish the job.

And no matter how faster the graphic card is, it's almost the same result with RX570.

RX570 is only slower 6mins to finish the job, but it's the cheapest graphic card in this test.

OK,

The bottleneck is not how we mount the graphic card, no matter internal or external.

No matter how fancy that graphic card is, it take almost the same time to finish the job.

OK, I got it.

The bottleneck is CPU, right ?

Intel Core i9-9900KS

Let's get the Intel Core i9-9900KS

When I grow up, I wish to have one of this.

What!!!!

I am so angry about CPU core 1 to core 16, they do almost nothing in this test.

But, what the hell about GPU usage ?

GPU almost get the full usage.

WE JUST CHANGE THE MODEL NAME !

Than it reduce 30 mins to export.

Now, we use core i9-9900, GPU take all the responsibility, and CPU doing nothing.

Final result is...

It reduces 15mins !!!

i9-9900KS + RX5700 XT + MacPro7,1 + macOS 10.15.1 + FCP X 10.4.6 is the best combination for this test.

But, Intel doesn't sponsor me for this test, so I only test one result with i9-9900.

Therefore I don't know will i9-9900KS do the same magic with RX570.

macOS 10.15.2 and FCP X 10.4.8

At the meantime,

macOS 10.15.2 is release to upgrade

FCP X 10.4.8 is release to upgrade, too.

Will it be better than before?

The result is 10.4.8 worse than 10.4.7, it fixed few bugs from 10.4.7, but no performance fix in this case.

What do you think about this?

Here is the result for the whole testing.

From the very beginning, we just though that eGPU not get the support for FCP X exporting video.

Then we mount the graphic card directly internal with hackintosh.

Then we just modified the model name with exactly the same hardware.

Even using the intel Core i9-9900KS

So, what is the bottleneck while FCP X exporting video ?

We just modified the model name with exactly the same hardware, it reduce 30mins in every test.

30mins is 1/3 exporting time!

Maybe

The real Mac Pro 2019 is the best solution for this test.

PS.

All test is base on the project of this video.

And using the same external USB3 SSD for reading and writing.

I have using much faster raid disk to test the project, it almost take the same time.

I am so confused.

Is there have some kind of hardware combination to unlock all the hidden power?

In my test, "Model Name" is one of this key to unlock the hidden power, what else?

This topic was modified 1 month ago

purge-wrangler ✧ 2014 iMac27 5K + VEGA 64


Username, artik and itsage liked
ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 3 years ago
 

@iphone4tw

Thank you for these tests. Extremely insightful. Simulating the right Mac model is important in a Hackintosh. Other users on other forums have discovered similar tricks to enable HEVC hardware acceleration on GPUs in macOS, for example. I’m also doing some testing, only on native Mac system though, in an attempt to examine the ongoing issues more closely. For now I can confirm that both the Vega 64 and 5700 XT aren’t used to their maximum on my system as well. I shall post those results once I’m done testing. I am also interested to see if the Pro AMD GPU models (not RX) provide any performance advantage in these scenarios over the standard models.

This post was modified 1 month ago

purge-wranglerpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance
2018 MacBook Pro 15" RP560X + RX 5700 XT (Mantiz Venus)

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


iPhone4TW liked
ReplyQuote
artik
(@artik)
Eminent Member
Joined: 2 years ago
 

@iphone4tw

Many thanks for your work... Very interesting indeed... I'm keen to build the hackintosh probably with Ryzen 3900x because I'm tired with the eGPU usage on FCPX (Mac Pro 2013).

Mac Pro 2013, 10 core, Sonnet eGFX 550, Vega 64


iPhone4TW liked
ReplyQuote
MorganDe
(@morgande)
New Member
Joined: 2 months ago
 

Hi! 

I've read your posts with great interest. I have the same 2014 5k iMac and do editing with it. I had some questions for you.

I shoot a lot of HLG footage with my GH5. The HLG footage 10bit, 420, 72mb/s h.265 play back just fine. The 10bit 422, 150mb/s h264, and v-log 10bit 422 footage stutter. And of course the wonderful display doesn't show the HDR very well. I've also been shooting some 5K footage in the 'anamorphic' or 'open-gate' 4:3. It stutters a lot and needs rendering.

so

1) Does your eGPU output good HDR to your LG HDR monitor? I kind of need this.

2) Does render time reduce, can you playback without stuttering?

 

3) Does HEVC (h.265) encoding/decoding work off the eGPU?

 

4) Would you recommend getting a eGPU for editing with the late 2014 5k iMac? I see your exports don’t benefit much, but I really need HDR monitoring and rendering/playback help.

 

Thanks Very Much

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


iPhone4TW liked
ReplyQuote
(@massimo_franzese)
Trusted Member
Joined: 1 month ago
 
Posted by: @morgande

Hi! 

I've read your posts with great interest. I have the same 2014 5k iMac and do editing with it. I had some questions for you.

I shoot a lot of HLG footage with my GH5. The HLG footage 10bit, 420, 72mb/s h.265 play back just fine. The 10bit 422, 150mb/s h264, and v-log 10bit 422 footage stutter. And of course the wonderful display doesn't show the HDR very well. I've also been shooting some 5K footage in the 'anamorphic' or 'open-gate' 4:3. It stutters a lot and needs rendering.

so

1) Does your eGPU output good HDR to your LG HDR monitor? I kind of need this.

2) Does render time reduce, can you playback without stuttering?

 

3) Does HEVC (h.265) encoding/decoding work off the eGPU?

 

4) Would you recommend getting a eGPU for editing with the late 2014 5k iMac? I see your exports don’t benefit much, but I really need HDR monitoring and rendering/playback help.

 

Thanks Very Much

LG monitors do not support HDR HLG but HDR-10 (PQ in Final Cut Pro terms)

Monitors support HDR-10 through DisplayHDR VESA specifications that is aimed at gamers and is not the same as your GH5 footage.

In order to potentially use an HDR-10 monitor you would need to convert fro HLG to PQ in FCPX once you do that in my experience connecting an HDR-10 monitor to an eGPU you don't get HDR-10 signal anyway

However in FCPX HLG has the same scopes of SDR and therefore all you need to have decent colours is a P3 display that I guess is not your case with your iMac

I still have to test if using an external eGPU as secondary AV output I can trigger HLG on my LG Tv from my macbook. The smallest one I think is 43" if not 49" a tad on the large side

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


iPhone4TW liked
ReplyQuote
artik
(@artik)
Eminent Member
Joined: 2 years ago
 

In order to get HDR from FCPX we need expensive AJA I/O 4K ... 

Mac Pro 2013, 10 core, Sonnet eGFX 550, Vega 64


iPhone4TW liked
ReplyQuote
(@massimo_franzese)
Trusted Member
Joined: 1 month ago
 

@artik

Blackmagic Ultrastudio Mini has now been updated with HDMI 2.0b so may work to connect your Tv HLG set

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


iPhone4TW liked
ReplyQuote
MorganDe
(@morgande)
New Member
Joined: 2 months ago
 

@massimo_franzese

Thanks. I've been trying to avoid the aja i/o box route for a long time now. But I'm guessing now with the Mac Pro supporting a whole range of AMD graphics cards, and the xdr display, there ought to be some way to connect the cards to an HDR display. I saw a youtube video where someone connected a whole bunch of macs and pcs to the XDR display with tb3 and it worked with HDR. (I wish I had $6k for an amazing monitor. Maybe someday)

I can put the HLG footage (10bit, 420, 72mb/s h.265 HEVC) on a usb drive and watch it directly from my ROKU which is awesome. Oddly enough if I use quicktime I can trim and combine footage into assemblies and output HEVC QuickTime without reencoding!! It's not frame accurate. But it saves like a day and a half per hour.

Have you heard of anyone having any luck getting HDR out of an eGPU?

I'm trying to figure out what's the best route eGPU, hackintosh, or mac mini, moving forward.

 

Wh

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


iPhone4TW liked
ReplyQuote
(@massimo_franzese)
Trusted Member
Joined: 1 month ago
 

From final cut pro point of view the application does not support HDR ful stop

it relies on an AV output device to do it or the new display XDR that has hardware built in and therefore is similar to an AV output device anyway

there are potentially 4 output devices for HLG edit

1. AJA IO - on the fcp documentation 

2. Blackmagic Ultrastudio Mini 4K

3. Apple XDR display (does not say mini is compatible)

4. EGPU connected as output device in a multi display scenario

1 and 3 are expensive. 2 and 4 comparable if you have a machine working well with final cut pro I would go with the blackmagic

you also have to consider that HLG scopes are the same as HDR and therefore is not that part that is the real issue but BT.2020

I can see that there is no visual difference between tone mapped and not on my P3 display and therefore right now my conclusion is that a screen with appropriate color coverage is more important than an HDR screen

consider also that you need a calibrated monitor and a Tv isn’t one

If you wanted to go the PQ route most displays for gaming have HDR-10 but only cover RGB so are not suited

you need P3 at least 70% BT.2020 and 1000 nits which is fairly rare too

i am editing my GH5 HLG with a Benq PD2720U

seems to work fine

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


ReplyQuote
iPhone4TW
(@iphone4tw)
New Member
Joined: 2 months ago
 

@morgande

Sorry, I don't have any experience with GH5 and HDR footage.

After all those tests, I give up the eGPU solution, and keep those fancy graphic card in my hackintosh with profile "macpro7,1".

I think it is the best and easy way for macOS or for PC Gaming.

purge-wrangler ✧ 2014 iMac27 5K + VEGA 64


Username liked
ReplyQuote
naticom
(@naticom)
Active Member
Joined: 5 months ago
 

Never thought @iphone4TW would show up here!

Another Taiwanese here!

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote