[Testing Project Finale] Dose THE REAL Mac Pro 2019 make a home run ?
 
Notifications

[Testing Project Finale] Dose THE REAL Mac Pro 2019 make a home run ?  

  RSS

iPhone4TW
(@iphone4tw)
Active Member
Joined: 3 months ago
 

Previously

It's all begin with several month ago, I want to make a video to share with my friends how eGPU rocks, and how eGPU helps me to save much time in video editing.

Especially when macOS 10.15.1 Catalina officially support AMD Radeon RX5700 XT with Metal .

I am so curious what's the difference between new RX5700 XT and my own VEGA 64 .

This is my first article : https://egpu.io/forums/pro-applications/testing-project-for-rx5700-xt-rx5700-vega-64-with-fcp-x-and-macos/

If you want to know the full story, you can check it out with the url above.

All those test base on this 38 mins video footage.

All media files, output video target storage is the same USB 3.0 flash drive with 1TB SSD inside the enclosure, reading and writing speed is around 400MB/s, I think it's not the bottle neck of the testing project.

 

The result is blow my mind, because no matter I am using PurgeWrangler or the 2018 Mac Mini for official Thunderbolt 3 eGPU support.

There are two variables with this test : macOS Version and Final Cut Pro X Version.

This is the video, that I get the best result of whole testing with my 2014 Late iMac27 5K with AMD R9 M295X internal graphic card. You can check out all information with the CPU, GPU, Disk i/O, sorry for that I recording with Mandarin, but you guys are pro, you know what you need to know.

Even VEGA 64 is faster 3.5 times than R9 M295X, but it still helps nothing.

R9 M295X internal graphic card is 30mins faster than all other result, finish time is 1 hour and 4 min 19 sec.

So, it must be the fault that I use it in external way .

Internal way

Few month later, I share another article : https://egpu.io/forums/pro-applications/testing-project2-is-external-gpu-faster-than-internal-gpu-while-exporting-video-in-fcp-x/

If it was the fault with external way, how about we all install it onboard directly?

Thanks to ASRock gave me the Phantom Z390 Gaming mainboard with Thunderbolt3.

And Kevin from XFastest support PC experience.

MAQ support dark side experience.

This time we have AMD Radeon VII for the test.

Full story: https://egpu.io/forums/pro-applications/testing-project2-is-external-gpu-faster-than-internal-gpu-while-exporting-video-in-fcp-x/

All graphic card we installed it onboard directly!!

The best combination of the result is macOS 10.14.6 with FCP X 10.4.6 at 1 hour and 30 mins 28 secs.

It was 30 mins slower than 2014 iMac27 5K internal graphic card !

WT__.........

I don't get it at all !

Most surprise thing was we find the third variable !

In system info, we change the Mac Model from MacPro6,1 to MacPro7,1, to see how will that change the result.

But we just modify the Mac Model, how would that be different with the same hardware ?

This is the result with Mac Model MacPro6,1 , it takes 1 hour and 30 mins 28 sec.

After we modify the Mac Model to MacPro7,1.

Damn, what the ...

As you can read from the result list above.

RX570 can run in 1 hour 18 secs, it's 30 mins faster than with the Mac Model is MacPro6,1.

But all other expensive graphic card such as Radeon VII, RX5700 XT, VEGA 64 only cut about 8 mins.

What?!

Ok, fine !

It seems graphic card do nothing with all these test, no matter it is internal or external.

And no matter how expensive, or how fancy it is, in FCP X export testing with "Video and Audio , H.264 .mov " , it takes all most the same time to export.

Now I got it, it must be the CPU's fault, right?

Intel Core i9-9900KS

I ask MAQ to unbox a brand new Intel Core i9-9900KS CPU for testing.

After start testing, my jaw drops again.

CPU do nothing at this time, but GPU with almost full loading.

What the .....

The best result of the all time is 44 mins 38 secs !!

Here is the all result.

The more test I made, the more doubt pop out to my mind.

Thought Thunderbolt 3 with eGPU -> won't work

Graphic card directly installed onboard -> won't work

Change Mac Model name -> it fxxking works!

Change CPU to i9-9900KS -> CPU do nothing but GPU works.

So, what it the bottle neck of all this test ?

The same hardware with Mac Model name from MacPro6,1 to MacPro7,1 can save 30 mins export time.

It's almost 1/3 times of the whole export time.

Maybe, I just think that maybe it's how apple want us to purchase the " Real Mac Pro 2019 " ?

 

The Real Mac Pro 2019

Ok, fine.

Last week, one of my friend got his Mac Pro 2019.

It's time to reveal all those question in our mind.

" The REAL " Mac Pro 2019 ( MacPro7,1 )

It's 3.2 GHz 16-core Intel Xeon W processor.

Test with 32 GB RAM onboard with macOS 10.15.2

With all the same footage and external flash drive.

I start test with latest version of macOS 10.15.2 + FCP X 10.4.8.

But I give up in export to10%, because it is taking 7 mins 17 secs, in my experience it will take over 1 hour at the end.

How about the best combination macOS 10.15.2 + FCP X 10.4.6 ?

I ask my friend after he spend so much money, how much time he wish to reduce in this test ?

The Real Mac Pro 2019 is really a master piece in the computer world.

And it is the real Mac Pro 2019 , not just we mess around with the Mac Model.

輸The result of the real Mac Pro 2019 is 43 mins 21 secs.

OMG !

It "ONLY" 1 min faster than my best result with i9-9900KS + RX5700 XT.

1 min ! 1 min! 1 min !

If I spend 13,999 USD for reduce 1 min, I will be kill by my wife.

I still don't get it.

It must be MY FAULT

This result is too shock to me.

It must be my fault that :

I am not expert in FCP X.

I just a spoiled brat that only shooting and editing 1080P60 video, does not deserve to own this holy grail.

It must be I using not enough transition, special effect and 38 mins footage is too short for this holy grail.

It must be I just filming with A7M3 and GoPro , it should use RED for filming everything with 8K60P

Finale

After 3 months later, today we have this finale to all you guys interested in my test.

From curiosity to the end,

It must be my fault to keep using FCP X for video editing.

But,

Hello!

FCP X is made by Apple, it's one of the Pro application that Apple very proud to announce to the video industry.

If you spend 13,999 dollars for all these results, what will you do ?

What is this story told us ?

Keep in Dark Side ?

Keep in eGPU ?

Keep in FCP X ?

It must be my fault that I don't have enough faith of apple.

Think about that:

Late 2014 iMac27 5K with internal graphic card AMD Radeon R9 M295X takes 1 hour 18 secs to finish the job.

Hackintosh with i9 CPU takes 44 mins. (it costs around $2,500)

The Real Mac Pro 2019 takes 43 mins to save 17 mins. (it costs $13,999)

I will choose keep using my 2014 iMac27 5K with internal GPU.

I still remember that FCP X 10.3.9 had great benefit performance with eGPU.

Thanks to all you guys and PurgeWrangler that made our old Mac to have a whole new life.

Sorry for my pool English, hope you guys get something with my whole test.

 

Last share with you the real Mac Pro 2019 testing time-lapse.

This topic was modified 3 weeks ago

purge-wrangler ✧ 2014 iMac27 5K + VEGA 64


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

iPhone4TW liked
ReplyQuote
galad
(@galad)
Active Member
Joined: 7 months ago
 

GPU can accelerate only a part of the video process: filters. If your workflow is just to cut and paste video clips around, a GPU has got nothing to do.

A GPU can't accelerate ProRes encoding/decoding. It has got a dedicated ASIC to decode/encode H.264 and HEVC maybe, but encoding quality is lower than a software encoder, and the decoder can decode only a limited number of video tracks simultaneously.

The H.264/HEVC encoder/decoder ASIC in your GPU or in QuickSync can be easily the bottleneck. You throw a lot of unneeded hardware without understanding where was the actual bottleneck.

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


iPhone4TW liked
ReplyQuote
Dope
 Dope
(@dope)
Active Member
Joined: 2 years ago
 

What is your test footage specs? H.264 8-bit? H.265 10-bit? What bit rates?

 

what was your output encode specs? H.264 8-bit? H.265 10-bit? What bit rates?

 

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


ReplyQuote
iPhone4TW
(@iphone4tw)
Active Member
Joined: 3 months ago
 

@dope

Most of footage shot by GoPro Hero8 , others shot by Hero7 and A7M3.

I use FCP X default export setting.

 

 2019 JCMS 2020 01 30 at 13.10.56

purge-wrangler ✧ 2014 iMac27 5K + VEGA 64


ReplyQuote
iPhone4TW
(@iphone4tw)
Active Member
Joined: 3 months ago
 
Posted by: @galad

The H.264/HEVC encoder/decoder ASIC in your GPU or in QuickSync can be easily the bottleneck. You throw a lot of unneeded hardware without understanding where was the actual bottleneck.

Thank you for your information.

In this case, do you have any suggestion to solve this bottleneck ?

purge-wrangler ✧ 2014 iMac27 5K + VEGA 64


ReplyQuote
Dope
 Dope
(@dope)
Active Member
Joined: 2 years ago
 

@iphone4tw

I assume when you did the test you used the same source test file..

I don't know if you care for fast encoding or quality cinematography but if you just want faster encoding, just set all of your cameras to record at codec your computers and software support with hardware acceleration. I don't know if you're using a hackintosh or a real Mac so I cannot give you exact advice but try checking if your computers work with h.264 or h.265 and then set your cameras to that codec. For quality cinematography, transcoding is the only way and there's really no fast way around it. Video editing is not just rendering time, better hardware allows you to edit "easier" and more enjoyably.

This post was modified 3 weeks ago

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


ReplyQuote