FCPX isn't using 280x GPU

[davehavers]

Hey Guys,

i recently replaced my GTX 770 with a dual R9 280x configuration, because i had lots of freezes with the nvidia card.

First, i ordered a single 280x and it worked out of the box. I tried FCPX and it worked, but 4K rendering seemed to be a bit slow. a few days ago i ordered a second 280x and rendered the same Clip, but it didn't changed anything. It still takes the same time to render the clip.

So i decided to take a look at the GPU Workload while rendering in FCPX. Well, iStat Menu shows me very little graphics card RAM and GPU load on the first card, but no load on the second card.

What can i do?

My specs:

Gigabyte GA-Z87-HD3
i7 4770
2 x Sapphire R9 280X Vapor-X OC
4x 4GB Corsair Vengeance
El Capitan installed with Clover


Edit:

I tried Motion and it uses both GPUs.

 

[sevoneone]

I don't know much about your build, but are you using an SSDT? I have R9 270x and noticed that when using an SSDT with Clover and El Cap that my GPU was under performing substantially: by 35%-50% in every benchmark. Without the SSDT it runs fantastic.

FCPX loves ATI/AMD. I was in the same boat as you with a GTX 760 that just lagged in FCPX. The Iris chipset in my 2013 13" MBP would runs FCPX more smoothly than that card ever did.

 

[rwillett]

I don't know much about your build, but are you using an SSDT? I have R9 270x and noticed that when using an SSDT with Clover and El Cap that my GPU was under performing substantially: by 35%-50% in every benchmark. Without the SSDT it runs fantastic.

FCPX loves ATI/AMD. I was in the same boat as you with a GTX 760 that just lagged in FCPX. The Iris chipset in my 2013 13" MBP would runs FCPX more smoothly than that card ever did.

I do not use any SSDT with my dual Sapphire 280X system. I use a Gigabyte Z97X UD5H but did use them both in a Z87X UD3H system as well. I have no issue with FCPX using both Sapphire cards. Works very well with both cards and zero (and I really mean zero) configuration.

I would check

1. Clover
2. SSDT (do you have one?)
3. Do you have the internal GPU adaptor (is there one?) and is it disabled?
4. Which version of FCPX?
5. Which version of OSX?

Rob

 

[mojimoji]

I am using Gigabyte R9 280x x2 and FCPX can utilised both card perfectly (monitor by iStat).
No any special config require, just only GraphicsEnabler=No

by the way...
- Cross Fire connector disconnect/connect its doesn't make any different.

 

[Total_Noob]

My new El Cap/Clover setup with a single 280x is currently running slow.
Bruce 5k runs at 32 seconds (although this isn't a catch-all perfect test of FCPX)
If I plug one of my monitors into the MOBO the Bruce time increases to 45 seconds.

Currently, if I boot back into Yosemite I get almost double the LuxMark result.
But the Bruce test is still around 30 seconds, probably because the Yosemite install is running off an HDD, and FCPX/Bruce might be pulling some of the generators used in the test from the HDD...

 

[rwillett]

My new El Cap/Clover setup with a single 280x is currently running slow.
Bruce 5k runs at 32 seconds (although this isn't a catch-all perfect test of FCPX)
If I plug one of my monitors into the MOBO the Bruce time increases to 45 seconds.

Currently, if I boot back into Yosemite I get almost double the LuxMark result.
But the Bruce test is still around 30 seconds, probably because the Yosemite install is running off an HDD, and FCPX/Bruce might be pulling some of the generators used in the test from the HDD...

This is what we would expect to see. If you plug a monitor into your system board then FCPX will try and use the system board graphics for rendering. It *appears* that FCPX prioritises internal graphics over dedicated GPU's.

32 secs for BruceX is pretty good. The biggest bottleneck in the system is NOT your HDD but the rendering. There is some disk activity but very little overall.

If you want to use FCPX properly then read this thread

http://www.tonymacx86.com/buying-ad...dware-software-z97x-motherboard-i7-4790k.html

and this one

http://www.tonymacx86.com/buying-advice/161933-final-cut-pro-x-graphics-card.html

Disable the internal graphics card when using FCPX is the best advice to speed things up.

Rob

 

[Total_Noob]

32 seconds under El Capitan is a little disappointing.
That's marginally slower than I was getting under 10.8, and El Cap should have improved the times quite a bit. Not that I'm complaining.

I mention the HDD in my Yosemite boot up because FCPX does call on the HDD to write to the library, pull in the generators and so on. It's not a huge data throughput, but lots of small files etc.

Anyway, more importantly for me is that the LuxMark score is currently crippled under El Cap/Clover and I'm sure the this is reflected in the Bruce 5k times.

Thanks for your links in your reply, kind of you to take the time.
I should point out that I'm a Pro, I was editing in Soho almost 20 years ago!

My current setup was built for FCPX in late 2012, and it has served me really well.
It plays 4k if I need it, though newer systems like yours would no doubt wipe the floor with it.

 

[rwillett]

32 seconds under El Capitan is a little disappointing.
That's marginally slower than I was getting under 10.8, and El Cap should have improved the times quite a bit. Not that I'm complaining.

I mention the HDD in my Yosemite boot up because FCPX does call on the HDD to write to the library, pull in the generators and so on. It's not a huge data throughput, but lots of small files etc.

Anyway, more importantly for me is that the LuxMark score is currently crippled under El Cap/Clover and I'm sure the this is reflected in the Bruce 5k times.

Thanks for your links in your reply, kind of you to take the time.
I should point out that I'm a Pro, I was editing in Soho almost 20 years ago!

My current setup was built for FCPX in late 2012, and it has served me really well.
It plays 4k if I need it, though newer systems like yours would no doubt wipe the floor with it.

I found the move to El Capitan AND to FCPX V10.2.2 to make a dramatic difference to the Bruce X benchmark. From memory I went from 23/24 secs to 16 secs.

The disk activity is still low, if you run the test a few times to get an average, then FCPX will have pulled any generators in and OS X will have cached most of them anyway just in case it does have to go out to disk. I'd be surprised if it made more than a few tenths of seconds worst case (assuming you have a reasonable disk). Whilst SDD's are fast, nothing beats having stuff cached in RAM unless its cached on the CPU.

No idea about Luxmark as never been interested in it.

Your nickname of Total_Noob tends to give the opposite of being a Pro

At the end of the day, the rendering speed doesn't make a massive difference, its just one part of the overall workflow. I used to commute to Belgium and would sometimes spend my journey doing FCPX on my 2012 Macbook Pro. Massively slower than my Hack, but you adapt to what you have.

32 secs would have been a good score on Yosemite and the older FCPX but I suspect something else is slowing you down for El Capitan and FXPX 10.2.2. Are you outputting ProRes or H264? And are you getting 32secs for 5K or 4K?

Rob

 

[Total_Noob]

I've just manually copied the EFI/Clover contents from my USB stick back into the SSD EFI partition, then using Text Edit edited the config file with the tweaks I've had to make to deal with the 'Scan Entries' error I was getting with my RAID array attached at boot up.

Now, FCP exports BruceX at 5k in 24 seconds and LuxMark 3.1 is back up to 13332 on the luxball test.
So it looks like it's up to full speed!

Re: my Total Noob tag, when I first selected the name in 2012 it was very suitable, yet it still feels right especially now we are using Clover. It doesn't fill me with any confidence, yet.

 

[rwillett]

I've just manually copied the EFI/Clover contents from my USB stick back into the SSD EFI partition, then using Text Edit edited the config file with the tweaks I've had to make to deal with the 'Scan Entries' error I was getting with my RAID array attached at boot up.

Now, FCP exports BruceX at 5k in 24 seconds and LuxMark 3.1 is back up to 13332 on the luxball test.
So it looks like it's up to full speed!

Re: my Total Noob tag, when I first selected the name in 2012 it was very suitable, yet it still feels right especially now we are using Clover. It doesn't fill me with any confidence, yet.

Thats a good speed for a single 280X on your kit.

I might try the Lexmark test just to see what the score is.

I use Clover over Multibeast. I am utterly clueless about Clover having done nothing but install it and use it. I honestly know more about major brain surgery than configuring Clover, however it "just works" for me. Clover was so much more stable than Multibeast for my setup, however I understand it can be complex and difficult (just not for me).

Rob