iMac Pro + eGPU for FCPX, does it actually do much?

[rondocap]

So I have a Vega 56 iMac Pro, and also an eGPU with a 5700XT. Most of the time, the eGPU doesn't seem to be doing much, only seems to really kick in if you do something like denoise or other heavy affect.

What else is it actually doing in FCPX? Does it help at all with timeline performance, render speed, export, etc?

Is there a list of things you've seen that it actually helps speed up? Can it actually hinder or slow down certain performance aspects?

I. know the internet Vega 56 is pretty powerful already, but just curious what the eGPU is actually doing.

 

[kohlson]

Have you tried using Activity Monitor (in Utilities)? Under Window is GPU History. When I activate this on my MBP, it shows activity for both iGPU and dGPU.

In FCPX, the GPUs are busier than the CPU - import/timeline/export. On export, for example, you can see that there is some cycling between all 3 compute resources.

 

[kohlson]

9to5

 

[Gwendolini]

9to5

Does the light panel help with rendering?

 

[sevoneone]

I assume you set the preference in Final Cut (dropdown in the editing tab) to use the 5700XT for Rendering/Sharing? FCP will use multiple cards for some tasks that it can split up, but most it is going to run on the primary card only from what I've seen.

If I remember right, 5700XT performance falls right between the Vega 56 and Vega 64. There is going to be some performance loss running an eGPU... So if you're looking for improved performance, the cards are probably about equal given their respective connections. The Vega 56 being internal, I wouldn't be surprised if it is even a bit faster than an external 5700XT.

 

[rondocap]

I assume you set the preference in Final Cut (dropdown in the editing tab) to use the 5700XT for Rendering/Sharing? FCP will use multiple cards for some tasks that it can split up, but most it is going to run on the primary card only from what I've seen.

If I remember right, 5700XT performance falls right between the Vega 56 and Vega 64. There is going to be some performance loss running an eGPU... So if you're looking for improved performance, the cards are probably about equal given their respective connections. The Vega 56 being internal, I wouldn't be surprised if it is even a bit faster than an external 5700XT.

The Vega 56 is definitely faster, so I keep that on the primary.

Related question - how about a 2013 Mac Pro with D700s? Would a thunderbolt 2 limited 5700xt be faster? I did a few tests and the D700s actually still seemed faster - maybe I can keep the 5700xt just for the occasional help as a secondary gpu.

 

[joema2]

Vega 56 iMac Pro, and also an eGPU with a 5700XT....What else is it actually doing in FCPX? Does it help at all with timeline performance, render speed, export, etc?...how about a 2013 Mac Pro with D700s? Would a thunderbolt 2 limited 5700xt be faster? I did a few tests and the D700s actually still seemed faster - maybe I can keep the 5700xt just for the occasional help as a secondary gpu.

I formerly owned a top-spec D700 Mac Pro. It's Achilles Heel was no hardware acceleration for video decode or encode. It was fast on ProRes and FCPX was fast on effects because it leveraged the dual D700s, but using H264 or most any Long GOP codec it was slow.

In general a GPU will not help encode or decode. The GPU can harness hundreds of lightweight threads to attack a graphical problem in parallel, but decode/encode of H264 and HEVC are inherently sequential. The core algorithm must be run faster, which requires dedicated hardware such as Intel's Quick Sync. Unfortunately Xeon doesn't have this.

The iMac Pro uses Xeon but has a T2 chip Apple uses instead of Quick Sync. It's way better than the "Trash Can" but it's still not as smooth and responsive as the latest generation Macs such as a 2019 MacBook Pro 16 or 2020 iMac 27.

GPUs have dedicated video acceleration which is totally separate from the normal GPU logic. It's bundled on the card and accessed with separate APIs. nVidia GPUs use NVDEC/NVENC, AMD has UVD/VCE, and more recently VCN. However each of these were released in various versions with differing capability, bugs and software frameworks. Quick Sync has had different versions but it was a more constant facility, so most developers used that.

Besides Long GOP there are other problematic codecs such as 4k 10-bit 4:2:2 All-Intra from Panasonic and the new Sony A7SIII. The latest versions of DaVinci Resolve on recent Macs are a lot faster than FCPX 10.4.9 on these. Apparently Apple needs to upgrade FCPX to more effectively leverage hardware video acceleration.

I have a 10-core Vega64 iMac Pro, so understand your concerns. the Vega 56 or 64 GPUs are actually pretty fast, but when doing stuff like Neat Video there is no such thing as fast enough.

Max Yuryev and his other site Max Tech have done lots of eGPU testing on various Macs and NLEs. In general the results are sometimes useful but overall not great.

My suggestion is stick with what you have or maybe as an interim step get a top-spec 2020 iMac 27. That would help some on Long GOP decode/encode, and it's possible Apple might later leverage AMD's new VCN video accelerator on the XT 5700.

I believe with the new Apple Silicon Macs Apple will unify and expand hardware video acceleration, and help clean up the mess of all the varying standards. We can see an early indication of this on the 2020 iPad Pro where the LumaFusion NLE runs most of these "difficult" codecs very well.

This coming week Max Yuryev will release extensive video editing benchmarks on the 10-core 2020 iMac 27 with 5700 XT. That will show the good, the bad and the ugly of how FCPX and other NLEs work on that hardware today. It may also give hints of what future improvement is possible on that same hardware while we wait for Apple Silicon Macs.

If you have specific workflow elements that are slow, examine these closely - don't just throw hardware at it. E.g, if Neat Video is a problem, the latest versions are a lot faster. Upgrade to those. Use the built-in Neat Video performance optimizer to evaluate and configure it for your hardware. For other plugins or Fx, do likewise.

FCPX is a lot faster if using proxies. The previous proxy system was fragile and difficult, but 10.4.9 has totally revamped this, including customizable proxy sizes and relink ability. Examine this closely - it could greatly improve performance on difficult cases, and using it costs nothing.

Edit/add: See these tests just posted by Max Tech comparing the base iMac Pro vs the new 10-core iMac 27 on various video editing tasks:

 

[rondocap]

I formerly owned a top-spec D700 Mac Pro. It's Achilles Heel was no hardware acceleration for video decode or encode. It was fast on ProRes and FCPX was fast on effects because it leveraged the dual D700s, but using H264 or most any Long GOP codec it was slow.

In general a GPU will not help encode or decode. The GPU can harness hundreds of lightweight threads to attack a graphical problem in parallel, but decode/encode of H264 and HEVC are inherently sequential. The core algorithm must be run faster, which requires dedicated hardware such as Intel's Quick Sync. Unfortunately Xeon doesn't have this.

The iMac Pro uses Xeon but has a T2 chip Apple uses instead of Quick Sync. It's way better than the "Trash Can" but it's still not as smooth and responsive as the latest generation Macs such as a 2019 MacBook Pro 16 or 2020 iMac 27.

GPUs have dedicated video acceleration which is totally separate from the normal GPU logic. It's bundled on the card and accessed with separate APIs. nVidia GPUs use NVDEC/NVENC, AMD has UVD/VCE, and more recently VCN. However each of these were released in various versions with differing capability, bugs and software frameworks. Quick Sync has had different versions but it was a more constant facility, so most developers used that.

Besides Long GOP there are other problematic codecs such as 4k 10-bit 4:2:2 All-Intra from Panasonic and the new Sony A7SIII. The latest versions of DaVinci Resolve on recent Macs are a lot faster than FCPX 10.4.9 on these. Apparently Apple needs to upgrade FCPX to more effectively leverage hardware video acceleration.

I have a 10-core Vega64 iMac Pro, so understand your concerns. the Vega 56 or 64 GPUs are actually pretty fast, but when doing stuff like Neat Video there is no such thing as fast enough.

Max Yuryev and his other site Max Tech have done lots of eGPU testing on various Macs and NLEs. In general the results are sometimes useful but overall not great.

My suggestion is stick with what you have or maybe as an interim step get a top-spec 2020 iMac 27. That would help some on Long GOP decode/encode, and it's possible Apple might later leverage AMD's new VCN video accelerator on the XT 5700.

I believe with the new Apple Silicon Macs Apple will unify and expand hardware video acceleration, and help clean up the mess of all the varying standards. We can see an early indication of this on the 2020 iPad Pro where the LumaFusion NLE runs most of these "difficult" codecs very well.

This coming week Max Yuryev will release extensive video editing benchmarks on the 10-core 2020 iMac 27 with 5700 XT. That will show the good, the bad and the ugly of how FCPX and other NLEs work on that hardware today. It may also give hints of what future improvement is possible on that same hardware while we wait for Apple Silicon Macs.

If you have specific workflow elements that are slow, examine these closely - don't just throw hardware at it. E.g, if Neat Video is a problem, the latest versions are a lot faster. Upgrade to those. Use the built-in Neat Video performance optimizer to evaluate and configure it for your hardware. For other plugins or Fx, do likewise.

FCPX is a lot faster if using proxies. The previous proxy system was fragile and difficult, but 10.4.9 has totally revamped this, including customizable proxy sizes and relink ability. Examine this closely - it could greatly improve performance on difficult cases, and using it costs nothing.

Edit/add: See these tests just posted by Max Tech comparing the base iMac Pro vs the new 10-core iMac 27 on various video editing tasks:

Hi, first- thank you so much for this very detailed and thorough reply, it’s exactly what I was looking for and explains it very well.

I experience with 4 systems:
IMac Pro w Vega 56
2013 Mac Pro 12 core with D700

PC- 9940x with 2080ti
10900k with 2080 Super

Running Davinci Resolve on the PCs and fcpx on the macs



1. Redcode raw works pretty well on all, especially with the new fcpx update - even the 2013 Mac Pro does decently. The powerful PCs are often twice or more as fast vs the older macs for certain things, but the new fcpx update really helps

2. Everything from the Canon R5 gives all the systems a hard time, aside from 8k raw which is easier to edit. Like you said, the codec Is the issue here.

You’re right, throwing hardware at things often isn’t the first answer, even with the Mac Pro 2019- unless it targets a specific workflow like prores.

 

[0279317]

I have an iMac Pro (base with 64 GB RAM) and a 5700 XT eGPU. I notice a considerable improvement in FCPX rendering. Not earth shattering but for sure better with the 5700 XT. Benchmarks show a ~30% improvement which seems about right. I had bought the card to pep up my old 2015 iMac which is now gone, deceased. So I tried it with the iMac Pro. I don't think it's worth the purchase for the purpose of using it with an iMac Pro though.