Poor eGPU Performance in DR - why?

Hi there,

I just got the ne Macbook Pro 16'' with i9 2.3 GhZ 8 Core, 32 GB Ram, 5500m 8 GB. I also got a Razer Core X with a Vega 64 (Sapphire Nitro+ 8GB).

I tested the system over some hours and while on benchmarks the performance gain with the eGPU is big, inside Davinci Resolve I'm seeing very poor results! With NR on a 4K Clip, I get 8fps - though the GPU is not even close to being maxed out, same as the GPU. Same with exporting, playback, stabilizing...the GPUs potantial is being cut down to around the same as the internal GPU. The best I get is around 35% usage of the GPU!

I understand that for Redcode the Cudas are missing, but also with BRAW or cDNG the GPU doesn't unfold its potantial. And I don't come across a CPU bottleneck (except with Redcode).

I also tried switched between Open CL and Metal without seeing a difference.

Is there something that I might be missing out or is this a software related problem?

Thanks for the help!!

Cheers

4K is hard, and Raw is harder.

How does the Razer compare to the MacBook Pro in performance?

BMD has not yet updated its Configuration Guide for 16, but the Disk Speed Test will give you an idea of what your system can handle. There are also known "Candle Tests" out there that will quantify the kind of speed results you can get in Resolve.

If you want maximum performance, I'm not convinced a laptop is an ideal solution. There's a lot of things laptops can do really well, but rendering 4K with heavy NR is not among them.

4K is hard, and Raw is harder.

How does the Razer compare to the MacBook Pro in performance?

BMD has not yet updated its Configuration Guide for 16, but the Disk Speed Test will give you an idea of what your system can handle. There are also known "Candle Tests" out there that will quantify the kind of speed results you can get in Resolve.

If you want maximum performance, I'm not convinced a laptop is an ideal solution. There's a lot of things laptops can do really well, but rendering 4K with heavy NR is not among them.

1. Razers performance is almost the same as no eGPU; but I link this back to the poor usage of the potential compared to benchmark tests. It's not just 4k: generally the GPU is underused and the CPU except with RED not at its limit.

2. Isn't Disk Speed Test for drives only? Do you mean the BRAW test? This one seems not to be able to use a discrete / external GPU unless I overlooked a setting.

3. I know, but I would like to have the potential of a device used. I know my Macbook is not gonna deliver as a desktop, for most work it's enough; but it would be more powerful if that eGPU gets used efficiently, which is does not momentarily.

Thanks for the help.

Update: With a project that has a lot of noise and GPU heavy effects on it, I got the eGPU working around 70%. Thats a little bit more promising. But still I should get more or less 100% eGPU power as with the benchmarks. I rendere the same project on a windows workstation with an RTX 1080: there thes GPU was during the whole render on approx. 100%.

I hope that BMD keeps working on the eGPU implementation i regards of efficiency. Having a powerful laptop and eGPU is such a nice combo.

Are you using Resolve (free) or Resolve Studio (paid)?

The free version does not use the GPU as heavily as the Studio version does.

Given your second test, I bet your thunderbolt or usb controller on the motherboard is the bottleneck. Basically you can't get data to the gpu fast enough to fully leverage it. Your second test requires more processing for a given amount of data transferred, thus the higher utilization.

What type of connection are you using, and have you tried others if your machine has them?


You can't compare this to your windows machine. In it, the cpu and gpu can talk directly over a much faster lower latency protocol.

What I would do is go to Preferences-Memory and GPU

GPU Processing Mode-Metal
GPU Selection - Manual

Select Vega 64
Deselect the 5500m

This should have Resolve use only the Vega64.

I have a 2018 MBpro with a WX7100 and that’s what I do. I get better performance from this way than if I leave it on auto. Try and see if this works.

I'm struggling in the same way. I have terrible eGPU performance with a Radeon VII via Core X on a 2016 15" Macbook Pro w/ Radeon Pro 460 internally installed. The Radeon Pro 460 works so much better than my Radeon VII in Resolve and I can't quite explain it. I'm running Studio 16.1.2.026 and have changed from auto to many different configurations and anytime I'm forcing the Radeon VII to be the workhorse, I get awful results with the Radeon barely being stressed.
I hear people say it's a bottleneck issue of TB3 but I'm a bit skeptical, since when I run gaming benchmarks, it easily maxes out my Radeon VII and blows my Radeon Pro 460 away. Resolve? It's more useful as an expensive paperweight. Would love to have better understanding here.

Ok I worked one week in this setup now:

First I can say - I really encourage people to go with the new Macbook Pro if they would like to have just one machine and need a laptop. It's a beast that does it for most of the work.

The egpu definitely helps to keep the temp down of the machine and therefore also noise. When working with Photos I directly go from TB3 to Display, because of the 10-bit I get with the internal radeon.

Interesting: When opening the project with red files without having the media linked but the smart caching on, the eGPU (also internal if no eGPU) crank up to 100% memory and 100% clock. At least one case where Resolve really exhausts the eGPU^^

Weird (?): both with the discrete as with the eGPU memory on MacOS quickly reaches around 100% and stays there (not just in resolve)...this is unusual compared to my old windows machine. Not sure though if this affects performance in any case.

I would doubt that the TB3 is the bottelneck. Yes, TB3 will not give you the same Performance with the same card as in a PCIE slot directly, but it seems rather to be to some extend a software thing to not use the eGPU enough...implementation could definitely be better.

Regarding Radeon VII: this card is not officially supported by Apple why it might be even less integrated.

Interesting stuff. I was thinking about the 16in MBP. Good to hear it's been great for you.

Regarding Radeon VII: this card is not officially supported by Apple why it might be even less integrated.

I think it is supported officially these days unless I'm confused (not unlikely ). https://support.apple.com/en-us/HT202239 seems to point to it being supported.

With Metal enabled and noise reduction the Processor of the Vega 64 almost reaches 100%.

Good news, patience is rewarded!

The only thing that distracts me is that the GPU memory is almost constantly on 95%+ and I don't know wether thats an issue. There are a number of threads online regarding this

Janis, i also have a MBP 16 for resolve and i am about to purchase en eGpu to help speed up my system. have you since you last posted about your underutilization of the egpu been able to use it to its full potential? there are countless videos on youtube showing that egpus do help all MBPs except the latest full specs 16 inches. which makes me confused. do you recommend complementing with an EGPU for resolve grading? i am less interrested in export time than the whole fluid grading experience (NR, etc)
thank you in advance

I found this topic and am having the same issue. My Macbook Pro 16 was overheating during Resolve use (and other things) so I grabbed a Razer Core X and dropped in a Radeon VII in hopes of better performance and heat/noise reduction. The heat and noise have dropped but the performance is worse than using the built-in 5500m, which is quite disappointing.

I changed the resolve preferences to Metal/Manual and left only the Radeon VII selected. I also enabled Smart Cache. When doing a 4K render it takes roughly 1.4X the time to render vs. using the 5500M integrated in the Macbook Pro (8 core i9, 16GB RAM, 1TB SSD).

I'd love to hear what's working for others or what might be going on.

samuch,
despite the 1.4x rendering time, did you find that egpu helped with the general grading experience, playback performance with multiple nodes, NR maybe, basically all the workflow up to the point where you had to render / export?

I just got the ne Macbook Pro 16'' with i9 2.3 GhZ 8 Core, 32 GB Ram, 5500m 8 GB. I also got a Razer Core X with a Vega 64 (Sapphire Nitro+ 8GB).

Although modern PC AMD graphics work out of the box in recent MacOS versions, they are actually performing significantly lower than their original Mac counterparts because of using a different framebuffer according to the macrumors forum user CMMChris. See the following thread for a possible performance boost.

The only thing that distracts me is that the GPU memory is almost constantly on 95%+ and I don't know wether thats an issue. There are a number of threads online regarding this

Are you using Temporal NR? How many reference frames? Does GPU memory usage (and/or performance) scale up and down with the number of reference frames used?

samuch,
despite the 1.4x rendering time, did you find that egpu helped with the general grading experience, playback performance with multiple nodes, NR maybe, basically all the workflow up to the point where you had to render / export?

I've noticed that some things are faster and some slower. That said, it feels random and the differences are trivial. Watching a NR clip in the editor now stutters to play (framerate drops to like 4fps and it stutters through) vs not really being viewable at all with builtin Radeon. But then things like crossfades between two clips might not show smoothly and hits the fps, which never happened with the builtin Radeon.

The article posted after this was interesting and seems to imply Apple is actually limiting the cards that aren't Apple provided. That would make sense.