Caprure 1 and Blackmagic eGPU for Mac

[Wesley]

I suppose you can try turning off OpenCL, quit CO, and turning it back on again if you haven't already done so.

Post your CO GPU benchmark in ImageCore. Also what does it list OpenCL and GPU driver version as?

How do I get those scores? I didn't do any systematic benchmarking when I was evaluating. I'm not sure how to turn off OpenCL either, but what would that do anyway? Wouldn't both my onboard Radeon 460 and the external Radeon RX 580 be similarly affected?

OpenCL is in CO Preferences -> General -> Hardware Acceleration

Turning it off and on should have CO benchmark all your GPU and let it know you got a new GPU.
You have the monitor connected to the eGPU, right?

I can find my scores in:
Main drive -> Users -> Shared -> Capture One -> ImageCore - > look at all the xml files

[hyperlinked]

Ok, here are my scores. The Blackmagic's benchmark score is under half of what it is for the onboard Radeon 460, which is about what I'd expect based on other benchmarks I've seen. So does this look like the issue I experienced was genuine and not just a case of the eGPU not being utilized? I did get about a 20% boost in export performance.

For the onboard Radeon 460

Code: Select all

<?xml version="1.0" encoding="utf-8"?>
<device>
  <Index>1</Index>
  <Platform>Apple</Platform>
  <Version>OpenCL 1.2 (May 24 2018 20:07:03)</Version>
  <Device>AMD Radeon Pro 460 Compute Engine</Device>
  <Driver>1.2 (Jun 29 2018 18:33:51)</Driver>
  <GlobalMemoryMB>4096</GlobalMemoryMB>
  <NumberOfKernels>1133</NumberOfKernels>
  <NumberOfCompiledKernels>1133</NumberOfCompiledKernels>
  <Benchmark>0.295320</Benchmark>
  <Status>Benchmarked:CL_SUCCESS</Status>
</device>


For the Blackmagic eGPU Radeon 580

Code: Select all

<?xml version="1.0" encoding="utf-8"?>
<device>
  <Index>2</Index>
  <Platform>Apple</Platform>
  <Version>OpenCL 1.2 (May 24 2018 20:07:03)</Version>
  <Device>AMD Radeon Pro 580 Compute Engine</Device>
  <Driver>1.2 (Jun 29 2018 18:33:51)</Driver>
  <GlobalMemoryMB>8192</GlobalMemoryMB>
  <NumberOfKernels>1133</NumberOfKernels>
  <NumberOfCompiledKernels>1133</NumberOfCompiledKernels>
  <Benchmark>0.120880</Benchmark>
  <Status>Benchmarked:CL_SUCCESS</Status>
</device>


[CorsairVelo]

I'm late to this thread but very interested in where it goes. One thing which seems a little "smoke and mirrors" is how C1 actually uses GPU (vs CPU vs RAM etc). I found this article which helps: https://www.phaseone.com/en/Search/Arti ... cleid=1720

It says:

Open CL will dedicate memory to several actions in Capture One.
The following outlines the softwares demand;
• Preview Update with different settings, styles etc: RAM
• Sorting/Rating: CPU cores and SSD speed
• Fit Image to Screen: GPU cores
• Process time: GPU processing units and CPU's and RAM

So GPU only aids in the bottom two (Fit image to screen and Process time). I did some testing on a couple of my machines and clearly Process time benefits greatly from GPU. I find the first bullet a bit wanting saying that Preview Update depends on RAM; I guess that includes CPU speed as well (?).

I think the "Fit Image to Screen" is the key for day-to-day performance when editing. Since this is highly dependent on GPU I would expect the eGPU to help here but it's hard to measure. It may also depend on proper setting of Preview size. If the "fit image" process depends on size of existing Preview, then if you generate unnecessarily large Previews, it could slow the whole process down. I've recently set my previews more appropriately for my iMac and it improved things. I just haven't figured out how to quantify it.

Anyway, my interest is based on a potential move to a laptop based solution with an external 4k monitor (or two) and the eGPU discussion interests me and it's very hard to determine how important GPU, CPU and CPU core count is in real world performance.

EDIT" What's disappointing in this thread is that the eGPU introduced a lag in normal editing and browsing images for one user here. WOuld love ot hear more followup on that.

[Wesley]

Ok, here are my scores. The Blackmagic's benchmark score is under half of what it is for the onboard Radeon 460, which is about what I'd expect based on other benchmarks I've seen. So does this look like the issue I experienced was genuine and not just a case of the eGPU not being utilized? I did get about a 20% boost in export performance.

For the onboard Radeon 460

Code: Select all

<?xml version="1.0" encoding="utf-8"?>
<device>
  <Index>1</Index>
  <Platform>Apple</Platform>
  <Version>OpenCL 1.2 (May 24 2018 20:07:03)</Version>
  <Device>AMD Radeon Pro 460 Compute Engine</Device>
  <Driver>1.2 (Jun 29 2018 18:33:51)</Driver>
  <GlobalMemoryMB>4096</GlobalMemoryMB>
  <NumberOfKernels>1133</NumberOfKernels>
  <NumberOfCompiledKernels>1133</NumberOfCompiledKernels>
  <Benchmark>0.295320</Benchmark>
  <Status>Benchmarked:CL_SUCCESS</Status>
</device>


For the Blackmagic eGPU Radeon 580

Code: Select all

<?xml version="1.0" encoding="utf-8"?>
<device>
  <Index>2</Index>
  <Platform>Apple</Platform>
  <Version>OpenCL 1.2 (May 24 2018 20:07:03)</Version>
  <Device>AMD Radeon Pro 580 Compute Engine</Device>
  <Driver>1.2 (Jun 29 2018 18:33:51)</Driver>
  <GlobalMemoryMB>8192</GlobalMemoryMB>
  <NumberOfKernels>1133</NumberOfKernels>
  <NumberOfCompiledKernels>1133</NumberOfCompiledKernels>
  <Benchmark>0.120880</Benchmark>
  <Status>Benchmarked:CL_SUCCESS</Status>
</device>


Try deleting all the xml files and relaunching CO.

[hyperlinked]

Try deleting all the xml files and relaunching CO.

Unforunately, I've already returned the Blackmagic eGPU so I won't be able to follow your suggestion and see if it helps.

Hopefully the next person who runs into the same issue I did will be able to give this a try and report back if what I experienced was normal or not. If it's not and I can get a serious boost in my culling workflow, I'd consider buying the Blackmagic eGPU again.

[NN636274225804497206UL]

This is a late reply which may shed light on hyperlink's problem with the Blackmagic eGPU. I ended up buying an OWC "Helios" eGPU chassis with a Radion "RX" 580 card. I bought both as a package from OWC for $549 (vs $699 for the blackmagic) as the card is virtually identical and you can replace it later on if need be. BTW, The helios chassis appears to be a Sonnet chassis rebranded for OWC.

In using this setup with a 2018 Macbook Pro 13", I noticed that sometimes it worked perfectly with eGPU, and sometimes there was a significant lag when using the eGPU. The lag would appear in various places but the most obvious was when I clicked on different tool tabs, the actual tools below didn't change for a period of time. So if I was on the "Library" tool tab and clicked on "Lens" tool tab, I'd still see the library tools! Eventually the screen would catch up but the effect was awful.

Then I noticed something: when the eGPU behaved badly, it appeared on the bottom of the Activity Monitor "GPU history" graph while the internal Intel Iris Plus 655 appeared above. (In activity monitor, select "window" then "GPU history" or command-4). When it was working perfectly, the eGPU appeared on the top of the "GPU history" graph instead of the bottom.

The difference: it depends what is hooked up when you boot. If you boot away from the eGPU and walk the laptop over to the eGPU then attach it, the problem exists. But if you reboot with the eGPU attached, problem goes away. Seems to be an OS thing to 'prefer' the best GPU attached at boot-time.

I'm running High Sierra 10.13.6 and with Mojave macOS will allow you to select the preferred GPU by app (which I will try when the time comes). But I write this only because it seemed to solve the problem forum member 'hyperlink' was having a couple months ago with the Blackmagic. Overall so far, now that I figured this out, the 13" Macbook Pro with eGPU is stellar. (And it could be better with a more powerful/expensive external GPU card).

[Mark T]

I recently upgraded to the Apple MBP 15" 2018 i9 32gb RAM. Top of the line. I saw almost no performance increase in preview generation - this biggest bottleneck to my workflow. I added a Sonnet eGFX Box with Radeon WX9100 (expensive bugger!) and I got a zero increase in preview generation performance. But I also saw no increase in this spec when upgrading my MPB. The published statement that Capture One Pro 12 relies on RAM to generate previews is hogwash. It barely touches the RAM on my system, let alone the processors.

I did a test and observed that Capture One Pro 12 uses 10% of my available 32GB of RAM and 10% of the processing power of my i9 chip when generating previews.

I just wrote the following e-mail to Capture One Support. I would appreciate it if you all write similar notes to them. The squeaky wheel gets the grease! Here is a link to my screen recording that shows the performance issue: https://youtu.be/8_59jtIl71A

Here is the support e-mail:

I recently upgraded to Capture One Pro 12. I also recently upgraded to the mid 2018 MacBook Pro 15 with i9 processor, 32GB RAM, and top of the line specs. Lastly I added a top of the line eGPU (Sonnet eGFX breakaway box 650 + Radeon WX 9100).

After all of these upgrades, I see a nice boost in exporting files from the raw. My system moves at more than double the speed as my previous MBP without eGPU. However, I have zero increase in speed in importing and rendering a new shoot. I often import 1,000 - 2,000 images per shoot, and the preview generation time is quite slow. I noticed that the GPU is not utilized at all for this. This is surprising to me because rendering previews seems to be the perfect task for a GPU.

I did some research and other users are reporting the same conclusions. I also saw an article that you posted that said RAM is the deciding factor in rendering previews. However, I disagree with this, and I would like a better explanation. I doubled the ram and processing power in my MBP and saw no change in rendering previews. Another user compared his 40GB RAM iMac Pro to his MBP with 16GB RAM and way less power, and saw no difference in preview generation time.

On my system, when I observe the apple activity monitor, with only Capture One running, generating previews, I can tell that the GPU is not used at all, and more shockingly, the CPU is only utilized to 10% of its processing power. Lastly, the RAM, which you write on your published paper, is most useful in generating previews, odd as that sounds to me, is also not utilized at all! I have 32GB of RAM, and C1Pro12 uses 10% of that RAM in generating previews!

I documented this on a short screen video recording. Please view it here: https://youtu.be/8_59jtIl71A

So, first order of business, please comment on this situation and let me know if I am missing something. Please also explain further why you say that RAM is the determining factor in preview generation performance when my tests show that it has no effect to speak of. Please also explain why a number crunching intensive task like preview generation is only using 10% of CPU power, and no GPU power at all.

Thank you for your time and help on this.

- Mark Tomaras

[yorickmeijdam]

Hi Mark,

I can't share any experience but I can say I'm curious for future developments since I'm interested in buying an eGPU (and therefore boosting preview generation and export times).
By the way, nice setup you got there! WX 9100, wow..

Cheers,
Yorick

[NN634969607133502601UL]

Hi Mark,

Silly question, but do you have Capture One running in a space/window on an external monitor? Or running on your MBP's built-in display?

I think I've read somewhere that eGPU's kinda only boost performance on apps that are running on an external display.

I recently upgraded to the Apple MBP 15" 2018 i9 32gb RAM. Top of the line. I saw almost no performance increase in preview generation - this biggest bottleneck to my workflow. I added a Sonnet eGFX Box with Radeon WX9100 (expensive bugger!) and I got a zero increase in preview generation performance. But I also saw no increase in this spec when upgrading my MPB. The published statement that Capture One Pro 12 relies on RAM to generate previews is hogwash. It barely touches the RAM on my system, let alone the processors.

I did a test and observed that Capture One Pro 12 uses 10% of my available 32GB of RAM and 10% of the processing power of my i9 chip when generating previews.

I just wrote the following e-mail to Capture One Support. I would appreciate it if you all write similar notes to them. The squeaky wheel gets the grease! Here is a link to my screen recording that shows the performance issue: https://youtu.be/8_59jtIl71A

Here is the support e-mail:

I recently upgraded to Capture One Pro 12. I also recently upgraded to the mid 2018 MacBook Pro 15 with i9 processor, 32GB RAM, and top of the line specs. Lastly I added a top of the line eGPU (Sonnet eGFX breakaway box 650 + Radeon WX 9100).

After all of these upgrades, I see a nice boost in exporting files from the raw. My system moves at more than double the speed as my previous MBP without eGPU. However, I have zero increase in speed in importing and rendering a new shoot. I often import 1,000 - 2,000 images per shoot, and the preview generation time is quite slow. I noticed that the GPU is not utilized at all for this. This is surprising to me because rendering previews seems to be the perfect task for a GPU.

I did some research and other users are reporting the same conclusions. I also saw an article that you posted that said RAM is the deciding factor in rendering previews. However, I disagree with this, and I would like a better explanation. I doubled the ram and processing power in my MBP and saw no change in rendering previews. Another user compared his 40GB RAM iMac Pro to his MBP with 16GB RAM and way less power, and saw no difference in preview generation time.

On my system, when I observe the apple activity monitor, with only Capture One running, generating previews, I can tell that the GPU is not used at all, and more shockingly, the CPU is only utilized to 10% of its processing power. Lastly, the RAM, which you write on your published paper, is most useful in generating previews, odd as that sounds to me, is also not utilized at all! I have 32GB of RAM, and C1Pro12 uses 10% of that RAM in generating previews!

I documented this on a short screen video recording. Please view it here: https://youtu.be/8_59jtIl71A

So, first order of business, please comment on this situation and let me know if I am missing something. Please also explain further why you say that RAM is the determining factor in preview generation performance when my tests show that it has no effect to speak of. Please also explain why a number crunching intensive task like preview generation is only using 10% of CPU power, and no GPU power at all.

Thank you for your time and help on this.

- Mark Tomaras

[yorickmeijdam]

eGPUs should also boost performance with only an internal display, although it's not an ideal situation since the bandwidth of T3 is being used going back and forth. Yet, it should boost somewhat the performance.
You can see on the video the graphics card is not being addressed by CO12.

[hyperlinked]

I think I've read somewhere that eGPU's kinda only boost performance on apps that are running on an external display.

They can definitely boost performance both directions, but as yorickmeijdam mentioned, they're not as efficient as an onboard GPU.

Support of external monitors is an especially hot topic for people with Macbook Pros though. It's because if you have an LG 5K Ultrafine monitor, the only eGPU you can use to accelerate the LG 5K is one of the pricey BlackMagic eGPUs. None of the other eGPUs offer Thunderbolt 3 support that can power a display. Those TB3 ports will only work for peripherals.

[Peter Svancar]

hello guys, is there any update on this topic with Mojave, which supports internal display? There is as well a ticbox in app properties to prefer external gpu....so maybe checking this box will help..
I have a dilemma now, since I have on hand RX580 gigabyte gaming box for good price, but I need to connect it via TB2 port...it is possible but wondering if anybody did try this so he can report back results...
I have imac 5k with R395 now

[Wesley]

hello guys, is there any update on this topic with Mojave, which supports internal display? There is as well a ticbox in app properties to prefer external gpu....so maybe checking this box will help..
I have a dilemma now, since I have on hand RX580 gigabyte gaming box for good price, but I need to connect it via TB2 port...it is possible but wondering if anybody did try this so he can report back results...
I have imac 5k with R395 now

I had the same eGPU connected to my TB1 - 2011 15" MBP + external display last year. You just need to get Apple branded TB3 to TB1/2 adapter.

I'm not using that setup anymore and haven't been keeping up with MacOS & eGPU support so I can't help you about internal display performance and such.

Going to https://egpu.io will be more useful with that.

[JCFP]

Hi all, did a little testing on my 2017 15" i7 MBP running the latest version of C1 (for Mojave, at least), both with and without a Sonnet/Radeon RX 580 eGPU. Not entirely sure what the data is telling here, so would love input. The test was set up to mimic conditions found while teching a studio job, so I tethered my 5DIV 30 megapixel camera and shot on Slow Motor Drive for 60 seconds, then wrote down how many images were imported at 60 seconds, how many total came in, what time C1 stopped doing it's quick preview loading, then how long it took to scroll through all images, waiting for the full screen (or nearly full screen, showing Dock and C1 Tools and Program Window) previews to load before advancing. Not rigorously scientific, I know, but it's something...

Hardware Acceleration (Open CL) was set to Auto for both Display and Processing.

I shot a "scene" that had a mix of light and dark, moving my camera about but overall, the images for every group all had the same amount of info.

Here's a spreadsheet showing the results:
https://www.dropbox.com/s/j2i7y8pojwt1q ... e.jpg?dl=0



What strikes me is there's no improvement on the best results using the eGPU - the highest amt of images imported from a 60 second capture came from not using it. The biggest takeaway, though, has to be the capture vs preview performance when mirroring your displays vs having your biggest be your primary window, your second monitor be the viewer window, and the laptop looking at Finder windows. You can capture far more images in 60 seconds when mirroring, but it takes C1 longer to ingest and then generate previews from them vs having all monitors show something different.

I was really hoping to see an improvement here when it came to previews, but it looks like I'd get about the same, if not better, results just foregoing the eGPU and using the screens for different purposes. Luckily, that's how I prefer to work. Given the option, if bringing the eGPU wasn't a big hassle, I will do it, as the benefit in exporting times will hopefully give me more time to check images as they come in, or allow me to capture and export simultaneously without the program hanging up.

It has to be said that these tests aren't precise, and that there's a certain amount of variability in these numbers.

Would love to hear what you all think, and have seen with your own setups.