Capture One Pro Performance w/Hardware Acceleration
To Utilize GPU or not Utilize GPU... that is the question!
I'm currently running 2012 Macbook Pro 15" and am looking to replace it.
In shopping for a new MacBook Pro, it is somewhat unclear "what is really important" for Capture One Pro (C1p) performance. Of course there are a lot of variables (disk/SSD speed, RAM, processor, GPU, cores, size of raw files, etc) and one could get mired in the permutations. The goal here is a bit more limited and attainable: to test a few machines and see what we get. I personally prefer the smaller form factor of the 13" MacBooks 90% of the time, but the 15" models offer discreet GPU 'oomph', more cores and, duh, a bigger screen.
Given that I use Capture One Pro to catalog my images, shoot tethered, and edit images, my goal is to find the best fit for my use cases and if possible stay at 13". I also sometimes execute large batch export jobs with hundreds of images, so I need efficient process recipe (batch) performance.
I use C1p Sessions when I shoot a project then once complete, I move that session to my iMac (with external RAID and all that) and import it into my 'master catalog' for future reference. My goal is to stick with the MacBook and dock it to a high-end monitor (or two) for of editing. Because of changes in technology (e.g., Thunderbolt 3), when docked I now have the option of plugging in an external GPU (eGPU) box to get the graphics performance of the best desktops. This is a bit new in the Mac world and I want to test that out.
This also creates and interesting option as I can now add an eGPU to a smaller MacBook and get desktop like performance. With that in mind, I quickly zeroed in on two options that are roughly within the same price range:
13" Macbook Pro with an external GPU
15" MacBook Pro (no additional eGPU)
The 2018 13" configuration in play (16GB, 1TB, 2.7Ghz/4.3Ghz) comes in at $2,899. The 15" high end (16GB, 1TB, 2.9Ghz/4.3Ghz) comes in at $3,899. Throwing in an eGPU, the 13" will cost me at $3,598 (with Blackmagic eGPU) or $3,448 (with OWC Bundle: OWC Helios FX + XFX AMD Radeon RX 580 GTS XXX Edition GPU bundle ); so with an eGPU, the 13" is still between $300 and $450 cheaper than the 15" .... more on that later.
The real question for me is not about specs, but performance using Capture One Pro. When does the GPU help? Is it relevant for my use case? And how does the 13" compare to , say, my iMac or my old 2012 15"?
Obviously there are compromises, but like all things computing, the specs don't always get reflected in real world performance perks and you can spend a boatload for a "better spec" that gains you next to nothing; I'm trying to avoid that so I started by seeing what PhaseOne has to say.
What does PhaseOne say about all this?
PhaseOne published an article on their website (quoted to the right).
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
... it's not a lot of detail to go on, but it is what it is.
Let's look at each of these four areas.
Testing Preview update and Process time
So in the end, I tested 5 machines for preview building and batch processing. I really didn't come up with a measurable way to test "Sorting/Rating" or "fit image to screen". Not that you can't notice improvements in those areas, I just wasn't sure how to measure it. While I could not get my hands on the 15" 2018 i9, I did test a 2017 i7 15" and an iMac which has a better GPU than the latest 15". I also did get some performance scores from a 2018 15" i9 owner which I mention in my notes later (I'm happy to update this article if I can get my hands on a 2018 i9).
So the machines in play are as follows:
1) 27" iMac 5k 2017: 40GB RAM, 4.2/4.5 Ghz i7, 4-core, Radeon Pro 580, 8GB
2) 15" MacBook Pro 2012: 16GB RAM, 2.7/3.7 Ghz 4-core i7, Intel HD Graphics 4000
3) 15" MacBook Pro 2017: 16GB RAM, 2.8/3.8 Ghz 4-core, Intel HD Graphics 630 1536 MB
4) 13" Macbook Pro 2018: 16GB RAM, 2.7/4.5 Ghz 4-core, Iris Plus Graphics 655 1536MB
5) 13" Macbook Pro 2018 + Mercury Helios FX eGPU: and XFX Radeon RX 580 GTS XXX Edition
Note: 4) and 5) are the exact same MacBook, only difference being the eGPU. For each test, I used the same exact C1p Session with 500 Sony A7Riii raw images. For the "export' test, I sent the resulting images to the same internal SSD drive at 2400px JPGs (long side), sRGB. Note that the 2012 MacBook has slower SATA SSD tech, the others had more modern PCI-E SSD drives.
Again, a primary interest is to see how a 13" MacBook performs docked with an eGPU. The idea being that on the road I don't need the GPU performance so much while using the small screen, but back in the studio when processing a large batch of images and using a large monitor or two, I'd like "desktop performance". The eGPU should theoretically take care of that nicely. Obviously, you can break the bank and purchase $1,000 - $1,500+ GPUs, but that's not happening in this case, I'm not a gamer and I don't render a lot of video; my goal is to get solid (not world record breaking) performance.
Why the Mercury Helios? Ok, so the Apple Blackmagic eGPU with Radeon Pro 580 card costs $699 and is NOT upgrade-able: you can't swap out the graphics card for a more powerful one later on. The Macsales.com (OWC) Helios FX eGPU can come as a kit with a similar card (the Radeon Rx 580) costs $150 less..... and it is upgrade-able, meaning you can swap the card in the future for a more powerful one, using the same chassis.
I should clarify: the two options do not utilize the exact same card, the Blackmagic has the Radeon 'Pro 580' with two Thunderbolt 3 ports, four USB 3 ports , one HDMI 2.0 port. The Helios has the Radeobn 'RX 580' version with one HDMI, three Display port and a single Dual-Link DVI. I actually wanted to test the Dual Link DVI so I could dust off my old 30" Cinema Display in the basement just for grins. The Blackmagic's USB ports didn't appeal to me as much as the Displayports in the Helios as well. A quick Google search shows they perform about identically. Helios it is.
TEST 1: Export results
The chart below shows each machine with "no GPU" (hardware acceleration off in Capture One preferences) and "w/GPU" (hardware acceleration turned on). Longer bars indicate better performance as I measured images rendered per minute.
In this case I exported five hundred 2400px (long edge) JPGs at 95% quality for each machine. The destination was always the Mac's internal SSD.
Note about the 13" options
This is key, the 13" with eGPU triples the export performance from 22.2 images per minute to 76.2.
PhaseOne indicates that GPU helps export performance and it rings true with one exception. The iMac saw a 4x improvement in export performance with hardware acceleration and the 13" Macbook Pro 2018 with eGPU wasn't far behind. And this makes sense since export processing leverages the GPU and both machines have similar 8GB Radeon 580 class GPUs.
Surprisingly perhaps, the old 2012 15" i7 actually suffered a bit with hardware acceleration, churning out an anemic 8.8 images per minute with it enabled (and 10.6 without).
About the 2018 15"MacBook Pro i9: If I'd gotten my hands on one, I would expect it to land around 66 or 75 images/min with hardware acceleration, but that's just a guess. In one forum, a user with that exact machine reported 45 images/min when exporting slightly larger (50mp) files. I didn't include that result since I didn't run it myself using the same setup, but it kind of matches expectations. Certainly it should fall somewhere in the 50 to 80 images/min range, similar to the 13" with eGPU (and the iMac).
TEST 2: Preview Generation
In this test I took the same 500 image Session, same A7r3 files, and regenerated all image previews at 2048px size. I didn't expect hardware acceleration to help here and it clearly didn't.
Hardware Acceleration clearly plays little to no role in preview generation performance. In this case, each computer regenerated 500 previews with and without GPU hardware acceleration and those numbers were normalized to "images per minute"; you can they are pretty consistent . As referenced above, PhaseOne mentions that RAM (not GPU, not CPU) is important in Preview Generation and the results support this.
What's interesting is that the iMac, with 40GB of RAM, didn't perform any better than the 13" with 16GB of ram. Perhaps RAM only matters to a point.
Overall: OPEN CL Hardware Acceleration doesn't help C1 "Preview" generation, but it helps a lot when using process recipes to export images.
Powerful discreet GPUs help a lot with C1P image exporting: The iMac has a decent GPU (Radeon Pro 580 8192 MB graphics) and with GPU processing turned on, it exported 4x as many images per minute (83 images/min using OPEN CL vs 20 images/min without). The 13" MacBook Pro, with external GPU, performed almost as well as the top-of-line 2017 iMac.
The newer integrated GPUs aren't half bad: The 2017 MacBook Pro showed more than double the export performance with it's non-discreet GPU enabled. The 2018 13" MacBook shows only a slightly better performance with acceleration on, but it was better nonetheless. The 2018 15" models with a discreet GPU and 4GB memory will probably fall in the 65+ images/min export range (see note above).
Older machines: The 2012 MacBook Pro ran SLOWER with hardware acceleration (8.8 images exported per minute with GPU acceleration on vs 10.6 images/minute with GPU off). If you are running old hardware, you may actually want to turn OPEN CL off or at least test it.
It's all about compromise at some point. In preferring the smaller form factor of the 13" model, the eGPU gives me a great "docked option" that didn't exist a few years back. And thus the 13" with eGPU hits a sweet spot (portability and performance) for my needs... and I keep the ability to upgrade to a more powerful GPU card at any time... a cheaper upgrade than a new MacBook. Obviously, you can also use an eGPU with the 15" model as well, but I believe you'd have to spend a bunch more on a faster GPU card to make it worth it as the internal GPU is quite good and the one I used in the Helios chassis wouldn't add much to the equation. So a 15" plus a better external GPU card would be a much more expensive play.
* * *