Performance goals for Capture One Pro
Here's a short video I made comparing the performance of making a basic exposure adjustment to the same 5D Mark III RAW file in both Apple Photos (beta) and Capture One Pro 8...
http://youtu.be/AV2EyxXbK84
My obsevations:
- In Apple Photos, response is smooth and instantaneous. I can't move the slider fast enough to introduce any lag.
- In Capture One Pro, response is slow... the slider lags behind the mouse and the image behind the slider, and the histogram behind the image.
I hope the developers can talk to someone at Apple about how to get this kind of performance in Capture One Pro. It apparently could be so much better.
http://youtu.be/AV2EyxXbK84
My obsevations:
- In Apple Photos, response is smooth and instantaneous. I can't move the slider fast enough to introduce any lag.
- In Capture One Pro, response is slow... the slider lags behind the mouse and the image behind the slider, and the histogram behind the image.
I hope the developers can talk to someone at Apple about how to get this kind of performance in Capture One Pro. It apparently could be so much better.
0
-
Interesting video! I wonder if C1 is constantly regenerating new previews from the original as the sliders are being manipulated or are we seeing incremental changes being applied to the last known state of the preview?
I also wonder how a basic adjustment in both platforms would compare against each other. For example, how well does exposure compensation in C1 compare to Apple Photos? Does one maintain white balance and color rendering better than the other?
I think pros are less likely to be throwing sliders back and forth in C1 than everyday shooters with an iPhone and Apple Photos. 😂 Pros are probably more likely to adjust incrementally or simply jump to well known values. Accurate rendering beats out speedy response ... but I do see your point. What did Apple figure out that C1 could have done? The Apple s/w team has nearly unlimited access to everything Apple so they may be leveraging internal trade secret knowledge.
I have yet to play with Apple Photos but from what I have seen it's not for me. Hence, I am here. 😉0 -
Well, a few pointers:
- We have some performance-issues with 4k-5k resolutions under certain situations, those will be fixed in the next service release.
- You are having 2 histograms and the navigator open. That will affect performance. (leave the upper most one, so the comparison is more equal).
- Is your OpenCL being used?0 -
Just as a note, I can't really speak of Photos app that I haven't tested myself, but on my MBPRO 15" Retina (Late 2013) 2.3Ghz/16Gb RAM/NVIDIA GeForce GT 750M 2048 MB sliders behave somewhat smoother than shown on video, although not as smooth as Photos app but there is a a noticeable lag. With one histogram open, browser is closed.
Lightroom 5.7 is also not as smooth but there is no noticeable lag.
Keep up with great work and make are awesome software even better!
Marko0 -
Ahhh, GPU acceleration! I didn't think of that. Yes, Apple is using all GPUs available in all model Macs. Capture One probably uses only select models? For example I don't think Capture One supports the AMD Radeon HD 6770M in my iMac. I would probably have jagged performance. 0 -
If it has above 1 gb of VRam, and the GPU capable enough, we support it, and also multiple GPU's, if available.
If you want to know in a visual way, enable the focus mask on a reset image, adjust contrast or exposure. If the mask disappears while adjusting, OpenCL is on. If not, CPU is used.0 -
I get what looks like the same performance using C1 on my Win 7 i7Intel notebook from 2012/2013. That was using a NEF from a D800 - the first big file I found when I went looking for one.
15" 1920px screen and no GPU assistance because the internal facilities are beaten by the CPU.
To be honest I don't find the response of C1 any sort of problem for normal work on this machine.
I have no idea how Apple have approached Photos development but a few years ago I compared the perceived "slow" performance and the convertor I was using at the time (not C1) with the major player in the market at the time (not Aperture of course being I was using a Windows machine.)
The major player looked more responsive on screen but, oddly, the total elasped time to edit and then commit the changes (they both worked a little differently to C1 in that respect) was longer in most case. Not by much but onscreen observations would have suggested it would have been finished, showered and gone before the competition had made it off the field.
Looking closely it seemed that what it was doing to appear fast meant that mostly it was editing using the embedded jpg as a source - if not that then a very 'lite' file sitting in the cache somewhere.
As you went through an edit step if you looked closely you could discern the image presented was still being processed in the background. That could take some time. However by working that way the tasks were more or less invisible to the user until committing the changes when they had to run the full process.
The application I was using, like C1, was doing a full bottom up recalculation after each "detected" change. Sliders can be challenging for that depending on how frequently the current position is sampled as we move it.
I have no idea how Photos works but it is quite possible that it is attempting less and seeking to offer a good onscreen experience as a primary objective in order to impress. There is no harm in that and offering an impressive GUI experience is clearly advantageous but how meaningful it is to the workflow on an individual basis may vary from one user to another.
My observations for what they are worth.
Grant0 -
This is like comparing the Tesla with a '66 Mustang. Both beasts in their own right...
But Photos is a complete re-write. It's a very light application and Apple are utilizing new internal Frameworks that are quite literally blazing fast.
http://sixcolors.com/post/2015/02/new-a ... framework/
I'm not saying we shouldn't push for better performance in Capture One, but it's just a touch unfair to compare against Photos. Unless of course your intentions were to show a proof of concept on how fast things really should in the future.0 -
My best guess would be, that Apple is using all kinds of Core Image functions exclusively, which are sort of OS level "preset system calls" for standard graphic manipulation, which again speed up image processing tremendously. It could be that Phase One's code base is more platform neutral or proprietary because of multiplatform-compatibility issues.
I'm just guessing here, maybe this is not the case and the difference in performance is due to some other things going on in the background. I remember the days when Aperture was criticised for being a lot slower than the first versions of Lightroom and when people digged into this a little deeper, it turned out that LR did never show you the end results on screen, just some sort of down-sampled "preview", while Aperture always rendered the end result practically in real time while you changed the sliders.
Maybe you've seen the latest preview video of a new Photoshop contender called Affinity, which is also a lot faster than Photoshop by relying completely on Core Image.
I would assume you'll see this kind of performance gap also when comparing to the latest version of Aperture.0 -
As a follow up... let me say a few things...
1. I'm happy to hear that some work is being done on performance on 4K and 5K displays. Sadly, I do find C1 works a lot smoother on my laptop with 27" Apple Display than it does on my much more powerful 2013 Mac Pro with 4K display (used in the video). My Mac Pro is the new trash can model with 3.5GHz 6-Core processor, 32GB RAM, Dual D500 GPUs, and 1TB SSD (which is home to my C1 catalog), driving a Dell P2715Q IPS Display. This is very good hardware.
2. I have OpenCL on auto... but as I've said numerous times before, I see no difference in the performance of any task in C1 with OpenCL on or off.
3. My intention here was not suggest that C1 should perform like Photos, but to demonstrate what kind of performance is possible.
4. It's difficult to get a feel for the difference from the video... You really need to try it for yourself. Of course as photographers, we don't spend our session swinging sliders back and forth but it illustrates the differences. Photos is like a speed boat and makes C1 feel like a oil tanker in comparison.
5. The video only shows a single adjustment being made to an otherwise unadjusted image. I'll try to do a video showing what happens when you've made a bunch of adjustments, along with a couple of local layers with their own adjustments. The other evening, I was trying to adjust saturation using the color editor on a local mask of the sky and it was taking a few second to refresh the viewer... Horrible.
6. Apple maybe using different frameworks or coding techniques, but the C1 development team would be remiss for not investigating if those same frameworks can be used for C1 to improve it's performance.
7. My name is probably mud within the walls of Phase One because I'm raising a ton of issues. In the last two weeks, I've logged at least 10 tickets with support. I love the app, but that's no excuse to have to deal with frustrating issues. And the issues I'm logging are not frivolous. I can't trust what I see in the viewer half the time and the performance is really slowing me down. I want the same thing everyone else wants... for Capture One Pro to be the best RAW converter and editor on the planet. I hope my drawing attention to issues is helping.0 -
[quote="VirtualRain" wrote:
6. Apple maybe using different frameworks or coding techniques, but the C1 development team would be remiss for not investigating if those same frameworks can be used for C1 to improve it's performance.
The framework used in Photos is called UXKit. It's based on UIKit and is implement atop of AppKit. It's a private framework and likely won't see the light of day for external developers (read more here: http://inessential.com/2015/02/05/uxkit_skepticism).
This isn't to say there's no ways to improve performance of the app. I've suggested a few ways privately and the response wasn't technical enough to lead me to believe they actually took my advice on board. Regardless I think Phase One are taking our criticisms seriously and we should see some headway in the next release and thereafter.0 -
Chris, you may not have missed it, but just in case: - You are having 2 histograms and the navigator open. That will affect performance. (leave the upper most one, so the comparison is more equal).
Sounds like that change may have the biggest impact on performance.0 -
I ran this same little experiment on my humble mid-2011 iMac. Following Christian Gruner's idea, I turned on my focus mask and it stayed on as I was adjusting the image. Capture One is using the CPU only here.
If I collapse all the histogram displays then I can get decently smooth performance. It's not as smooth as the Apple Photos vide but it's pretty close. If I expand the histogram tool then there is noticeable lag. In this case it's not as jagged as what was in the video but the word I would use again is noticeable.
The image came from a Pentax K-3.0 -
I did some further testing of the viewer performance incorporating Christian Gruner's suggestion to have just a single histogram (so I removed the levels and focus adjustment tools).
I have also confirmed that OpenCL is active as making an adjustment to a reset image with the focus mask will make the mask disappear while adjusting. However, I don't know if this indicates anything, but the mask does NOT disappear when making an adjustment to an image which already has adjustments.
Here are two videos I've prepared which show the issues I am seeing with performance...
1. Capture One Pro 8 Viewer Update Performance (making adjustments)
http://youtu.be/8OUe_49GwVA
The first demo is on an image with no adjustments. You'll see that the screen refresh is a bit choppy but it keeps up reasonably well.
The second demo is the same image with several adjustments including five different local adjustment layers. You'll notice that the performance of updating the viewer is incredibly laggy. In fact, making even a slight adjustment to exposure requires a wait of 10 seconds or more to see the result rendered in the viewer and the spinning wheel to stop.
2. Capture One Pro 8 Viewer Update Performance (browsing)
http://youtu.be/OY82BfqHpnc
This video demonstrates the time to load adjusted images when selecting them in the browser. The delay can be anywhere from a few seconds to 10-15 seconds for images with many adjustments (including local adjustment layers).
This is done on a 2013 Mac Pro, 3.5GHz 6-Core, 32GB RAM, Dual D500 GPUs, and 1TB SSD driving a 4K 3840x2160 Dell Display.
I have submitted this to Support. Case # 1730040 -
What size do you have previews set to? 0 -
[quote="meanwhile" wrote:
What size do you have previews set to?
Previews are set to 2880px. And my workspace is configured such that the viewer is slightly smaller than 2880px. There's a couple of strange issues if the viewer window size is larger than your preview images.0 -
Have you tried setting them to 3840? There's a couple of strange issues if the viewer window size is larger than your preview images
Did you mean smaller than? You might just need to wait until Phase fixes the 4K issues.0 -
[quote="meanwhile" wrote:
Have you tried setting them to 3840?There's a couple of strange issues if the viewer window size is larger than your preview images
Did you mean smaller than? You might just need to wait until Phase fixes the 4K issues.
I'm waiting patiently. And I do run my displays at native resolution (3840x2160). There are issues I've witnessed when the viewer window exceeds the size of your preview images or if it's much smaller than your preview images. The former can easily happen on a 4K display... Even with 2880px previews, you can have a viewer that's 3000px or more and even bigger on a 5K iMac. The latter can happen on any screen if you size your workspace smaller than your previews. So far I've found it's best to setup your workspace so your viewer is just slightly smaller than your preview size.
I wonder if dynamic previews would make more sense... When you first load an image in the viewer, it creates a preview of the exact size of the viewer. As long as you don't resize things on your workspace, that preview would be ideal. If you do resize your workspace, then a new preview would have to be created at the new size. For people that work with a consistent workspace layout, this would probably provide the most accurate view without scaling. I don't know... But the current solution is not working well. If you can't trust what you're seeimg in the viewer then you're really screwed.0 -
Sure the Previews are dynamic in so far as they are filed for quick read at a "preferred" resolution but in use are resized in line with current settings. Moreover they are dynamic in that will be automatically tweaked to reflect, inter alia, the currently selected (or just set) Process recipe.
The performance problem people experience is, I suspect, often a comparative issue because C1 makes these recalculations in real time. If we choose to display an image larger than the pre-saved preview a whole new interpretation is calculated, not just a quick upsizing of the preview. It the same for downsizing although binning pixels is likely to be slightly faster than calculating for new ones, especially if an image has been cropped and browsing to it results in upscaling rather than just recalculating within its current pixel dimensions.
The rewards of having plenty of compute power are high. In fact vital.
Grant0 -
[quote="SFA" wrote:
Sure the Previews are dynamic in so far as they are filed for quick read at a "preferred" resolution but in use are resized in line with current settings. Moreover they are dynamic in that will be automatically tweaked to reflect, inter alia, the currently selected (or just set) Process recipe.
The performance problem people experience is, I suspect, often a comparative issue because C1 makes these recalculations in real time. If we choose to display an image larger than the pre-saved preview a whole new interpretation is calculated, not just a quick upsizing of the preview. It the same for downsizing although binning pixels is likely to be slightly faster than calculating for new ones, especially if an image has been cropped and browsing to it results in upscaling rather than just recalculating within its current pixel dimensions.
The rewards of having plenty of compute power are high. In fact vital.
Grant
Whatever C1 is doing is not very efficient... and prone to issues that mean you can't trust what you're seeing in the viewer... The first is a pain in the ass, the second is a showstopper. As the comparison to Apple's own apps (Aperture and Photos) shows in the first post, it seems Phase One could learn a lot from how Apple is handling updates to images during editing.0 -
Are there any news on this Topic?
I'm going to buy a 5k iMac soon, and want to be sure that C1 performs good on it.0
Post is closed for comments.
Comments
20 comments