Skip to main content

Improve C1P's exporting performance as it does not use all CPU powers.

Logged

Comments

45 comments

  • Official comment
    Bjørn
    Product Manager

    Hi Benjamin

    Thanks for your suggestion for improvements and the details you have provided. Improving hardware utilisation is something we are exploring. 

    Bjørn, Product Manager, CO

  • David Fay

    While I enjoy still being able to use my computer while c1 exports, the options to max out the processor would be useful when setting a large export and leaving the computer (as I often do).

    1
  • BeO
    Top Commenter

    My experience with performance of any multistep process is that there is always one step which is a bottleneck. In the case of C1 on your computers it might be that the step which the bottleneck for the specific C1 processing is not measured by the graphs, e.g. loading data into an internal processor cache, or something alike.

    -1
  • Thomas D.

    Hi,

    It would be great if you could parallelize everything.

    A single GPU can easily manage two to three images simultaneously.

    A style can also be copied in parallel to several photos at the same time, the same applies to the creation of masks (luma masks).

    Images can also be created in parallel side by side during import, use the GPU for import.

    Capture One is a powerful software, but it is designed so that you usually only deal with one photo a time.

    Greets Tx

    -1
  • FirstName LastName

    This is one of my biggest let-downs in Capture One.  It's speed in browsing pre-processed images is pretty good.  It's speed in processing is not nearly as good as it could be considering poor to no multi-core utilization.  As a software developer myself, I can't think of a good reason why this should be a technical challenge.  It's a data-parallel type of parallelism, which is the easiest kind of parallel processing problem.  Multiple images can go through multiple corresponding threads running different instances of the exact same processing code.  In cases where the GPU is involved, that adds some additional work, but still.  I'm not saying it's something one person can do in a few days, but I am saying it's a "just do it."

    1
  • Peter Svancar

    Perhaps i am mistaken…i believe that capture one is using GPU for exporting images and all processing. Therefore the CPU cores are not under load…so basically you can use computer while exporting..perhaps you can force C1 to use CPU cores for export by disabling GPU acceleration for processing…in the C1 menu

    0
  • Benjamin Kim

    Peter Savancar/ LR CC also use GPU for exporting but the big difference is LR CC use almost CPU cores. My testing already proves that C1P is way slower than LR CC by 30 min vs 18 min for same files. That's almost 2 twice slower which is not acceptable. I had a job for high volume production works and exporting speed was bothering even if I export them per folder. Yes, I export them whenever we finish with each product and move on which takes only 2 min to shoot and I have to process them.

    0
  • Peter Svancar

    LR is using primary CPU for exports..if i am not mistaken, it was atleast in the times when they started use HW acceleration. It would be interesting if the test can be repeated with C1 gpu acceleration off

    0
  • Benjamin Kim

    Peter Savancar/ I already tested without GPU and it took more than 1 hour. The CPU uses was still low compared to LR CC.

    0
  • Thomas D.

    Peter, did you use LR with its new denoising on?

    This will slow down the process by 8x.

    ___

    To me LR is like using a software that is processing images with 8 bit precision.

    With Capture One, to me, its like processing with 32 bit precision.

    Also using layers a lot does slow down a lot because every layer acts as a own image.

    0
  • Peter Svancar

    I do not have LR so i cant compare

    0
  • Martin Knowles

    Definitely running into this on 16.2.2.1 on an M1 Max MacBook Pro (as Hiarch reported above). In my testing, it's possible to max out either the set of GPU cores on export, or the CPU cores on export if you have "Use hardware GPU" set to Off rather than Auto. On an M1 Max, at least (and probably true for anything with more GPU cores), using GPU is the faster thing to do...but it still leaves you with your CPU cores sitting mostly idle at ~20% while your GPU usage pegs at about 75%.

    For big batch processes (exporting, in particular; preview generation on import is another one) that can be pipelined to either the GPU or the CPU, it would be great if C1 did both provided you have enough RAM. On my M1 Max, exporting about 100 ~28MP Sony ARW's only pegs ~32GB RAM tops. 

    Relatedly, applying LCCs causes GPU acceleration to not work at all, which leads to somewhat annoyingly slow export behaviour if you stack LCC'd images--export a dozen Phase IQ4150 images, and your CPU utilization across all cores sits at 75% while your GPU sits largely idle. This is one case (creating masks from Color Editor selections is another) where optimizations would lead to big workflow speed improvements.

    0
  • Francis Principe-Gillespie

    Yes please, if possible, allow for a 'use all computer resources' feature where Capture one will use as many resources as possible. I do lots of microscopy so export for stacking 300-500 images per batch to create a larger gigapixel mosaic. This slows to a crawl and yet I can do loads of other tasks on my machine at the same time . Lettting Capture One use all system resources eg multiple processors with multiple cores and multiple threads plus gpu. This would be fantastic though perhaps a dream wish list item.

    2
  • Benjamin Kim

    It has been more than 8 months. How come you are not fixing this?

    1
  • Chelaru Ionut

    Probably next year we will see something.

    0
  • BeO
    Top Commenter

    It has been more than 8 months. How come you are not fixing this?

    Probably because they had other priorities, e.g. AI masking?

    Which I find way more interesting than catching up with export performance of LR. And speaking purely about performance, 0.9 sec per image vs. 0.54 sec per image on export, with AI masking one can save more time than this difference.

    I understand though that you seem to have other priorities.

     

    0
  • BeO
    Top Commenter

    Yes, I export them whenever we finish with each product and move on which takes only 2 min to shoot and I have to process them.

    As long as you don't have a better performing C1, if you don't need to deliver the highest quality immediately, consider to export them with a JPEG QuickProof recipe, which exports the C1 preview very quickly, and deliver the high quality exports later. If that is a possible workflow for your job.

    0
  • Benjamin Kim

    I have many different requests for each export recipe and that won't gonna work that way. Clients give very specific recipes so it's not me to make decisions. They know this well. Beside, I have to export each image with 2~3 different recipe such as JPEG in different sizes, TIFF, and more. Also, if you dont export them right after you finish each folder, you dont have time to export them and give all files to photographers and clients right away after we finish. Each folder takes 10~20 min to export images and I have 20~50 folders per session. 

     

    Perhaps you are not working as a digitech so probably won't gonna know how it works. This is why C1P really need better optimization with CPU uses cause LR is literally way faster.

    0
  • Chelaru Ionut

    Given how it works some step/steps in the rendering of the image goes from gpu to cpu and its single thread. C1 uses all the resources but somewhere in there its moves data from gpu to cpu then back to gpu and that is why we see so low utilisation. 

    0
  • BeO
    Top Commenter

    Hi David,

    I understand, different cameras, different output formats, all determined by others not you, and files need to be delivered soon after the last capture.

    And I don't argue that performance is of utmost importance for you, for the above reasons.

    That out of the way, I am interested in your numbers. 

    Each folder takes 10~20 min to export images and I have 20~50 folders per session. 

    That is between 3 and 16.6 hours, 8.75 hours average export time. If shot with Nikon D850, given your latest comparison charts, it is still almost 4.5 hours on average with LR. So you would have to export folder by folder in LR too, and if LR has not changed its export facility you cannot export several file formats in parallel. Does this reflect your situation?

    Hi Chelaru,

    I don't know how exactly it works but indeed your are right, when more than one step or more than one resource is used (e.g. data read (storage), data transfer to the computing units (e.g. bus), computing time (CPU/GPU), writing the results back etc.) then there is always one bottleneck, especially if the process is run on a multi-purpose machine or it there are variations in input (e.g. file size) or computation details (like layers or specific tools in the case of C1 image edits). The job is to find and widen the bottleneck so that the other steps/parts have to wait less for this weak link. To complicate further, this might possibly be different things on Intel / Windows systems than on Macs, and might also vary for each of CPU vs. GPU variation. 

    rendering of the image goes from gpu to cpu

    On the Apple chips, memory is shared between cpu and gpu, so this should give an advantage, theoretically, over Intel chips.

    0
  • Benjamin Kim

    Hi David,

    I understand, different cameras, different output formats, all determined by others not you, and files need to be delivered soon after the last capture.

    And I don't argue that performance is of utmost importance for you, for the above reasons.

    That out of the way, I am interested in your numbers. 

    Each folder takes 10~20 min to export images and I have 20~50 folders per session. 

    That is between 3 and 16.6 hours, 8.75 hours average export time. If shot with Nikon D850, given your latest comparison charts, it is still almost 4.5 hours on average with LR. So you would have to export folder by folder in LR too, and if LR has not changed its export facility you cannot export several file formats in parallel. Does this reflect your situation?

     

    I have no idea what you are talking about. Clearly, you dont know how exporting workflow works. Each session has different amount of folder and files so therefore, it's totally different and I dont see the relationship between C1P and LR in terms of performance. Btw LR Classic can also export in several file formats. 

    Doesn't really change how fast LR is which proves that C1P is poorly optimized despite being expensive. 

    0
  • BeO
    Top Commenter

    Clearly, you dont know how exporting workflow works.

    You can believe what you want.

    0
  • Benjamin Kim

    Beo/ That's what I wanted to say.

    -1
  • BeO
    Top Commenter

    Clearly, I don't know how your export workflow works, as otherwise I would understand how it helped you if C1 would reach LR performance, on avgerage 15 min. per folder x 35 folders per session (as per your numbers) = 8.75 hours cannot be done at end of your day as no one wants to wait so long for your output, so you have to export during the day, but would they wait 4+ hours if C1 would be twice as fast as today? So what would change? You still would need to export during the day. Clearly you don't understand this consideration.

    0
  • Benjamin Kim

    BeO/ It's totally simple and yet you failed to understand it.

    First of all, who really export all files and folders at the end? Whenever you finish EACH folder, we move on and you export them right away while people are prepping for the next look or setup which takes 1~10 min. Since C1P takes 10~20min per folder, it would be way better if they optimized the performance by using all CPU cores just like LR Classic in order to reduce the export time while minimize the stress for the computer while tethering. 

     

    You DONT export them all at the end. You export each folder whenever we finish each setup or look. Which means instead of 4 or 8 hours, you export them by 10~20 min per each so that you dont need to wait after we finish the session.

     

     

    0
  • Raymond Harrison

    David Lee

    You don't have to explain your pain. Everyone gets it. I'm not a digitech and except for knowing it must be very stressful, I don't know what your day is like. Similarly, you won't know what most everyone else's day is like either. I don't think anyone here would disagree with you that Capture One can always use performance improvements, sometimes major. Full stop. Like LrC, too, by the way, and probably in ways you don't encounter in your particular role.

    As this is a feature request forum, requests are not time-bound. It doesn't matter that it was 8 months ago that it was requested (or whatever it was) and I think you feel you're reporting a bug here and that people are arguing against it. People are not. They're just trying to understand your specific scenario so they can add their support to the request. I'm sure YouTubers are interesting, sometimes, but it's your numbers on your machine(s) that are important. It's your difference in export times that are important to the discussion. It's understanding how you're measuring apple-apples that are important.

    As others have indicated, since LrC and C1 are completely different programs, with different code bases and different developers, the bottlenecks may be (will be) in different places. Spending time here trying to tell people you've never met what they do or don't know isn't a winning formula for getting upvotes, to me. I always find that just being fact-based and outcome-centric, without the drama, is a great approach to getting people on board. This request has 16 or so upvotes. It will possibly need more. That said, I also know that Capture One are always working on performance (e.g. the recent improved preview generation) so maybe just discussion on this thread is enough, who knows? Preview generation has been a long-standing bottleneck and maybe with major work done on that, they can focus more on export timings (which themselves have improved a lot over the last couple of years).

    If you feel you have a bug or want to represent this as such, you should file a bug report with as much detauil as you can manage. 

    Note that a 100% CPU or GPU usage regime is great in that when a process needs it, it gets the work cycles, but it isn't a measure of efficiency of use of those cycles or a measure (necessarily) of when a process gets done. I can write a process right now that throttles every core (CPU & GPU) but is highly inefficient :-).

    Improving the efficiency of exporting is a great feature request. Truly. Let the Capture One team figure out the "how". They're the ones who know the code base. No one here does. 

    For what it's worth, LrC is somewhat faster on my machines across a range of export types and camera bodies, and it's down to mostly GPU utilization, other efficiencies and differences in the applications themselves. If there's little to no GPU utilization, but all CPUs are throttled (100%), LrC is appallingly slow. CPU utilization isn't adding much. C1 seems to heavily utilize the efficiency CPU cores but doesn't do much with the performance ones, at least on my machine. If it doesn't really need CPU, then it probably doesn't matter much. No need to use it if you don't need it. I don't have insight into their code base though.  

    By eye on Apple's monitoring, Capture One's GPU usage on export sits around 80-85%, LrC probably at 90-95%. I have no idea if that's a bad or good thing or whether that's affecting performance. Capture One's GPU usage earlier in the M1 Max regime for me was around 40-50%, so it has definitely improved on the utilization front. 

     

     

     

    0
  • Benjamin Kim

    Raymond Harrison/

    As others have indicated, since LrC and C1 are completely different programs, with different code bases and different developers, the bottlenecks may be (will be) in different places.

    That's such a poor excuse when LR or Adobe is well known for terrible optimization while Capture One Pro already fundamentally redesign their software after M1 released on May 27th, 2021. Beside, Capture One Pro is way more expensive and yet, their optimization is extremely poor. Exporting requires heavily on CPU cores but C1P barely used 2~3 cores based on my testing. Using GPU didn't really improve the performance and still, LR was way faster. Other YouTuber I mentioned already proven that in terms of actual results. 

     

    https://openletter.earth/open-letter-from-a-digital-technician-to-capture-one-6cca76e8

    Believe it or not, C1P heavily criticized especially after 2022 or 2023 from many professionals as they kept failing to satisfied C1P users. We've been requesting many features and improvements for a long time and yet, C1P literally ignored them. Software optimization is just a tip of the iceberg of Capture One Pro's problem and this is not the first time we disappointed toward C1P. I am not the only one complaining about how and what C1P is doing so far. Fixing and improving C1P is already failed and ignored instead of wasting their time on meaningless mobile apps. Is this how they treat professionals? You better read the context from the link above cause quite a lot of digitechs extremely worried about C1P since version 23. 

     

    At this point, it's just a poor excuse as they keep failing to meet our expectations. 

    0
  • Raymond Harrison

    David Lee

    OK - I'll leave you to approach problem solving in your own way. Good luck with the open letter (assuming Capture One have read it)!

    0
  • Raymond Harrison

    And the letter is interesting, by the way. They should advertise it more since there's not as many signatories as I'd have expected. If the 600+ people who signed it came here and upvoted this feature request, it'd definitely get the company's attention. 

    0
  • Brian Jordan
    Moderator

    “Our” ……meeting my expectations pretty well so you might want to rephrase that.

    -1

Please sign in to leave a comment.