Skip to main content

12 image CPU - GPU exporting benchmarks.

Comments

94 comments

  • cdc
    I'm interested in the results as well as I'm thinking about building a PC. Though computing power is only one aspect to how fast images will be processed. The difference between a set of images of outdoor scenes rich in detail takes nearly twice as long as the same number of images of a model in front of a white background. So unless everybody is processing the same exact images the results will not be to definitive.

    I can tell you however that my little mid-2014 13" macbook with 2.6 GHz Intel Core i5/16GB DDR3/Intel Iris 1536 MB/ that these tests were run on does fine running capture one tethered to 36mp cameras all day. But when it comes time to process out 3000 images it will take this thing 10 hours compared to a base model iMac's 2-3 hours.

    Anyway here is the test from my macbook, I know, another mac....

    Full size RAW to JPG
    80% quality
    No output sharpening

    5DmkIII 12 CR2 -> JPG Subject model on white background

    Acceleration on: 1:06
    Acceleration off: 1:14

    D800 12 NEF -> JPG Subject model on white background

    Acceleration on: 1:11
    Acceleration off: 1:16

    D800 12 NEF -> JPG Subject outdoor scenes rich in detail

    Acceleration on: 2:06
    Acceleration off: 2:05
    0
  • Dave R
    CraigJohn wrote:

    So yeah! GPU acceleration is pretty significant for the Sony and Nikon files. Not so much with the Fuji Xtrans sensor.

    You need the very latest version of Capture One to get full Fuji Xtrans GPU acceleration support.

    Dave
    0
  • Robert Whetton
    CraigJohn wrote:

    would love to see various benchmark lists with these...
    CPUs
    Intel: 7700K, 5820K, 6800K, 6850k, 7800x, 7820x, 7900x
    Ryzen: 1600, 1600x, 1700, 1700x, 1800x, Threadripper 1900x, 1920x

    GPUs
    Nvidia: GTX 970ti, GTX 1050, GTX 1060, GTX 1070, GTX 1080, GTX 1080ti
    AMD: Radeon RX 480, Radeon RX 560, Radeon RX 570, Radeon RX 580 (8GB variant), RX Vega 56, RX Vega 64

    If you have anything different, please post it - the more, the better...😄


    My 7D2 files [5472x3648] take just under a second to process with my system, i5 2500K with R9 390 8GB

    I've tested with model on black backround on following 100% size JPG Quality 90:

    Nikon D750 [6016x4016] = 2 seconds x 12 = 24 sec

    Sony A7R2 [7952x5204] = 3 seconds x 12 = 36 sec

    Canon 5D3 [5760x3840] = 3 seconds x 12 = 36 sec

    Looking at the processing power of Vega, the 64 is going to rip though files!

    For a more accurate bench, would be helpful to have a link to the RAW files you used.
    0
  • Dave R
    System
    Windows 10 Pro v1703
    i7-3820
    32Gig 1600MHz RAM
    Samsung SSD 850 EVO 1TB
    AMD Radeon PRO WX 5100 Graphics card

    12x XPro-1 RAF to full size JPG
    with GPU acceleration 20 sec (1.7 secs per image)
    without GPU acceleration 43 seconds (3.6 sec per image)

    That WX 5100 was a good investment 😎
    0
  • CraigJohn
    cdc wrote:
    I'm interested in the results as well as I'm thinking about building a PC. Though computing power is only one aspect to how fast images will be processed. The difference between a set of images of outdoor scenes rich in detail takes nearly twice as long as the same number of images of a model in front of a white background. So unless everybody is processing the same exact images the results will not be to definitive.

    I can tell you however that my little mid-2014 13" macbook with 2.6 GHz Intel Core i5/16GB DDR3/Intel Iris 1536 MB/ that these tests were run on does fine running capture one tethered to 36mp cameras all day. But when it comes time to process out 3000 images it will take this thing 10 hours compared to a base model iMac's 2-3 hours.

    Anyway here is the test from my macbook, I know, another mac....

    Full size RAW to JPG
    80% quality
    No output sharpening

    5DmkIII 12 CR2 -> JPG Subject model on white background

    Acceleration on: 1:06
    Acceleration off: 1:14

    D800 12 NEF -> JPG Subject model on white background

    Acceleration on: 1:11
    Acceleration off: 1:16

    D800 12 NEF -> JPG Subject outdoor scenes rich in detail

    Acceleration on: 2:06
    Acceleration off: 2:05





    My i7 (1.7GHz) MacBook Air 6.2 is perfect for shooting tethered - at least for my needs - with the Nikon D610/750 and D810. It's dog slow for exporting. My 2015 2.5GHz MacBook Pro Retina was great at everything but exporting.

    Yes. It's true about indoor/outdoor (complexity of) scenes - and if you wanted to do this scientifically, sure. You'd use the same 12 images shot with each camera. Considering I shoot in all sorts of settings, and with all sorts of cameras, I get the feeling this will give a reasonably good enough idea.
    0
  • cdc
    CraigJohn wrote:
    cdc wrote:

    Yes. It's true about indoor/outdoor (complexity of) scenes - and if you wanted to do this scientifically, sure. You'd use the same 12 images shot with each camera. Considering I shoot in all sorts of settings, and with all sorts of cameras, I get the feeling this will give a reasonably good enough idea.


    Right, I was just pointing out that these tests you are asking people to make would need to be a little more controlled if you truly wanted to build a resource of benchmarks with various CPU/GPU combinations.
    0
  • CraigJohn
    Bobtographer wrote:
    CraigJohn wrote:

    would love to see various benchmark lists with these...
    CPUs
    Intel: 7700K, 5820K, 6800K, 6850k, 7800x, 7820x, 7900x
    Ryzen: 1600, 1600x, 1700, 1700x, 1800x, Threadripper 1900x, 1920x

    GPUs
    Nvidia: GTX 970ti, GTX 1050, GTX 1060, GTX 1070, GTX 1080, GTX 1080ti
    AMD: Radeon RX 480, Radeon RX 560, Radeon RX 570, Radeon RX 580 (8GB variant), RX Vega 56, RX Vega 64

    If you have anything different, please post it - the more, the better...😄


    My 7D2 files [5472x3648] take just under a second to process with my system, i5 2500K with R9 390 8GB

    I've tested with model on black backround on following 100% size JPG Quality 90:

    Nikon D750 [6016x4016] = 2 seconds x 12 = 24 sec

    Sony A7R2 [7952x5204] = 3 seconds x 12 = 36 sec

    Canon 5D3 [5760x3840] = 3 seconds x 12 = 36 sec

    Looking at the processing power of Vega, the 64 is going to rip though files!

    For a more accurate bench, would be helpful to have a link to the RAW files you used.



    In theory, that should work. I've exported singular files, and it's taken a couple of seconds. But when there's a batch of images, seems like it takes Capture One a bit longer to process the set than it should going by the math.
    0
  • Gustavo Ferlizi
    Crap old OC'd AMD triple-core + Geforce GTX 670 2GB + Old Corsair SSD + 8GB DDR2 + Windows 10 Pro

    On my machine export time is consistent for most scenes, with really busy ones taking only up to 2 seconds longer.

    12 Fuji XT-2 uncompressed raw, varied scenes, varied ISO's, varied editing; to 100% sRGB JPEG.

    I tested with no render penalty for added grain and varied editing.

    GPU accelerated: 1:08 (5-6 seconds each)
    CPU only: 4:56 (24-26 seconds each) (version 10.0.2 does 4:50)
    0
  • Christian Gruner
    Here are a few on AMD Ryzen 7 1800x, with 2 x AMD R9 Nano (benchmark at 0.55), processing to and from a M.2 disk:

    D810, 12 images
    CPU-jpeg: 45sec
    GPU-jpeg: 19 sec
    GPU-tiff: 6 sec

    5Dmk II/III, 12 images
    CPU-jpeg: 34 sec
    GPU-jpeg: 13 sec
    GPU-tiff: 5 sec

    X-Pro2, 12 images:
    CPU-jpeg: 31 sec
    GPU-jpeg: 15 sec
    GPU-tiff: 9 sec

    Bonus-info, the machine does 99 5dm3 raws in 39 seconds when processing to to 8 bit tiff, quite nice performance!

    When doing these benchmark, call up the Task Manager, and look at the Performance graphs under the Performance tab. When processing to uncompressed tiff, if the either the CPU or the Disk in use maxes out, there is your bottleneck.
    If processing to any other format, that uses compression, the loading is a bit more complex, and thus can't be simplified like that.
    0
  • Gustavo Ferlizi
    Christian Gruner wrote:
    Here are a few on AMD Ryzen 7 1800x, with 2 x AMD R9 Nano (benchmark at 0.55), processing to and from a M.2 disk:

    D810, 12 images
    CPU-jpeg: 45sec
    GPU-jpeg: 19 sec
    GPU-tiff: 6 sec

    5Dmk II/III, 12 images
    CPU-jpeg: 34 sec
    GPU-jpeg: 13 sec
    GPU-tiff: 5 sec

    X-Pro2, 12 images:
    CPU-jpeg: 31 sec
    GPU-jpeg: 15 sec
    GPU-tiff: 9 sec

    Bonus-info, the machine does 99 5dm3 raws in 39 seconds when processing to to 8 bit tiff, quite nice performance!

    When doing these benchmark, call up the Task Manager, and look at the Performance graphs under the Performance tab. When processing to uncompressed tiff, if the either the CPU or the Disk in use maxes out, there is your bottleneck.
    If processing to any other format, that uses compression, the loading is a bit more complex, and thus can't be simplified like that.


    For me individual XT-2 uncompressed 8bit tiffs take less than a second each. In a 12 image batch it takes 55 seconds. What do you think accounts to that toll?
    0
  • CraigJohn
    Christian Gruner wrote:
    Here are a few on AMD Ryzen 7 1800x, with 2 x AMD R9 Nano (benchmark at 0.55), processing to and from a M.2 disk:

    D810, 12 images
    CPU-jpeg: 45sec
    GPU-jpeg: 19 sec
    GPU-tiff: 6 sec

    5Dmk II/III, 12 images
    CPU-jpeg: 34 sec
    GPU-jpeg: 13 sec
    GPU-tiff: 5 sec

    X-Pro2, 12 images:
    CPU-jpeg: 31 sec
    GPU-jpeg: 15 sec
    GPU-tiff: 9 sec

    Bonus-info, the machine does 99 5dm3 raws in 39 seconds when processing to to 8 bit tiff, quite nice performance!

    When doing these benchmark, call up the Task Manager, and look at the Performance graphs under the Performance tab. When processing to uncompressed tiff, if the either the CPU or the Disk in use maxes out, there is your bottleneck.
    If processing to any other format, that uses compression, the loading is a bit more complex, and thus can't be simplified like that.



    Balls - those CPU marks are faster than my GPU marks. Maybe I need to rethink my build and consider an AMD Ryzen chip. Even if Photoshop processing with a 1700x is generally 10-15% slower than the 7820x (in testing theory), it should still be wildly faster than my 2009 2.66GHz QuadCore Xeon W3520. LOL

    And your GPU marks nearly a 1/3 what mine were. I also wonder if Capture One Pro 10 can use both the 16 available threads as well as the GPU efficiently. If so, this would be outstanding.

    ...I wonder how much the M.2 drive comes into play. This is why I'm itching to have two or more NVMe drives in my system. I'd eventually like to have a 256GB Boot/system and Applications drive, a 1TB active jobs drive and a 500GB scratch disk. Then keep cued, backups, finished/archived jobs on 7200RPM drives.

    EDIT! Forgot to ask - how does Capture One Pro behave in terms of regular use? Are the slider actions smooth and responsive, thumbnail rendering, library scrolling, etc, retouching with and without layers, etc...

    Many thanks for posting this. ...still floored your D810 CPU export is practically the same speed as my GPU export. 😂
    0
  • Christian Gruner
    Likely because of caching, so steps can be skipped on a viewed image (i.e. loading of the raw-file and so on)
    0
  • Christian Gruner
    CraigJohn wrote:
    Christian Gruner wrote:
    Here are a few on AMD Ryzen 7 1800x, with 2 x AMD R9 Nano (benchmark at 0.55), processing to and from a M.2 disk:

    D810, 12 images
    CPU-jpeg: 45sec
    GPU-jpeg: 19 sec
    GPU-tiff: 6 sec

    5Dmk II/III, 12 images
    CPU-jpeg: 34 sec
    GPU-jpeg: 13 sec
    GPU-tiff: 5 sec

    X-Pro2, 12 images:
    CPU-jpeg: 31 sec
    GPU-jpeg: 15 sec
    GPU-tiff: 9 sec

    Bonus-info, the machine does 99 5dm3 raws in 39 seconds when processing to to 8 bit tiff, quite nice performance!

    When doing these benchmark, call up the Task Manager, and look at the Performance graphs under the Performance tab. When processing to uncompressed tiff, if the either the CPU or the Disk in use maxes out, there is your bottleneck.
    If processing to any other format, that uses compression, the loading is a bit more complex, and thus can't be simplified like that.



    Balls - those CPU marks are faster than my GPU marks. Maybe I need to rethink my build and consider an AMD Ryzen chip. Even if Photoshop processing with a 1700x is generally 10-15% slower than the 7820x (in testing theory), it should still be wildly faster than my 2009 2.66GHz QuadCore Xeon W3520. LOL

    And your GPU marks nearly a 1/3 what mine were. I also wonder if Capture One Pro 10 can use both the 16 available threads as well as the GPU efficiently. If so, this would be outstanding.

    ...I wonder how much the M.2 drive comes into play. This is why I'm itching to have two or more NVMe drives in my system. I'd eventually like to have a 256GB Boot/system and Applications drive, a 1TB active jobs drive and a 500GB scratch disk. Then keep cued, backups, finished/archived jobs on 7200RPM drives.


    It will eat all the cores it can lay its hands on.
    An M.2 drive will only make sense if you have identified the bottleneck as being the disk. On my machine, as mentioned above, it get a 10% increase in performance when using it, moving the bottleneck to the CPU instead (it basically can't feed the GPU's fast enough).
    0
  • CraigJohn
    Fascinating. Makes me wonder what the Threadripper 1920X would do? Would that feed the GPU more quickly? Would 16 cores be a point of diminishing returns, or steadily increase the CPU to GPU performance.

    Do you know if your dual GPU set up is a benefit over a single GPU? Could you run the same test with a single GPU?
    0
  • Christian Gruner
    CraigJohn wrote:
    Fascinating. Makes me wonder what the Threadripper 1920X would do? Would that feed the GPU more quickly? Would 16 cores be a point of diminishing returns, or steadily increase the CPU to GPU performance.

    Do you know if your dual GPU set up is a benefit over a single GPU? Could you run the same test with a single GPU?


    Everything is interconnected. So, if you double the number of GPU's, but your CPU is marginal, you won't see much improvement. If you double the amount of cores on your CPU, but your GPU is already saturated during processing, you again won't see much improvement.
    And again, if you do both GPU and CPU, but your disk is on its limits performance wise, you won't get the expected bonus.

    So it is not straight forward sadly, and it does need careful inspection of the saturation of each hardware component to be able to identify your current bottleneck.
    0
  • CraigJohn
    Christian Gruner wrote:
    Everything is interconnected. So, if you double the number of GPU's, but your CPU is marginal, you won't see much improvement. If you double the amount of cores on your CPU, but your GPU is already saturated during processing, you again won't see much improvement.
    And again, if you do both GPU and CPU, but your disk is on its limits performance wise, you won't get the expected bonus.

    So it is not straight forward sadly, and it does need careful inspection of the saturation of each hardware component to be able to identify your current bottleneck.



    For sure. But if the scalability is this efficient, where Capture One Pro 10 can efficiently utilize 10, 12 or even 16 cores, and....and multiple GPUs....that would be entirely counter to Lightroom's capability - which is neither multicore efficient, nor GPU efficient. Lightroom seems to be minimally efficient with either of those factors - it still relies on core frequency for performance gains.

    That is a fantastic testament to the Capture One Pro development team.
    0
  • CraigJohn
    Christian,

    I can't remember if I asked - how is the develop mode with the Ryzen on Capture One Pro? How quick are the preview renderings, library scrolling rendering, image snapping when selecting a new image to edit? My computer can take a bit to render the previews, taking a while to scroll through the library. When selecting an new image for editing, it can take a second or two for the image to snap into focus.

    Are sliders smooth, is the feedback on the image editing quick: cropping, layers, brushes, white balances, exposure, etc.

    My computer can be a little choppy when using the WB and Exposure Balance sliders - it just isn't always smooth and responsive.

    Thanks for all of your info. 😊
    0
  • Gustavo Ferlizi
    CraigJohn wrote:
    When selecting an new image for editing, it can take a second or two for the image to snap into focus.


    Don't get your hopes up just yet.

    I'm pretty sure that's a behaviour issue with CO because even on my old slow computer, I sometimes get lightning to strike twice in the same place, and it will cache most images in a folder for instant preview.

    Look at the other threads of people complaining about the same thing with fairly powerful systems.

    Also pay close attention at the webinars, where the same behaviour can be seen, albeit less noticeable.
    0
  • Nikolai Vassiliev
    Ryzen 1700, DDR 3200, AMD R9 380x
    12 Sony A7r files
    CPU 58 seconds
    GPU 25 seconds
    0
  • CraigJohn
    NN174596UL wrote:
    Ryzen 1700, DDR 3200, AMD R9 380x
    12 Sony A7r files
    CPU 58 seconds
    GPU 25 seconds



    Are you kidding me? With a 36MP files? This is depressing 😂

    Is your 1700 overclocked? ...and that's with a single GPU, not a dual like the 1800x set up a few posts above.

    How well does Capture One Pro perform with the sliders, scrolling through the library, do the thumbnails pop into focus fairly quickly? Hoping everything is smooth...

    How well does 1700 handle photoshop tasks? Arbitrary rotate, resize, sharpening, cropping...retouching, using the airbrush. I've heard using the airbrush is kind of laggy.

    Thank you for posting.

    Please keep these little tests coming.
    0
  • Stephan Riedelbeck
    CO 10.1.2 Ryzen 1700x no OC, DDR 3000, Powercolor AMD R9 280x
    D810, 12 images
    CPU-jpeg: 42s
    GPU-jpeg: 21s
    GPU-tiff: 11s
    0
  • MadManAce
    C1 10.1.2.23,

    Ryzen 1700 OC, 32GB DDR 2666, GTX 1080 Ti
    12 JPEG, all photos are outdoor portraits with lots of detail, 100% size, 90% sRGB JPEG, No Output Sharpening

    Canon 5Dmk4 CPU: 29.98
    Canon 5Dmk4 GPU: 9.75

    Nikon D800 CPU: 40.95
    Nikon D800 GPU: 11.75

    Fuji XT2 Uncompressed CPU: 26.46
    Fuji XT2 Uncompressed GPU: 10.06

    Fuji XT2 Compressed CPU: 38.80
    Fuji XT2 Compressed GPU: 23.22

    C1 runs smoothly, no issues with sliders or brushes, same with Photoshop. Maybe those reports are from early Ryzen builds, new drivers and agesa 1.0.0.6 have really made Ryzen a rock solid platform.

    To test PS performance I ran this test in 8s, my prior i7 6700k OC to 4.8 MHz build took 15.5s :
    http://ksimonian.com/Blog/2010/02/24/im ... lter-test/
    0
  • CraigJohn
    That Keith Simonian test doesn't test PS performance, but multi-core CPU performance. There's only a few filters that use multicores, whereas the majority of PS prefers single (or dual) core with higher clock frequencies.

    My computer pulled in a whopping 38 seconds on that CPU test.

    I am surprised at your 6700K score (overclocked or not). That's as fast as some 6 and 8 core scores.
    0
  • Robert Whetton
    Intel has the advantage with single core speed. Ryzen is all about multi-core...
    0
  • CraigJohn
    I restarted my computer - made sure nothing was running in the background and reran a the X-T2 test, and ran a Nikon D3s (12MP) test...

    C1P10.1.2.26,
    Mac Pro 2.66GHz QuadCore Xeon W3520, Sapphire Radeon HD 7950, 24GB ram, Internal 2TB 7200RPM WD Black Drive

    Nikon D3s
    GPU = 22 seconds
    CPU = 46 seconds

    Fuji X-T2
    GPU = 44 seconds
    CPU = 1 minutes 10 seconds

    The X-T2 also improved quite a bit.
    0
  • jlV
    Hello!

    my little contribution.

    Phenom X6 1045T, GT430, 16GO, 1TB HDD CO1 10.1.2 Win 10

    Sony Alpha 99

    2 test:
    12 raw (without modification) -> jpg with GPU 1"46; without GPU 2"54
    12 raw (with modification auto)->jpg with GPU 2"09; without GPU 3"30
    0
  • Tx
    Hey Guys,

    have a new build.

    7820x, Vega Froniter, 64GB Ram 2666,
    Nvme 512GB <-for OS, Software, Sysfile <cache).

    Getting a BSOD at the Moment with CO, think it's the GPU beta driver.

    Expect Benchmark at the end of the week.
    0
  • CraigJohn
    Tom, I'm literally salivating in anticipation of your benchmark. I have 7820x build waiting in the wings on PC Part Picker. ...but also have a Threadripper 1900x and a Ryzen 1700x waiting. ...I can't decide, 😂 ...but I know I'm leaning to the 7820x.
    0
  • Tx
    CraigJohn wrote:
    Tom, I'm literally salivating in anticipation of your benchmark. I have 7820x build waiting in the wings on PC Part Picker. ...but also have a Threadripper 1900x and a Ryzen 1700x waiting. ...I can't decide, 😂 ...but I know I'm leaning to the 7820x.



    ok, first numbers ... at the moment it's not that faster in Exporting my last Wedding with 1296 images (with up to 3 full image Layers). It's only 4 to 5 Minutes Faster as the system you see in my signature.
    35 Minutes vs 30 Minutes.

    But Creating Proxy Files is almost double as fast. <--- i have to measure it, this is Feeling. You can count it, almost 2 images a Second.

    Using Capture One before, old system, was also Fluid, loading images to screen was fast an is still fast.

    Using Adobe Bridge is much faster, i exported these images to Tiff 16bit ProPhoto RGB, and Bridge was loading these Files while scrolling, not like before, you scroll and then stop while Bridge is loading.

    Some Numbers,

    All Cores while export, 4 GHz, max 64° (Prolimatech Genesis + Liquid Pro)<--- Best Air Cooler of all Time!
    System Still unstable, did get a Bluescreen with FireFox with no reason while writing this Text. (Unmature: Need BIOS Updates)

    My Thoughts: If you work with the Build, wait to December to get this Thing, cause of Bios updates.<---- you need a Stable Sys to work with.
    If you know what you do <--- get it and watch for Bios updates, this is including 1900X <-- it's a new System.
    Don't get the 1700X and Overclock it, get the 1800X<---- OC is a waste of Time. ( i did oc my 3930K to 4.2 ghz <-- was fuck.d by a stock 6700k with a RX 480 (5 Minutes faster, 1102 images (25 Minutes).

    Don't take my Numbers too serious at the Moment, if you add Layers to CO the CPU has much more to do while export and the drivers for the Frontier are not WHQL (final)( <--- will come at September 13)
    I also should export my Tiff files to another drive not the same (Controller has much to do).

    Next Days i will write down my number with 12 images with no layers.

    Greets
    Tom-D
    0
  • CraigJohn
    Thanks for chiming in with this. Is it possible to run this test without overclocking? Would love to see what your rig can do with stock set-up. I doubt I'll overclock, not for critical client work.

    Also - had a few other questions

    • How is the overall feel when working with Capture One Pro? Are the exposure/WB sliders smooth, and thumbnail draws/library scrolling quick? Mine is painfully slow, especially when loading a new library and trying to scroll through the images.

    • How well does it work with Photoshop? How fast can you open 25 Hi-res images? My computer seems to take forever to open 25 hi-res 24MP Fuji X-T2 files. ...Is Arbitrary rotating a layer on a high-res file smooth?

    many many thanks.
    0

Please sign in to leave a comment.