Exporting benchmarks v2

[garrison]

There is a thread with attempt to collect CO export benchmark statistics on different machines
https://forum.phaseone.com/En/viewtopic.php?f=52&t=26478
So everyone could use this information to build/buy a good computer for CO workflow.

I suggest to improve this bench with fixed set of RAW files, so the collected bench data would be more accurate: it will depend mostly on computer specification (CPU and GPU first), but not on RAW files type (Canon, Nikon, Sony and all other RAW types affect export speed).

So here is the benchmark algorithm:

1. Download the following zip with RAW files and unpack them to some folder
https://drive.google.com/open?id=1sqW34MAafXvRHsBmZmynGcjTVN-LDpyz
50 photos made with Canon 5D mk3, size about 1.3GB.
Theese photos were made by myself so no property rights are violated
2. Start CO and ensure that GPU acceleration is enabled: menu Edit - Preferences - General tab - Hardware acceleration - Processing set to Auto (if it was set to "Never" before, you need to restart CO so changes are applied)
3. Open RAW files that were downloaded on step 1 and wait until they are fully imported (previews are built by CO).
4. Set up the following export parameters:
Format: JPEG
Quality: 100%
ICC Profile: sRGB IEC61966-2.1
Resolution: 300 px/in
Scale: Fixed 100%
Open with: None

It is very important to set up export with theese parameters! Otherwise the bench results won't be reliable!
5. Make the first export (GPU accelerated) noting the duration of this process.
At least one export process is needed. However it would be great if you'll be able to make it 2 or 3 times, because results may vary depending on HDD caching, computer backround processes and others. Take the shortest time duration - it is the first bench result.
6. Disable GPU acceleration: menu Edit - Preferences - General tab - Hardware acceleration - Processing set to Never. Restart the CO.
7. Make the second export (no GPU acceleration) in a similar way as in step 5.
That will be the second bench result.
8. Make a post in this thread with the following format:
Computer type (PC/Mac) and model (if any) and OS version
CPU+GPU - benchmark time 1
CPU only - benchmark time 2
CPU model
GPU model
CO version

You can also make the second benchmark for TIFF export format - it would be good addition to previous JPEG benchmark. Current version of CO (11.0) is limited in perfomance when exporting to JPEG format due to internal algorithms. So TIFF export would show better results in some circumstances, consuming more hardware power. Theese TIFF export parameters are needed to be set up:

. . .
Here is a myself-written Windows utility that will help to easily calculate benchmark results: it analyzes all files in selected folder and automatically calculates time span between opening of first file for writing and last file modifying.
https://drive.google.com/file/d/1mcT6CIaBOraPBCFGLH7s4I_B8By3WP9R/view?usp=sharing
. . .
It is also very welcomed to add your bench values via this form
https://phodograf.com/capture-one-batch-export-benchmarks/
Here you can see all collected benches in user-friendly spreadsheet format (thanks to WPNL) .
. . .
Suggestions for benchmark improvements are very welcomed!
Hope my post doesn't look to rigorous
The only goal of this thread is to help each other with selecting hardware components for best CO workflow experience.

[garrison]

Windows PC (Win10 b1703)
C1 version - 11.0
CPU - Intel Core i7 3770 (4-core, 8-threads, 3.40 GHz)
GPU - NVidia GTX1070

JPEG
CPU+GPU - 0:55 (55s)
CPU only - 2:39 (159s)
The GPU speed up factor is about 2.89x.

TIFF
CPU+GPU - 0:28 (28s)
CPU only - 2:11 (131s)
The GPU speed up factor is about 4.68x.

[David532]

Windows PC (WIN 10 1709)
CPU+GPU 1:04
CPU Only 2:44
CPU - Intel CoreI7 3820
GPU - AMD Radeon Pro WX5100
C1 version 11.0

Dave

[gnwooding]

My Specs
Windows 10 (1703)
Intel i7 5820k
32Gb RAM
2 x Nvidia GTX 1080ti
Capture One 11

For each scenario I ran it with the session first on my NVME, SSD, old mechanical drive and then a 12GB RAM drive. I always saved the files to the same disk the raws were on.

I also did a test with a single 1080ti as well as with both.

Code: Select all

I first ran the tests with my CPU at stock speeds (3.3GHz)
                 NVME          SSD           MECH          RAM
CPU+GPU SLI:     55.15s        52.17s        53.52s        52.62s
CPU+GPU    :     55.29s        53.36s        55.27s        53.37s
CPU        :    116.11s       115.73s       114.48s       113.30s

I then ran the tests with my CPU overclocked to 4.4GHz
                 NVME          SSD           MECH          RAM
CPU+GPU SLI:     43.67s        43.76s        45,31s        44.03s
CPU+GPU    :     44.93s        44.68s        45.38s        45.30s
CPU        :     97.91s        97.51s        98.55s        96.63s


I can see both GPUs being used by Capture One but it seems to alternate between them hence why there is basically no improvement. I suspect even with the single GPU I am Severely CPU limited. Although during test CPU usage sits under 50%.

[garrison]

gnwooding
Very interesting bench! And there is much to think about...
At least several conclusions can be drawn:
1. Disk speed almost does not affect overall time result.
2. GPU SLI mode advantages are not used by C1 to speed up conversion. Or very fast GPU needs proper CPU speed to unleash all its power.
3. CPU with higher single core speed is needed to take more processing power from the fast GPU (maybe I'm wrong).

And it looks very strange that results are the same in GPU accelerated conversion mode comparing to my bench (with slower CPU and GPU) - about 0:55

Windows PC (Win10 b1703)
CPU+GPU - 0:55
CPU only - 2:39
CPU - Intel Core i7 3770
GPU - NVidia GTX1070
C1 version - 11.0

I can't explain that except internal C1 limitations that prevent using all power speed from CPU and GPU. Maybe C1 team expert could help us understad theese results.

[garrison]

Windows PC (Win10 b1703)
CPU - Intel Xeon X5670 (6-core, 12-threads, 2.93 GHz)
GPU - NVidia GTX670
C1 version - 11.0

JPEG
CPU+GPU - 1:22 (82s)
CPU only - 3:57 (237s)
The GPU speed up factor is about 2.89x.

TIFF
CPU+GPU - 0:56 (56s)
CPU only - 3:24 (204s)
The GPU speed up factor is about 3.64x.

[CraigJohn]

Windows 10
CPU - i7 7820x - 8 cores, 16 threads
MOB - ASRock Extreme4
GPU - Sapphire Radeon RX 580 Nitro+
Capture One Pro 11
Sata 6 SSD

3.6GHz (stock speed)
CPU + GPU = 49.2s
CPU = 100.9s
temps averaged 35C

4.4GHz OC
CPU + GPU = 44.1s
CPU = 88s
temps averaged 55c

4.6GHz OC
CPU + GPU = 41.7s
CPU = 85.4s
temps average 70c <--- surprisingly good. I was half expecting 85-90c

I was at the mercy of my budget when I built this computer, and I believe my GPU is bottlenecking my CPU in a big way. I'd like to see what this system would do with a 1080ti.
Would also like to see what this system would do with an NVMe M.2 SSD for my active jobs.

At the time I built this workstation, I had a choice between the 7820x for $375, and the 8700K for $415 USD on Black Friday. I went with the 7820x.

What I especially love about this build, is how smooth and responsive everything is. The adjustment sliders a butter, with zero lag. I can run a massive export, and a large photoshop batch, go back and start making RAW adjustments to a third job, and this computer doesn't break a sweat. Still silky smooth and responsive.

Compare that to my old workstation below...


2009 Mac Pro (2.66GHz Xeon W3520 4 core, 8 threads)
Sapphire Radeon HD 7950 Mac Edition
Mac OSX El Capitan
Capture One Pro 10.2.1

CPU + GPU = 139.2s
CPU = 514.5s

[garrison]

Windows PC (Win10 b1709)
CPU - Intel Xeon L5630 (4-core, 8-threads, 2.13 GHz)
GPU - NVidia GT710
C1 version - 11.0

JPEG
CPU+GPU - 7:00 (420s)
CPU only - 7:23 (443s)
The GPU speed up factor is about 1.05x.

Slow CPU & very slow GPU. This bench is just for ridiculous comparison to fast ones

[garrison]

Windows PC (Win10 b1709)
CPU+GPU - 5:28 (328s)
CPU only - 5:35 (335s)
CPU - Intel Xeon L5640 (6-core, 12-threads, 2.26 GHz)
GPU - NVidia GT710
C1 version - 11.0

The GPU speed up factor is about 1.02x.

Just a little bit faster CPU with the same GPU as in previous bench. It seems like the GPU almost doesn't improve performance with this CPU.

[Chad Dahlquist]

OS - Win10 b1709
CPU - 7820x O.C to 4300MHz
GPU - EVGA 1080 FTW also O.C.
MOB - ASUS strix x299 gaming
AIO - kraken 62 cooler
mem - 32GB Gskill @3200
storage- 1TB NVMe evo for file storage in use files were on this
storage 2 - 250 samsung 850 boot in use
storage 3&4 - WD red 6TB not in use for this
capture one - 11.01

CPU+GPU call it jpeg 32 sec
31.37 2nd time 31.97 3rd 32.39 4th 32.14
I had both chrome with 15 tabs and firefox running and Photoshop and evernote and a couple others really just to kinda do what is my norm
CPU only 67 sec something I would never run anyway

edited to add in Tiff timings
CPU+GPU TIFF 8 bit(Adobe RGB) 23 sec
CPU+GPU TIFF 16 bit(Adobe RGB) 28 sec
I tend to work in TIFF or PSB and 16 bit then output to what is needed
also again same things running PS etc..




I wonder how much human finger error we have ahahahhahah
for fun tried to OC to 4700 but kept locking up also OC the GPU and had a couple crashes as I was trying to push it to far without more tweaking I can get mine to 4500 stable but run at 4300 and can do some work at 4600 but get random lockups just in case folks are curious want stability first quiet 2nd 3rd any extra Mhz is fine but not at the cost of the first 2
my CPU is at 42c right now my gpu at 56c about ten programs running and 6 browser windows
I want to put more fans into the box (positive pressure) the kraken is a pretty impressive cooler I must say

one thing both Craig and I have chatted about with the idea to go to the 7820 is multi tasking I can be doing this AND working in PS with no lag for fun I had C1 up building 3 sessions previews and 1 in LR and could still do other things without the lag mouse feel

[StephanR]

Here some results from my machine

Windows PC (Win10 1709)
CPU+GPU - 54s (GPU use around 30% in taskmanager)
CPU only - 108s
CPU - AMD Ryzen 1700x 3,4GHz 8 Cores 16 Threads
GPU - AMD Radeon 280x
C1 version - 11.0.1

Compressing to jpg seems to be a single core operation (I think I've read this somewhere in this forum) as it uses the JPG functions from Microsoft.
This means first the GPU is converting the raw picture and then only the CPU saves the jpg. At this time the GPU is idle until the next raw file is done. As the AMD Ryzen processor is not as fast in single core operations as Intel it will take a bit more time to store the jpg.

If converting to tif the GPU is used much more:
TIF uncompressed 8 Bit CPU+GPU - 26s (GPU use around 63% in taskmanager)
TIF uncompressed 8 Bit CPU only - 81s

Some time ago I've had an idea to improve the speed!
I'm using a small program made by myself to convert the TIF to JPG with the "open with" operation.
After the file is stored as TIF Capture One starts my program in parallel. The program just converts the TIF to JPG
(keeping meta data) and then erases the TIF file.

Here the results with my extra program:
TIF uncompressed 8 Bit -> open with to jpg CPU+GPU - 27s
TIF uncompressed 8 Bit -> open with to jpg CPU only - 88s
As you can see the speed with GPU is only 1s slower then only the TIF conversion.
Even the CPU only test is faster (all cores are running around 100%) as the jpg conversion in CO1
And as you can see my AMD 280x is very old in comparison to a NVidia 1080.

In my opinion this is also the reason why two fast GPU cards do not show much performance differences.
Therefore the times for TiF would be also very interessting for other CPU/GPU combinations.

[CraigJohn]

The GPU and NVMe drive is clearly bottle necking my system. I'll be looking to upgrade to a 1080ti and NVMe later this year when I have a more $$$. For now I'm good to go.

[WPNL]

Windows 10 Build 1709
CPU+GPU - 0:40 (40 sec)
CPU only - 1:44 (104 sec)
Intel Core i7-6700K (overclocked, delidded and watercooled)
MSI Radeon R9 290X Gaming 4G
C1 11.01

GPU temp, when used while rendering +/- 60 Degrees Celcius
CPU temp, when used while rendering +/- 45 Degrees Celcius

Less than a second difference between working on a 7.200RPM HDD (150 Mb/s) vs RAID0 SSD (1.000 Mb/s).
That's not what I expected! Drive activity stays below 10%...

[photoGrant]

Here some results from my machine

Windows PC (Win10 1709)
CPU+GPU - 54s (GPU use around 30% in taskmanager)
CPU only - 108s
CPU - AMD Ryzen 1700x 3,4GHz 8 Cores 16 Threads
GPU - AMD Radeon 280x
C1 version - 11.0.1

Compressing to jpg seems to be a single core operation (I think I've read this somewhere in this forum) as it uses the JPG functions from Microsoft.
This means first the GPU is converting the raw picture and then only the CPU saves the jpg. At this time the GPU is idle until the next raw file is done. As the AMD Ryzen processor is not as fast in single core operations as Intel it will take a bit more time to store the jpg.

....

Good research Stephan, thanks.

[photoGrant]

Some time ago I've had an idea to improve the speed!
I'm using a small program made by myself to convert the TIF to JPG with the "open with" operation.
After the file is stored as TIF Capture One starts my program in parallel. The program just converts the TIF to JPG
(keeping meta data) and then erases the TIF file.

Here the results with my extra program:
TIF uncompressed 8 Bit -> open with to jpg CPU+GPU - 27s
TIF uncompressed 8 Bit -> open with to jpg CPU only - 88s
As you can see the speed with GPU is only 1s slower then only the TIF conversion.
Even the CPU only test is faster (all cores are running around 100%) as the jpg conversion in CO1
And as you can see my AMD 280x is very old in comparison to a NVidia 1080.

In my opinion this is also the reason why two fast GPU cards do not show much performance differences.
Therefore the times for TiF would be also very interessting for other CPU/GPU combinations.

This is a great solution but it simply tells me that Phase One can do a much better job at utilizing available resources (you're parallel tasking where Capture ONE should be).

I'd give this information to Phase in a technical support case and hope it lands on good ears..

G