⚠️ Please note that this topic or post has been archived. The information contained here may no longer be accurate or up-to-date. ⚠️

"OpenCL benchMark"

NNN635158767546269381

February 10, 2015 14:24

In the Home/Library/Logs directory, one can find the CaptureOne logs.

I just noticed this:
2015-02-10 15:16:19> OpenCL initialization...
2015-02-10 15:16:19> OpenCL : found platform Apple, OpenCL Version : OpenCL 1.2 (Jul 29 2014 21:24:39)
2015-02-10 15:16:19> OpenCL Device : HD Graphics 4000
2015-02-10 15:16:19> OpenCL Driver Version : 1.2(Dec 17 2014 19:37:58)
2015-02-10 15:16:19> OpenCL Compute Units : 16
2015-02-10 15:16:19> OpenCL : Loading kernels
2015-02-10 15:16:19> OpenCL : Loading kernels finished
2015-02-10 15:16:19> OpenCL : Benchmarking
2015-02-10 15:16:21> OpenCL : Initialization completed
2015-02-10 15:16:21> OpenCL benchMark : 2.664000

Pretty interesting is the OpenCL benchmark. my question: Higher number means faster or slower performances ? Could anyone post its values on another machine ?
Regards

Comments

33 comments

Christian Gruner

February 10, 2015 14:46
The higher, the slower.
Other examples:
Ati D700 bencmarks to 0.11
Ati Radeon R9 290 benchmarks to .083
0
Rolf Vassen

February 10, 2015 15:14
[quote="Christian Gruner" wrote:
The higher, the slower.
Other examples:
Ati D700 bencmarks to 0.11
Ati Radeon R9 290 benchmarks to .083

Radeon R9 M295X (iMac 5K)

2015-02-09 19:22:38> OpenCL Device : AMD Radeon R9 M295X Compute Engine
2015-02-09 19:22:38> OpenCL Driver Version : 1.2 (Dec 23 2014 00:17:17)
2015-02-09 19:22:38> OpenCL Compute Units : 32
2015-02-09 19:22:38> OpenCL : Loading kernels
2015-02-09 19:22:38> OpenCL : Loading kernels finished
2015-02-09 19:22:38> OpenCL : Benchmarking
2015-02-09 19:22:38> OpenCL : Initialization completed
2015-02-09 19:22:38> OpenCL benchMark : 0.169608
2015-02-09 20:31:44> Exiting
2015-02-09 20:31:44> OpenCL allocations : 4198416 bytes
2015-02-09 20:31:44> Shutting down
0
Grant Hodgeon

February 10, 2015 15:40
This is huge. Great find.

Can we have more details on this? How and when are benchmarks performed? Does this data help us isolate issues with performance?

Does this mean you have a database of graphics cards and their respective Open CL performance? If you don't I think that would be a good thing to release/share.

Otherwise we can start one ourselves 😊
0
Grant Hodgeon

February 10, 2015 15:46
[quote="Christian Gruner" wrote:
The higher, the slower.
Other examples:
Ati D700 bencmarks to 0.11
Ati Radeon R9 290 benchmarks to .083

Does this mean you don't take advantage of the Dual GPU's?

My R270X in a 2011 Mac Pro runs as a 7950 when not flashed for Mac and pulls in at 0.11.

2015-02-09 13:35:17> OpenCL Device : AMD Radeon HD 7950 Compute Engine
2015-02-09 13:35:17> OpenCL Driver Version : 1.2 (Dec 23 2014 00:17:17)
2015-02-09 13:35:17> OpenCL Compute Units : 28
2015-02-09 13:35:18> OpenCL : Loading kernels
2015-02-09 13:35:18> OpenCL : Loading kernels finished
2015-02-09 13:35:18> OpenCL : Benchmarking
2015-02-09 13:35:18> OpenCL : Initialization completed
2015-02-09 13:35:18> OpenCL benchMark : 0.111000
0
LSdigi.com

February 11, 2015 00:03
[quote="photoGrant" wrote:
[quote="Christian Gruner" wrote:
The higher, the slower.
Other examples:
Ati D700 bencmarks to 0.11
Ati Radeon R9 290 benchmarks to .083

Does this mean you don't take advantage of the Dual GPU's?
/quote]

Would be interested to find out as well. Isn't one of the big benefits of C1 8 dual GPU support?
0
VirtualRain

February 11, 2015 02:23
I believe this is just a dump of the OpenCL discovery process which is part of the API. Nothing should be implied from it other than if it's detecting your hardware correctly.
0
Christian Gruner

February 11, 2015 08:10
We can use multiple GPU's but they are all benchmarked seperately, hence my "x 1".

[quote="photoGrant" wrote:
[quote="Christian Gruner" wrote:
The higher, the slower.
Other examples:
Ati D700 bencmarks to 0.11
Ati Radeon R9 290 benchmarks to .083

Does this mean you don't take advantage of the Dual GPU's?

My R270X in a 2011 Mac Pro runs as a 7950 when not flashed for Mac and pulls in at 0.11.

2015-02-09 13:35:17> OpenCL Device : AMD Radeon HD 7950 Compute Engine
2015-02-09 13:35:17> OpenCL Driver Version : 1.2 (Dec 23 2014 00:17:17)
2015-02-09 13:35:17> OpenCL Compute Units : 28
2015-02-09 13:35:18> OpenCL : Loading kernels
2015-02-09 13:35:18> OpenCL : Loading kernels finished
2015-02-09 13:35:18> OpenCL : Benchmarking
2015-02-09 13:35:18> OpenCL : Initialization completed
2015-02-09 13:35:18> OpenCL benchMark : 0.111000
0
meanwhile

February 11, 2015 10:43
"2015-02-10 15:16:19> OpenCL Device : HD Graphics 4000"

Given that I have the same device ... does C1 8.x use it for OpenCL?

Or do numbers over 1 mean it would be slower?
0
Christian Gruner

February 11, 2015 11:15
The hd4000 will be used, however, if you have a fast CPU, the CPU will likely be quicker.
0
Grant Hodgeon

February 11, 2015 14:57
[quote="Christian Gruner" wrote:
We can use multiple GPU's but they are all benchmarked seperately, hence my "x 1".

[quote="photoGrant" wrote:
[quote="Christian Gruner" wrote:
The higher, the slower.
Other examples:
Ati D700 bencmarks to 0.11
Ati Radeon R9 290 benchmarks to .083

Does this mean you don't take advantage of the Dual GPU's?

My R270X in a 2011 Mac Pro runs as a 7950 when not flashed for Mac and pulls in at 0.11.

2015-02-09 13:35:17> OpenCL Device : AMD Radeon HD 7950 Compute Engine
2015-02-09 13:35:17> OpenCL Driver Version : 1.2 (Dec 23 2014 00:17:17)
2015-02-09 13:35:17> OpenCL Compute Units : 28
2015-02-09 13:35:18> OpenCL : Loading kernels
2015-02-09 13:35:18> OpenCL : Loading kernels finished
2015-02-09 13:35:18> OpenCL : Benchmarking
2015-02-09 13:35:18> OpenCL : Initialization completed
2015-02-09 13:35:18> OpenCL benchMark : 0.111000

I don't mean to nitpick but there was no mention of 'x 1'.

Can you clarify whether this is an internal benchmark or something separate to C-1 entirely?
0
Christian Gruner

February 12, 2015 08:18
My bad, should properly have said the lack of any plural tendencies 😉

The benchmark is a number calculated from a representative step in our image-pipeline, and is as such completely separate and unique for CO.
0
Photocor

February 12, 2015 08:34
To know the performance the benchmark on open gl is interesting but is one part of the performance.
It would be interesting to know if C1 uses all cores, and one or two graphics cards together. (for me it is hexacores and 2 graphic cards), or is it only the frequency of the processors which is significant?
Do you know this?
Regards.
0
Christian Gruner

February 12, 2015 08:50
[quote="Photocor" wrote:
To know the performance the benchmark on open gl is interesting but is one part of the performance.
It would be interesting to know if C1 uses all cores, and one or two graphics cards together. (for me it is hexacores and 2 graphic cards), or is it only the frequency of the processors which is significant?
Do you know this?
Regards.

Please note that CO uses openCL, not openGL .

CO uses all supported and initialized GPU's found, even different models and vendors at the same time (we have a test-machine running here with 2xGTX780 and also using the embedded HD4600 GPU at the same time)

When running on GPU's, we use the CPU for loading raw-file into RAM, and for saving/compressing the processed files to disk. This prevents the system from being so loaded/saturated, that you can still work in CO and other programs.'

With regard to the number of cores vs clockfrequency and so on, it's a complex question, so the simplified answer will be "the more of everything (cores, clock, cache) the faster", as we can utilize all cores when processing.
0
LSdigi.com

February 12, 2015 11:41
[quote="Christian Gruner" wrote:

With regard to the number of cores vs clockfrequency and so on, it's a complex question, so the simplified answer will be "the more of everything (cores, clock, cache) the faster", as we can utilize all cores when processing.

I track my CPU performance using iStat Pro which shows all 6 cores (2013 mac pro), and during processing - either JPEG or full res tiff's, I never see C1 using 100% of each core at the same time - so 600% / 1200%. It seems like there's a lot more room for improvement!

Only video software and fully optimized ones like JPEGmini Pro use 1200% when processing.
0
Christian Gruner

February 12, 2015 11:50
[quote="LSdigi.com" wrote:
[quote="Christian Gruner" wrote:

With regard to the number of cores vs clockfrequency and so on, it's a complex question, so the simplified answer will be "the more of everything (cores, clock, cache) the faster", as we can utilize all cores when processing.

I track my CPU performance using iStat Pro which shows all 6 cores (2013 mac pro), and during processing - either JPEG or full res tiff's, I never see C1 using 100% of each core at the same time - so 600% / 1200%. It seems like there's a lot more room for improvement!

Only video software and fully optimized ones like JPEGmini Pro use 1200% when processing.

When only using CPU or ?
0
Grant Hodgeon

February 12, 2015 13:45
[quote="Christian Gruner" wrote:
[quote="Photocor" wrote:
To know the performance the benchmark on open gl is interesting but is one part of the performance.
It would be interesting to know if C1 uses all cores, and one or two graphics cards together. (for me it is hexacores and 2 graphic cards), or is it only the frequency of the processors which is significant?
Do you know this?
Regards.

Please note that CO uses openCL, not openGL .

CO uses all supported and initialized GPU's found, even different models and vendors at the same time (we have a test-machine running here with 2xGTX780 and also using the embedded HD4600 GPU at the same time)

When running on GPU's, we use the CPU for loading raw-file into RAM, and for saving/compressing the processed files to disk. This prevents the system from being so loaded/saturated, that you can still work in CO and other programs.'

With regard to the number of cores vs clockfrequency and so on, it's a complex question, so the simplified answer will be "the more of everything (cores, clock, cache) the faster", as we can utilize all cores when processing.

This is fantastic information, Christian. I know I and a lot of others would be interested in more of this information and transparency, it's great!

[quote="Christian Gruner" wrote:
When only using CPU or ?

Exactly. My findings are that with CPU only, all 12 cores get maxed.
0
VirtualRain

February 12, 2015 18:00
[quote="Christian Gruner" wrote:

When running on GPU's, we use the CPU for loading raw-file into RAM, and for saving/compressing the processed files to disk. This prevents the system from being so loaded/saturated, that you can still work in CO and other programs.'

It's still not clear to me what hardware is doing what tasks?

When is OpenCL used? All the time? For specific adjustments?

What tasks are CPU bound vs GPU bound?
0
Christian Gruner

February 13, 2015 13:31
[quote="VirtualRain" wrote:
[quote="Christian Gruner" wrote:

When running on GPU's, we use the CPU for loading raw-file into RAM, and for saving/compressing the processed files to disk. This prevents the system from being so loaded/saturated, that you can still work in CO and other programs.'

It's still not clear to me what hardware is doing what tasks?

When is OpenCL used? All the time? For specific adjustments?

What tasks are CPU bound vs GPU bound?

For CO8:
The CPU can do every single task in CO.
The GPU is used (if hardware is supported) for processing to file and rendering Viewer-previews when adjusting an image.

Various examples:
- LCC, CA analysis, Proxy (Preview & Thumb file generation to disk) are always performed on CPU.
- If using the GPU for processing, the CPU will support the processing by preloading raw files (while the GPU is processing another one) and saving the file down to disk (including compressing).
0
LSdigi.com

February 16, 2015 20:08
[quote="Christian Gruner" wrote:
[quote="LSdigi.com" wrote:
[quote="Christian Gruner" wrote:

With regard to the number of cores vs clockfrequency and so on, it's a complex question, so the simplified answer will be "the more of everything (cores, clock, cache) the faster", as we can utilize all cores when processing.

I track my CPU performance using iStat Pro which shows all 6 cores (2013 mac pro), and during processing - either JPEG or full res tiff's, I never see C1 using 100% of each core at the same time - so 600% / 1200%. It seems like there's a lot more room for improvement!

Only video software and fully optimized ones like JPEGmini Pro use 1200% when processing.

When only using CPU or ?

Here is a screenshot processing around 300 full res JPEG's (OpenCL Enabled):

http://i62.tinypic.com/72em14.png

http://i57.tinypic.com/sfz1iw.png

Here is C1 8.1.1 Processing TIFF's

http://i61.tinypic.com/avkfa0.png

And here is JPEGmini processing JPEG's

http://i61.tinypic.com/bgrz2v.png

Notice the spike at the end of the graph compared to C1 activity

http://i57.tinypic.com/b69h0w.png

Any thoughts on this? It does feel like the Mac Pro should be processing JPEG's much faster than it currently is. The system doesn't even slow down, I can still shoot, check focus and process without any noticeable slow-down, however it does feel like you could squeeze the CPU a bit further?

EDIT::

Tried it with Auto CL disabled, does appear it uses more of the CPU and processing speed difference between enabled and disabled is marginal!

http://i60.tinypic.com/2mzxlk7.png
0
Grant Hodgeon

February 16, 2015 20:22
[quote="LSdigi.com" wrote:

I track my CPU performance using iStat Pro which shows all 6 cores (2013 mac pro), and during processing - either JPEG or full res tiff's, I never see C1 using 100% of each core at the same time - so 600% / 1200%. It seems like there's a lot more room for improvement!

Wait CPU goes from 177.9% to 969.9% and that's marginal?

Here's what I'm getting with OpenCL Disabled

https://s3.amazonaws.com/f.cl.ly/items/1N1i2l3h1B2Y0G1Z3527/Screen%20Shot%202015-02-16%20at%202.19.02%20PM.png

Or are you saying the time it takes to process files GPU vs CPU is marginal? Because you're right about that, and that's why OpenCL should stay enabled. The GPU is much better at these tasks.
0
Claudio Lazzaretti

February 17, 2015 07:45
how do you enable and disable OPEN CL ?
0
LSdigi.com

February 17, 2015 22:43
[quote="photoGrant" wrote:
[quote="LSdigi.com" wrote:

I track my CPU performance using iStat Pro which shows all 6 cores (2013 mac pro), and during processing - either JPEG or full res tiff's, I never see C1 using 100% of each core at the same time - so 600% / 1200%. It seems like there's a lot more room for improvement!

Wait CPU goes from 177.9% to 969.9% and that's marginal?

Here's what I'm getting with OpenCL Disabled

https://s3.amazonaws.com/f.cl.ly/items/1N1i2l3h1B2Y0G1Z3527/Screen%20Shot%202015-02-16%20at%202.19.02%20PM.png

Or are you saying the time it takes to process files GPU vs CPU is marginal? Because you're right about that, and that's why OpenCL should stay enabled. The GPU is much better at these tasks.

Sorry Grant I didn't explain myself very well, I meant the processing speed of OpenCL enabled/disabled makes little difference.

And LAZZA - it's in the preferences (Display OpenCL, Processing AutoCL)
0
Christian Gruner

February 17, 2015 23:14
Hmm, you where on a Mac Pro with Dual d500 and what CPU ? You should see quite some difference. Dig a bit more, but I think it would be a good idea to let our Support guys have a look at what is up.
0
Grant Hodgeon

February 17, 2015 23:59
[quote="LSdigi.com" wrote:
[quote="photoGrant" wrote:
[quote="LSdigi.com" wrote:

I track my CPU performance using iStat Pro which shows all 6 cores (2013 mac pro), and during processing - either JPEG or full res tiff's, I never see C1 using 100% of each core at the same time - so 600% / 1200%. It seems like there's a lot more room for improvement!

Wait CPU goes from 177.9% to 969.9% and that's marginal?

Here's what I'm getting with OpenCL Disabled

https://s3.amazonaws.com/f.cl.ly/items/1N1i2l3h1B2Y0G1Z3527/Screen%20Shot%202015-02-16%20at%202.19.02%20PM.png

Or are you saying the time it takes to process files GPU vs CPU is marginal? Because you're right about that, and that's why OpenCL should stay enabled. The GPU is much better at these tasks.

Sorry Grant I didn't explain myself very well, I meant the processing speed of OpenCL enabled/disabled makes little difference.

And LAZZA - it's in the preferences (Display OpenCL, Processing AutoCL)

Strange. GPU should be somewhat faster.

Can you process 100 JPG's to a subfolder called CPU with OpenCL disabled then re-process the same JPG's in a new subfolder called GPU with OpenCL enabled and post the creation time-stamp of the first and last JPG in each folder?

That should give us some solid numbers to go off.
0
Claudio Lazzaretti

February 18, 2015 08:08
You know how to set C1. Wondering if the operating system of Apple Yosemite you can activate or deactivate OpenCL!

I have a iMAC 5k Retina with 32Gb RAM and hard but I Fusion performance of C1 lower the Macbook year 2012 in my office!
0
BeO

December 15, 2015 00:00
[quote="Christian Gruner" wrote:

For CO8:
The CPU can do every single task in CO.
The GPU is used (if hardware is supported) for processing to file and rendering Viewer-previews when adjusting an image.

Various examples:
- LCC, CA analysis, Proxy (Preview & Thumb file generation to disk) are always performed on CPU.
- If using the GPU for processing, the CPU will support the processing by preloading raw files (while the GPU is processing another one) and saving the file down to disk (including compressing).

Christian, am I right assuming that if the GPU is working on a task, the CPU is not contributing to that specific task at that same moment? In other words, GPU is disabled then CPU would do every task in CO, if GPU is enabled it takes certain kind of tasks "completely" away from the CPU?

Is there a list of supported and/or recommended graphic cards? (Windows please)
Is there a list of your capable and reliable test systems?
Is there a list of capable and reliable P1 staffs' privately owned hardware?
(Edit: including drivers....)

The hardware market is full of new and used machines, hard enough to get into it, but even harder if there is no concrete information from P1 regarding C1 best performing hardware; and that is especially sad as C1 is quite demanding compared to other raw converters/image editors, from my experience.

Is it clear what I am looking for? Some guidance other than high-level "faster and bigger is better" (with shooting in the dark then you can spend 10,000$ or more).

Thanks in advance
BeO
0
Permanently deleted user

December 15, 2015 07:46
I can't see any OpenCL log file in home/lib. Does this mean OpenCL is not used? OpenCL is set to auto in C1P (iMAC 27 with NVIDIA GeForce GTX 680MX 2048 MB)
0
BeO

December 15, 2015 10:44
[quote="FL" wrote:
I can't see any OpenCL log file in home/lib. Does this mean OpenCL is not used? OpenCL is set to auto in C1P (iMAC 27 with NVIDIA GeForce GTX 680MX 2048 MB)

Search for the file ImgCore.log
In Windows it is here ( C:\Users\[user]\AppData\Local\CaptureOne\Logs )

It doesn't tell you however, if OpenCL is used. Maybe in the application.log?
0
Permanently deleted user

December 15, 2015 11:42
No such file on my iMAC. Any other ideas / hints?
0
BeO

December 15, 2015 12:27
Thread opener wrote: "In the Home/Library/Logs directory, one can find the CaptureOne logs"

Also, this might be useful:

cheers
0

Post is closed for comments.