Comparative Speed-Test 2

[PhaseoneUser52346]

Comparative Speed-Test - PART 2
Apr 08, 2013

This is a Test I found in this german magazin "http://shop.heise.de/katalog/ct-digitale-fotografie-2-2013".
(They have also an english edition but not allways with the same stories | http://shop.heise.de/katalog/ct-digital-photography-11)

In this very interesting Test they compare the speed, OpenGL, Open CL of 13 GPU with different software.

For me there are 2 interesting things to see.
1. You can see how important a new, powerfull GPU can be to speedup your workflow also in C1 vers 7.x
2. You can see C1 vers. 7.x compared to Corel ASP (former Bibble). Corel made also a lot of improvements
but was able to save the speed in their workflow.

GPU_________________CPU_________________Corel ASP_____________________CaptureOne
(100NEF-Photos > 8Bit Tiff)

Intel HD Graphics 2000..... Intel i3 2120..................117 sec.....................................809 sec
AMD Radeon HD 7750.......Intel i3 2120....................40 sec.....................................223 sec
Intel HD Graphics 3000......Intel i7 2600K..................58 sec.....................................523 sec
AMD Radeon HD 6450........Intel i7 2600K..................44 sec.....................................525 sec
Nvidia Geforce 210...........Intel i7 2600K..................53 sec.....................................533 sec
AMD Radeon HD 7750........Intel i7 2600K.................30 sec......................................207 sec
Nvidia Geforce GTX 650 Ti..Intel i7 2600K.................37 sec......................................216 sec
AMD Radeon HD 7870........Intel i7 2600K.................30 sec......................................150 sec
Nvidia Geforce GTX 660.....Intel i7 2600K.................35 sec......................................170 sec
AMD Radeon HD 7970........Intel i7 2600K..................30 sec..................................... 134 sec
Nvidia Geforce GTX 680.....Intel i7 2600K..................33 sec......................................156 sec
Intel HD Graphics 4000......Intel i7 3770K...................50 sec.....................................530 sec
AMD Radeon HD 7750........Intel i7 3770K..................28 sec......................................206 sec

AMD Radeon HD 7750 with Intel i7 3770K seems to be the best bundle for Corel, it converts the 100 NEF files in 28 sec.
For C1 it is the AMD Radeon HD 7970 with Intel i7 2600K, it converts the 100 NEF files in 134 sec.
Thats a difference of 106 sec (1min 46sec)

Corel converts the 100 files in (min) 28 sec - (max) 117 sec
Capt1 converts the 100 files in (min) 134 sec - (max) 809 sec



Comparative Speed-Test - PART 1
Dez 06, 2012:
viewtopic.php?f=47&t=13583

Test PC is my Laptop: Sony Vaio VGN-Z11VN
RAW-files from my Canon EOS 1Ds MIII

Tests:
1) Importing Images and calculate Previews (36 images)
2) Developing 36 images WITHOUT adjustments or crops into Jpeg sRGB 8Bit + Tiff 16Bit Adobe RGB

Results:

C1 vers. 3.7.10 (32bit):

Test 1 - Importing Images and calculate Previews (36 images): Time: 1:23 Min
Test 2 - Developing 36 images WITHOUT adjustments or crops: Time: 12:56 Min

CPU-Usage from 52-100% - RAM from 43-45% (lowest an highest during Test)
=======================================================================
C1 vers. 6.4.4 (64bit):

Test 1 - Importing Images and calculate Previews (36 images): Time: 1:05 Min
Test 2 - Developing 36 images WITHOUT adjustments or crops: Time: 15:28 Min

CPU-Usage from 64-100% - RAM from 50-100% (lowest an highest during Test)
=======================================================================
C1 vers. 7.0. (64bit):

Test 1 - Importing Images and calculate Previews (36 images): Time: 1:48 Min
Test 2 - Developing 36 images WITHOUT adjustments or crops: Time: 28:49 Min

CPU-Usage from 94-100% - RAM from 74-84% (lowest an highest during Test)
=======================================================================
C1 vers. 7.1. (64bit): OpenCL completely disabled UPDATE

Test 1 - Importing Images and calculate Previews (36 images): Time: 1:25 Min
Test 2 - Developing 36 images WITHOUT adjustments or crops: Time: 27:09 Min

CPU-Usage from 89-100% - RAM from 49-61% (lowest an highest during Test)
=======================================================================
UPDATE:
good old 3.7.10 (32bit) is more than 2x faster than brand new 7.1 (64bit),
12:56 to 27:09 Min.... on the same machine under the same conditions!

[Ulf]

Hi

Very interesting test.
But to be honest, if you have a new software which draws advantages of hardware features such as GPU then it should also tested on a computer where such features can be taken advantage of.

The test computer magazine did, all graphic card which has a processing time around 200sec and lower are supported graphic cards. While the ones that is around 500 sec and longer are not supported and the time difference are CPU based rather then GPU based (big difference between i3 and i7)

On your laptop (Sony Vaio VGN-Z11VN ) the graphic card is to poorly spec'ed in order to utilized OpenCL, so enabled or disabled makes no difference. Reason why there is a small difference was probably due to some background service.

I tested a laptop with i7 2.3Ghz CPU, SolidState disk and AMD Radeon HD6750M

9x 1DsMKii Images.

Capture One 7.1.1 OpenCL Enabled:
22.4 Sec
Capture One 7.1.1 OpenCL Disabled:
43,5 Sec
Capture One 3.7.9
47.2 Sec

The more powerful graphic card, the faster it is.
Another point regarding 3.7 is that it can only utilize 2 CPU's while Capture One 7 can utilize up to 32 cores.

[Alain]

Hi

Very interesting test.
But to be honest, if you have a new software which draws advantages of hardware features such as GPU then it should also tested on a computer where such features can be taken advantage of.
...

But then the GPU has to be utilized by the software. I have a AMD 7870 on a i5 2500K and C1 7.1.1 only uses the 7870 for about 25% while doing display work and I don't see a visible speed difference with GPU enabled or not.

BTW. I don't find processing images (even to a 16-bit tiff file) a speed problem, I can do something else then. When I'm using c1, culling, going to 100%, drawing masks etc, the speed if far more important.

Alain

[PhaseoneUser52346]

@ Ulf

The computer magazine published a 13 pages long story around that test - I can only recommend to read it because there is a lot more interesting informations how and why they used this or that configuration and so on. Due to copyright I can´t show it here, sorry...

How ever, of course, you are right, the newest hardware should work better with the newest software - but can you imagine that everybody can/will buy a new computer with the latest hardware for every software update or new software????

You start your reply with the words: "But to be honest" - Ok, let´s be honest!
For me C1 Vers7.x is a beta version because you can´t sell a software which works at 100% only with the newest hardware
What you also can see in that test is that another software (Corel, former Bibble) is able to be so much faster than C1 - are they magicians? Of course not!
That is those people which abandoned C1 a few years ago and it seems that they also kept their know-how...

What concerns my laptop, you are right again, the graphic card is poor - no question - but it works fine with CS5 and even faster with CS6. And C1 3.7.x is 2x faster than C1 7.x and it can only utilize 2 CPU's while Capture One 7 can utilize up to 32 cores. For my job I need a small but powerfull laptop - my Vaio was, at the time (2008), the best available machine (2.500€) 1,5 kg and 13,1" screen with 1600x900 pix non glare screen, ssd, 8GB RAM, Win7 64bit, ...

But I have similar problems on my desktop, Win8 64bit, Intel Core 2 Quad CPU Q6600 2,4GHz, 8GB RAM, ATI Radeon HD5700, SSD - it isn´t that much faster than my laptop...

So the only official answer for all of us having those speed troubles is to buy a new hardware = computer?
And what do wee need when C1 vers. 8.x will come out?

[SFA]

I am part of a test team for a non-image processing related Windows only software application for which a clear requirement from the users and for future strategic development needs for the product dictated a change of "engine" for the underlying processing. The challenge is similar (but without any real focus on GUI needs) to the step up undertaken for larger files and bigger data needs for the C6 to C7 changes. Continuing with the original set up was really no longer an option.

The new product development has to find ways to reproduce what the existing product does. Both core engines are Microsift products but most of the development means cutting new code. In many situations a feature or function in the old engine, used for speed of perceived response, is simply not available in the new engine. It does things quite differently and is more suited to large scale processing at speed than to instant on screen responses that we all like to see. However it is more scalable and productive in a heavy workload environment so not all of the news is bad.

Being newer software core technology it has been developed with newer hardware developments in mind. However some considerations are not absolutely clear in terms of best use. There are, for example, different opinions about the benefits of Hyperthreading or even the use of multi cores depending on both hardware configurations (many millions out there) and the application software in use 0 maybe even some settings parameters for the application software.

Got an interactive virus checker that assesses every new file created? Any engine that maybe uses multicore/multithreading is likely to be creating a lot of small files in memory and each may be checked by the virus checker. All of that saps processing speed to some extent. Running a RAID array? Is that having a performance impact? Does the RAID benefit outweigh the potential foro processing speed reduction? Will you really notice?

I'm running C1 V7 on a i7 notebook processor with 8Gb RAM and Win 7 64. It's not all "instant" but the speed generally seems fine to me. Hardly anything takes more than a second or so, most things much faster especially bulk processing of any sort. All much faster than C1 6 on my old Core Duo notebook or Pentium 4 based desktop machine.

As part of an experiment for the business software testing I have been using the i7 machine with Hyperthreading turned off. So just 4 primary cores in use.

For the business application testing some very rough timings suggested performance for the current level of development was hardly affects. Large data loads, for example, might take half a second or so longer - just under 17 seconds rather than just over 16 seconds for about 6 million records to be manipulated in some way.

Leaving hyperthreading turned off and running C1 V7 some things 'seem' a little less responsive - which I guess probably means they might be APPEARING to take about twice as long but if so we are talking 1.5 or 2 seconds rather than 1. Not something one can measure without doing some internal timing logs. Appearances can be misleading. However, if actual rather than just an appearnce, it means that C1 V7 looks well placed to take advantage of new hardware and firmware developmentas as they come along and we all eventually have to buy new equipment as older machines fail at the end of their natural lives.

The reality is that newer technologies come together to produce "better" or "Faster" or more consistent performance overall but can be constrained by processing objectives that are not nearly the same as they were in earlier technologies.

From a business perspective a strategic decision is required. We all know this of course. We all make such decisions every time we think of buying a new camera (or lens) from which we expect some benefit or other. Why do we keep buying new cameras when it is clear the manufacturers could simply offer all of the benefits and advantages we feel we need with inexpensive upgrades (often to firmware) for our existing equipment?

There are a lot of variables in this game. I'm not at all sure that detailed technical testing will tell us much of practical value in real time use. Or at least not much that is widely useful to everyone all the time.

With or without Hyperthreading turned on for me C1 V7 is working about as fast as I can work with it. Maybe I am not pushing it hard enough? Or perhaps I am just slow?


Grant Perkins

[Alain]

I am part of a test team for a non-image processing related Windows only software application for which a clear requirement from the users and for future strategic development needs for the product dictated a change of "engine" for the underlying processing. The challenge is similar (but without any real focus on GUI needs) to the step up undertaken for larger files and bigger data needs for the C6 to C7 changes. Continuing with the original set up was really no longer an option.

The new product development has to find ways to reproduce what the existing product does. Both core engines are Microsift products but most of the development means cutting new code. In many situations a feature or function in the old engine, used for speed of perceived response, is simply not available in the new engine. It does things quite differently and is more suited to large scale processing at speed than to instant on screen responses that we all like to see. However it is more scalable and productive in a heavy workload environment so not all of the news is bad.

Being newer software core technology it has been developed with newer hardware developments in mind. However some considerations are not absolutely clear in terms of best use. There are, for example, different opinions about the benefits of Hyperthreading or even the use of multi cores depending on both hardware configurations (many millions out there) and the application software in use 0 maybe even some settings parameters for the application software.

Got an interactive virus checker that assesses every new file created? Any engine that maybe uses multicore/multithreading is likely to be creating a lot of small files in memory and each may be checked by the virus checker. All of that saps processing speed to some extent. Running a RAID array? Is that having a performance impact? Does the RAID benefit outweigh the potential foro processing speed reduction? Will you really notice?

I'm running C1 V7 on a i7 notebook processor with 8Gb RAM and Win 7 64. It's not all "instant" but the speed generally seems fine to me. Hardly anything takes more than a second or so, most things much faster especially bulk processing of any sort. All much faster than C1 6 on my old Core Duo notebook or Pentium 4 based desktop machine.

As part of an experiment for the business software testing I have been using the i7 machine with Hyperthreading turned off. So just 4 primary cores in use.

For the business application testing some very rough timings suggested performance for the current level of development was hardly affects. Large data loads, for example, might take half a second or so longer - just under 17 seconds rather than just over 16 seconds for about 6 million records to be manipulated in some way.

Leaving hyperthreading turned off and running C1 V7 some things 'seem' a little less responsive - which I guess probably means they might be APPEARING to take about twice as long but if so we are talking 1.5 or 2 seconds rather than 1. Not something one can measure without doing some internal timing logs. Appearances can be misleading. However, if actual rather than just an appearnce, it means that C1 V7 looks well placed to take advantage of new hardware and firmware developmentas as they come along and we all eventually have to buy new equipment as older machines fail at the end of their natural lives.

The reality is that newer technologies come together to produce "better" or "Faster" or more consistent performance overall but can be constrained by processing objectives that are not nearly the same as they were in earlier technologies.

From a business perspective a strategic decision is required. We all know this of course. We all make such decisions every time we think of buying a new camera (or lens) from which we expect some benefit or other. Why do we keep buying new cameras when it is clear the manufacturers could simply offer all of the benefits and advantages we feel we need with inexpensive upgrades (often to firmware) for our existing equipment?

There are a lot of variables in this game. I'm not at all sure that detailed technical testing will tell us much of practical value in real time use. Or at least not much that is widely useful to everyone all the time.

With or without Hyperthreading turned on for me C1 V7 is working about as fast as I can work with it. Maybe I am not pushing it hard enough? Or perhaps I am just slow?


Grant Perkins

Hi Grant

Did you test the effect hyper threading with local adjustment brushes?
Any change till c1 9?