Photoshop and Multicore Processors:

[Moppie]

Had a bit of a chat with someone in another thread about Photoshop's ability to use more than more Core.

Rather than argue logic, it was suggested some benchmarks be set, and some tests run.

Tim then asked about performance when converting files, Raw to Tiff, JPEG etc.

So, I have grabbed 10 random RAW files, and started to do some tests.

I will add more as I do them, and others are welcome to do the same, and please feel free to use different methods, just make sure you describe what was done.

For those with multicore Windoz machines, you can choose how many cores Photoshop can use by going to:
Task Manager -> Processes -> right click "photoshop.exe" -> set affinity

For Multi core Mac machines, please see this thread by Steve: http://photography-on-the.net/forum/...d.php?t=463404


I will use the same 10 files for all the following tests:

All tests are carried out on my homebuilt PC.
Intel Q6600 Quad Core CPU
Intel DP35DPM Dragontail motherboard with P35 chipset
APACER DDR2 PC6400 Ram - 4 x 1gb
1 x Seagate 80gb HDD
2 x Seagate 500gb HDD
1 x Western Digital 320gb (primary scratch disk)
8800GT Graphics Card



OPEN RAW FILE IN PHOTOSHOP.
Method:
Open PS CS3, and drag and drop in 10 random Canon .CR2 files.
This opens Adobe Camera Raw.
Select all 10 images.
Preform auto adjust and add 15 saturation.
Select "OPEN".
This opens all 10 images as editable 16bit files in PS, but still with the .CR2 file tag.

Results:

• 1 core: 45sec
• 2 core: 33sec
• 4 core: 29sec

SAVE AS 16bitt .TIFF
Method:
Open PS CS3, and drag and drop in 10 random Canon .CR2 files.
This opens Adobe Camera Raw.
Select all 10 images.
Preform auto adjust and add 15 saturation.
Select "SAVE".
This saves all 10 images in the file type and folder of your choice. I saved in the same folder on an 80gb SATA2 Seagate Raptor, that RAW files are in.

Results:

• 1 core: 1:06sec
• 2 core: 45sec
• 3 core: 40sec
• 4 core: 40sec

It is interesting that useing 3 or 4 cores made no difference. Its possible that HDD read write speed is bigger factor than CPU speed.



CONVERT 16bit .TIFF TO 8bit .JPEG
Method:
Open PS CS3, and drag and drop in 10 random Canon .TIFF files.
Open the Photoshop Image Processor, and use it to save all 10 open files in the same folder as level 12 JPEG's.


Results:

• 1 core: 38sec
• 2 core: 31sec
• 4 core: 30sec

Again, it would appear HDD access speed is having an effect on it.


RETOUCH ARTISTS SPEED TEST, The most conclusive yet.

Exactly the same test as run in this thread: http://photography-on-the.net/forum/...d.php?t=170063

This test uses a variety of filters, mode and layer changes, as well as different tools, so a whole variety of what photoshop can do is tested.

RESULTS

• 1 core: 1:29sec
• 2 core: 1:07sec
• 3 core: 1:04sec
• 4 core: 46sec

Clearly the more cores availiable here, the faster Photoshop can work.



NOTE: All tests so far show that the more Cores Photoshop CS3 has access to, the faster it can work, up until it reaches the limits of other system Hardware.

It also appears there is not a linear relationship between the number of cores and performance, and the supporting hardware is just as important as the CPU.

It does however show that the more cores you have, the faster your photoshop experiance will be.

My next plan is to run Photoshop and Lightroom together, have Photoshop run the Retouch Artists speed test while Lightroom converts a large batch of Raw files.

More tests to follow.

[CyberDyneSystems]

Well, allow me to remove the hard drive bottle neck from the testing scenario by running the Retouch Artist test on Skynet, which uses RAM for a scratch Disk..

SuperMicro H8DCE Dual Opteron Motherboard Nforce Professional
Two Dual Core Opteron 275s @ 2.2GHZ per core
8 GB DDR 400 ECC RAM
Windows XP 64 bit
Swap file 1: 3.5 GB System RAM full speed in virtual drive "ramdisk"
Swap File 2: 4GB DDR 400 on a Gigabyte I-Ram (SATA controller is the bottle neck here )

RETOUCH ARTISTS SPEED TEST

Exactly the same test as run in this thread: http://photography-on-the.net/forum/...d.php?t=170063

** except I did NOT reconfigure PSCS this time to limit histories etc and squeeze out more performance. Since I am looking for a comparison here, not top speed, and want real world results, I left all my defaults as is



RESULTS

• 1 core: 2:34 sec
• 2 core: 1:58 sec
• 4 core: 57 sec

Not only did going to four cores have more affect than doubling the cores did, but for both the dual core and quad core tests I watched on the Performance Tab of Task Manager.
In the first Image you can see that though the system has dropped down to 33% CPU usage at the moment I hit print screen, the history of the two cores indicates they were both in use 100% for some time;

bench 2 cores.JPG

In the next image we can see that all four cores were lit up for some time, and at the moment I hit print screen we were just shy opf all four at 100%

bench 4 cores.JPG

It was clear that all four cores were being used for most of the actions involved. There were some points when CPU activity dropped all together even with the RAM bandwidth available,. Some filters or actions seemed to use only two cores. But most of the time it was using all four.

So back to Tom's Hardware test.
It is one thing to do a test that shows that a hardware bottleneck has been reached, and another thing to see results that actually indicate exactly what that bottleneck was.
Assuming a given bottleneck without trying to remove the far more likely sources, is bad science.

On my system I was able to remove the most common bottlenecks in PSCS performance, Hard drive swap file speed and RAM capacity. This allowed the true power of the CPU cores to shine.

Why was going to four cores so much more of an impact?
Well in addition to adding TWO more cores instead of one when I went from 1 to 2 cores, I was also enabling broader RAM access as the 3rd and 4th cores were on the second CPU with it's own RAM controller and Hypertransport!

(if only these aging Opterons and the Hypertransport they run on were as fast as the recent intels, then I'd be cutting these speeds into 1/4 yet again! )

[tim]

Thanks for doing the tests guys, i'm interested to see 4 cores doesn't seem to much help. I'm still interested in converting from RAW directly to JPG, perhaps with a slightly larger sample size - make ten copies of the same RAW files perhaps? This would reduce the disk activity from the equation somewhat, since RAW and JPG files aren't really that big, but TIFF files are.

I don't think filters are a useful test, unless you personally use a lot of filters. 90% of my work is in ACR and batch converting images.

[CyberDyneSystems]

On my system going from two to four cores cut the time down by more than half..
How is that "doesn't seem to help much" ?

[bsmotril]

The CNet web site has a photoshop benchmark test they run on systems and post when they do a review of a system. You could also look there to get an idea how the various CPU's perform.

[BiikeMike]

I'd be happy to chime in on this test with my Quad Core Mac Pro, if anyone knows how to choose how to choose the number of processors on a Mac....?

[Steve Beck]

I'll try it when I get home with my dual quad core 3.2 mac pro. See if ps utilizes all 8 cores and how it affects everything.

[CyberDyneSystems]

8 cores... wow.. SkyNet is looking positively antediluvian !
I recall your 19 second speed test Steve, if it's doing that on less than all 8 cores, then jeeez man,. !

[BiikeMike]

Originally Posted by Steve Beck

I'll try it when I get home with my dual quad core 3.2 mac pro. See if ps utilizes all 8 cores and how it affects everything.

But can you assign programs to use different processors?

[Steve Beck]

Originally Posted by CyberDyneSystems

8 cores... wow.. SkyNet is looking positively antediluvian !
I recall your 19 second speed test Steve, if it's doing that on less than all 8 cores, then jeeez man,. !

I tweaked a little and got it down to 16 seconds. I have been trying on and off to get to 15 but can't.

[Moppie]

Originally Posted by CyberDyneSystems

On my system going from two to four cores cut the time down by more than half..
How is that "doesn't seem to help much" ?

I think he's talking about file conversion, from RAW to JPEG etc.
It appears ram and PageFile are more important than CPU cycles.

With your ultra fast pagefile, prehaps you could do a test on 1,2 and 4 core RAW conversion?

Originally Posted by BiikeMike

I'd be happy to chime in on this test with my Quad Core Mac Pro, if anyone knows how to choose how to choose the number of processors on a Mac....?

That would be awsome to see what sort of a difference going from 1 - 8 cores makes
There are some smart Mac users here, or might have to find a good Mac tech forum.

[CyberDyneSystems]

I'll have to try some timed batch conversions soon.
Trouble is, I don't have a clue how to run a batch in PSCS ACR?

But that leads me to the next question, do we even need to test this?

Is not PSCS ACR the same engine as Lightroom ACR?
And do we not KNOW that LR will use all the cores we can throw at it?

And if not, why are people using that bloated program? Bibble absolutely uses all four cores as did RSP, as does C1.

[CyberDyneSystems]

Moppie, what OS are you running?
Are you able to use all 4GB of your RAM?

[tim]

Originally Posted by CyberDyneSystems

I'll have to try some timed batch conversions soon.
Trouble is, I don't have a clue how to run a batch in PSCS ACR?

Drag a bunch of RAW files into photoshop, ACR will open, hit save down the bottom left, choose JPG, hit ok. Easy

[Moppie]

Originally Posted by CyberDyneSystems

Moppie, what OS are you running?
Are you able to use all 4GB of your RAM?

Vista Business, 32bit.
With a 512mb graphics card, I'm somewhat limited to how much of the 4gb I can see. Its currently showing 3325mb.

I think if I was building a dedicated photo processing machine I would use 64bit and as much ram as I could cram on the board.
But, my current machine also has to do everything else, including gaming, and I really can't be bothered with the potential limitations, or having a 3rd OS to dual boot.
At the end of the day a 32bit OS is more than fast enough for what I need given the current number of photos I process.

When you do the conversion test, might you have time to try under 32bit and 64bit XP? Be interesting to see how that effects things as well
(it might also distract skynet from its path to sentience.)

Originally Posted by tim

Drag a bunch of RAW files into photoshop, ACR will open, hit save down the bottom left, choose JPG, hit ok. Easy

Don't forget to "Select All" before hitting save