ACR Speed Test Mystery

[Redcrown]

My Win7 desktop system is 5 years old. It's still adequate for Photoshop, but I spend a lot of time waiting for ACR conversions. So I bought a new system. The new system is way more powerful than the old one. But I'm discovering that it is not much faster, and in some tests is actually slower when doing ACR conversions.

This has me baffled, so I'm asking for help. Please review the following and offer comments on my testing procedure, or suggestions for other things to look at.

Test procedure: I made 10 duplicates of a raw file (Canon 5D3). Then I made several tests, each with a different set of ACR adjustments appled to all copies. From Bridge I would "Load files into Photoshop layers", and measure the time it took. Only Bridge and Photoshop were loaded during the tests.

I can't see any other Windows services stealing CPU time. Using Task Manager to monitor a test, I see the old system peaking at 88% CPU usage, while the new system peaks at only 30%.

The following is a table of my results.

http://kellyphoto.smugmug.com/photos/i-XsqKXSf/0/O/i-XsqKXSf.jpg


Note the incremental times for each separate adjustment type, but most important, note the times for noise reduction and sharpening.


Old System: I5 model 750 4 core processor, 8gb memory, 2 basic 7200rpm hard drives.

New System: I7 model 5820, 6 core processor, 16bg memory, 1 SSD and 1 7200rpm hard drive.

The systems have different graphics cards, but I've determined that the GPU has no impact on ACR conversions. I can turn GPU on and off inside Photoshop and I get the same results.

I've also run two Photoshop benchmarks I find via Google. Those tests do show significant differences between the systems. But those tests all involve running various filters via an action, or generating huge images that exceed available memory to test the scratch/swap file function. None of them test ACR conversion.

[Redcrown]

I have discovered the root cause of this mystery, yet that discovery has turned up another mystery.

I realized I was using different versions of ACR. I used Creative Cloud to load the new computer, so I got ACR version 8.7.1.311. My old system is one version back, ACR version 8.6.0.254.

So I stole ACR 8.6 from the old system and installed it on the new system. I just copied ACR. The new system is Photoshop 2014.2.2 while the old is Photoshop 2014.1.0.

Lo and behold, all my test conversions on the new system ran fast. Under ACR 8.6 the new system does ACR conversions 150% to 200% faster than under ACR 8.7. Good news, that's what I wanted and expected.

Here is a new table showing the results:



I can only conclude from this that there is something seriously wrong with the ACR 8.7 release. However, it gets more strange...

I loaded the new ACR 8.7 on the old system and ran a couple tests. They showed almost no difference. ACR 8.7 shows the same conversion speeds as ACR 8.6 on my old I5-750 system. Some conversions are even slightly faster.

Then I was pointed to this Adobe blog post:

http://blogs.adobe.com/lightroomjournal/2014/10/1366.html

Which says ACR 8.7 has "Improved performance on processors with AVX/AVX2 vector instruction support (Intel Sandy Bridge (2011) processors and later". I Googled "AVX/AVX2" and found it a bit beyond my pay grade, but it seems to be a likely explanation for my issue. My old system does not have that feature, my new system does have it.

I would greatly appreciate it if others would test the conversion speed of ACR 8.7 vs. ACR 8.6 on their system. Hopefully users with newer I7 based PCs. We need more evidence if ACR 8.7 is crippling some users.

[Erland]

What OS are you using on both computers?
And have you checked the Hz the CPU is using during the import from bridge?
Going from 4 to 6 cores isn't always an advantage.. And then you have Hyperthreading.

[Redcrown]

As in the OP, the operating system is Windows 7 SP1 on both systems.

On the new I7-5820 I have enabled/disabled hyperthreading. Disabling makes it slightly slower.
On the new I7-5830 I have used Task Manager and "Affinity" on the Photoshop process to kill 2 cores, leaving 4 active same as old system. Makes it slightly slower.

On both systems I have monitored CPU activity during conversions. All 4 cores are active on the old I5-750 system, peaking at about 85%. On the new system, 8 to 10 logical cores appear active, total CPU peaking at about 35%.

In other words, hyperthreading and number of cores do not appear to the problem. That leaves the AVX/AVX2 feature as the most likely suspect. I can't find a way to disable that in BIOS or inside ACR. That would be a good test if possible.

[Erland]

found this:

"I think you could use SDE http://software.intel.com/en-us/articles/intel-software-development-emulator to lower capabilities of your CPU. I haven't tried it though."

Perhaps the newer CC uses AVX instead of SSE4? Could be the problem. But still, the new CPU should be a big improvement.

[tony field]

Is it possible that the location of the raw cache files have something to do with the timing??  Maybe SSD, maybe different drive fragmentation, etc ??

[Redcrown]

I/O speed (SSD vs. HDD) will certainly effect speed of ACR conversions, but not much. I'm just guessing, but from experience and testing it looks like ACR conversion is 10% or less I/O and 90% or greater CPU processing. So a big improvement in I/O speed only impacts 10% of the total.

Regardless, in the tests I've shown above all files and cache were in the same location. The only difference is the version of ACR.

I'm getting depressed over this. I've posted here, on Adobe Forums, and on dpReview, and can't get a single user to duplicate my test to confirm or deny the findings. Submitted an Adobe bug report too. No response to that either.

If I'm right, everyone running ACR 8.7 on a PC made since 2012 is suffering.

[aidanw]

You could disable AVX and see if that is the difference:

https://randomascii.wordpress.com/2013/03/11/should-this-windows-7-bug-be-fixed/
(about 4 pages down)