HyperThread & Photoshop - The Test
Clyde
for me. Of course, I recommend that you do your own testing on your own
system for the way that you work in Photoshop.
Because I couldn't find any hard numbers to support a lot of claims, I
just ran my own tests to see if Photoshop ran faster with HT on or off.
These tests were run on my home-built computer. The motherboard is Intel
D865GBF. The processor it Intel P4 3.2 GHz Prescott - and HT, of course.
I have 2 GB of memory in 512 KB chunks of DDR400 using Dual Channel
Mode. This is Corsair CL 2.5 memory bought in pair, but not the
expensive Dual Channel tested stuff. I have a 120 GB Seagate (7200 rpm)
as my system drive and second Scratch Disc for Photoshop. I use a
Western Digital Raptor 36.7 GB 10,000 rpm drive for Window XP Pro's
Pagefile AND for Photoshop's first Scratch Disc. Both HDs are connected
by SATA 150.
I still use Photoshop CS. I haven't had a compelling reason to upgrade
yet. I'm also waiting for that first bug fix release; then I'll upgrade
regardless. I am using the > 1 GB plugin from Adobe.
My primary test was an action that included almost all the steps that I
use in my pro wedding photography business. This is somewhat my workflow
and a bit more. i.e. I never do all of this on any one picture. I do use
all of this on some pictures. I didn't include ACR in this because that
is usually interactive for me. My sharpening tool is Focus Magic and it
only works interactively. Here is a high level summary of my action;
default setting unless otherwise indicated:
Convert to RGB
Flatten Image
Run CurveMeister 2.0 - loading a setting
Create a Hue/Sat adjustment layer - set Saturation at 20% - Merge Down
Run Noise Ninja - using automatic noise profile
Run PTLens - this calls pano12.dll
Shadow/Highlights
Convert to 8 bit
Run digital background filters - Sprayed Strokes, Spatter, Lighting
Effects (not quite like I do backgrounds)
Save As - JPEG 2000 lossless - in different folder
Convert to B&W - using an action that does the 2 adjustment layers
method (See Dr. Brown)
Convert to Duotone - Tritone in my loaded setting
Save As - PSD
Close
Well, that's what's important to me. You are free to do what's important
to you. Of course, the filters are the big time eaters in this or any
action. Focus Magic can take longer than any of these - if a lot of
sharpening is needed. I doubt it would changed much though.
I also ran Contact Sheet II. I put 4 on a page, but that didn't take
very long to run. I then ran Dr. Brown's Image Processor 2.2 resizing
and saving them all to JPEG, PSD, and TIFF with LZW.
To use a more standard test I also ran SiSoftware Sandra Lite 2005.SR2a.
I ran the standard Combined Performance Index Wizard. For CPU that seems
to be a combination of Dhrystone and Whetstone. It did test CPU
Arithmetic, CPU Multi-Media, Memory, and Storage. I have it set to not
test Network, because I don't have a server to pound.
I did all this on nine (9) pictures. I picked a pretty good variety of
pictures that I regularly do. These were a variety of file formats,
color depths, color spaces, sizes, etc. The action ran for between 4 and
6 minutes for the action tests. I ran the tests multiple times, but it
didn't matter; there was no significant deviation.
I ran the tests on my computer with no other applications running with
both HT on and off. I left the standard boot apps running. So, my avast!
AV software and Sygate firewall were running, but not doing anything
that I could see.
Since HT's big benefit seems to be the ability to run multithreaded apps
and/or multiple apps at once. I also ran all the tests with a background
app running. Frankly, I don't have any app that runs consistently. I do
have Windows Media Player and iTunes. They run pretty consistently, but
not hard enough for testing. So, I used SiSoftware Sandra's Burn-In
Wizard to pound the two CPU burn-in tests. I ran these as Normal
Priority with the CPU level set at 33%. That was just my own gut feeling
of a realistic background level. Most of the time I don't have that much
running in background, but I do get spikes higher than that too.
Since the action test is most significant to me (and probably you in
this newsgroup), I'll start with those results....
With no background application running, my computer runs 7% faster with
HyperThread turn on. With the background app simulation running, my
computer ran 15% faster with HT turned on. This is just the Photoshop
action test.
The running of the Contact Sheet II showed no significant difference
between HT on or off. Part of the reason for this may be because 9
pictures wasn't long enough test. Ideally I should run this on many more
files, but it still may not be that big of a difference. The other
reason may be described below in the Sandra Memory and Storage results.
The Dr. Brown's Image Processor test the times were longer, but still
pretty short. Without a background app, there was no difference between
HT on or off. With a background app running, HT gave a 10% increase. I
don't think this is very significant though.
Sandra's Memory and Storage test results show that there is no
difference between HT on or off. Since HyperThreading is a CPU
utilization enhancement, it doesn't affect the the slower technologies
of memory and hard disks. So, the tests show no improvement.
Since Contact Sheet II and Dr. Brown's Image Processor are mostly doing
disk access (the slow part anyway), they shouldn't be much affected by a
CPU enhancement. They do some image resizing, but that appears to be
pretty efficient and doesn't change things enough to matter.
I did not run Sandra's benchmarking with a background app. For one
thing, it wouldn't let me run it's benchmark and burn-in wizards at the
same time. Then again, benchmark apps do take up all the tested
resources. So a background app doesn't make much sense.
Sandra's CPU Arithmetic benchmark is 21% faster with HT on. The CPU
Multi-Media benchmark runs 34% faster with HT on.
In no test did I find that anything ran slower with HyperThreading turn
on. Everything ran just as fast or faster with HT on. It seems to run
just as promised; multithreaded apps and multi apps running at the same
time can use efficiencies in the CPU allowed by HT.
It doesn't seem to help IO intensive processes, but it shouldn't either.
Then again, it doesn't hurt them either. If you have CPU intensive apps
- and who doesn't - HT can only help. Photoshop certainly seems to be
improved by having HT on. My gut feeling matches this. My experience
using Photoshop told me that it was most helpful when I had other
applications running at the same time. Since I usually do have other
apps running, it has been useful. It is good to finally have some hard
numbers to support that gut feeling.
Of course, this isn't the final, universal answer either. It is the
first hard numbers that I have seen - from someone other than Intel or
AMD. Don't forget that you are welcome to change anything in my testing
setup and run your own tests. I encourage you to do just that. Then post
your own results in this august newsgroup.
A comforted and happy user,
Clyde
Hunt
>
>This is the definitive test for Photoshop using HyperThread. Well, it is
>for me. Of course, I recommend that you do your own testing on your own
>system for the way that you work in Photoshop.
First, thanks Clyde. It seems that you have taken the time to address the
question of HT in real-world PS situations. While not covering every possible
variable, your tests seem to cover a wide gamut, and, as you said, applied to
what you do.
Again, thanks for your efforts, and for posting the results.
Hunt
Clyde
You are welcome. Glad to be of service.
Clyde
Hecate
>> First, thanks Clyde. It seems that you have taken the time to address the
>> question of HT in real-world PS situations. While not covering every possible
>> variable, your tests seem to cover a wide gamut, and, as you said, applied to
>> what you do.
>>
>> Again, thanks for your efforts, and for posting the results.
>>
>> Hunt
>>
>
>You are welcome. Glad to be of service.
>
Yes, let me second that - it was interesting. :)
--
Hecate - The Real One
Hec...@newsguy.com
Fashion: Buying things you don't need, with money
you don't have, to impress people you don't like...
Clyde
Of course, Hecate. You were the one I most trying to impress. Well, we
were the start of the discussion.
Thanks,
Clyde