[Marco Fanciulli]

Originally posted by squish:
the G5 was a lot slower on radial blur, water color, pointillize and accented edges. why?

Hi, Photoshop and other application developed with G4 and G4's Altivec in mind, have several coding techniques that are strongly inefficient on G5. If you can follow the program execution at low level, you can see that an high number of DST and ST derivative instructions are dispatched and the same can be said about DCBZ. The problem whith DCBZ is that it works on 32 byte chunks while the G5 operates on the entire 128 bytes cache with as much as four times the effort.

There are a number of other underperformant code lines and several branch misprediction which cause a drop down in performances.

You can read more about G5's performance dont's at http://developer.apple.com/technotes/tn/tn2087.html

Marco

[Chryx]

Originally posted by sparkplug:

Last I checked 64bit windows was alive and well.

64Bit IA-64 (read 'Itanium') is available, 64Bit x86-64 (read 'Opteron / Athlon64') has, as far as I'm aware, not entered beta stage yet, it certainly isn't available to buy.

so no, 64bit windows, in the context of the Opteron, is NOT alive and well.

[Chryx]

Originally posted by Hash:
Sorry to disappoint AMD opteron fans, but Opteron is not something we are discussing here. We are talking about Photoshop test.

And they were talking about how the Opteron would cope with Photoshop..


TBH, the current SSE2 implementation on the Opteron is somewhat weak (they've apparently improved it in the C stepping chips, but I've yet to see one of them benchmarked), and it can't be an x86-64 version of photoshop as of yet, which basically means I would expect it behave like.. well, a similarly clocked AthlonXP with a meg of L2 cache and a memory controller on crack.

AND __ People use Pentium/Athlon PC and Windows as well as Macs and OS 9/X for Photoshop work (probably, majority of pros Macs). I never heard about anyone working on Photoshop on Opteron server.

Opteron based workstations are available... some would argue that they're _as_ available as G5s are (although I'd argue that they aren't, but I'm in the UK, they seem to be fairly widespread in the USA), ergo it's valid to mention them in a discussion of graphics workstation performance.

Now, does anyone have access to an Opteron so they could throw PSBench at it?

[Bill Harrison]

Originally posted by stingerman:
The Opteron's would run slower right now since they could only run photoshop in 32-bit compatibility mode which is slower than its native mode. Opteron's are at Microsoft's mercy right now unles you are using them for Linux servers. $1000 / processor is pretty expensive too, so you'll not see one in a dual G5 desktop price range anytime soon; though they may call it a desktop but charge 5K for it.

This is not true. The Opteron runs 32 bit code faster than an equivilant Athlon XP speed wise. There is no "emulation" and it runs 32 bit apps just as fast as it does 64 bit.

[Chryx]

Originally posted by Bill Harrison:
This is not true. The Opteron runs 32 bit code faster than an equivilant Athlon XP speed wise. There is no "emulation" and it runs 32 bit apps just as fast as it does 64 bit.

You're mostly right, x86-64 adds registers, which means that in 64bit mode code needs to hit cache/memory less, which means that an Opteron running 64bit code is faster than it is in 32bit mode.

not because 64bit is inherently faster (because it isn't) but because the implementation of the Opteron is faster on 64bit code.

in 32bit mode it runs the code faster due to its memory controller and extra L2 cache, the core is effectively the same as an AthlonXP if you switch off the nifty stuffs.

[johneee]

I guess what it comes down to is that the top end chips of all the major vendors come out pretty close to the same... Just like it always has, with some shuffling here and there as to who is on top, but nobody really having enough of a lead to make it something that is best for everyone.

It's all just marketing. Bah Humbug I say.

[videian28]

someone needs to throw a dual 1.42 into that table

[legacyb4]

Arn,

Thanks. Having the highlights really makes it easy to get a feel for results.

Cheers.

Originally posted by arn:
Hope no one minds, I threw the numbers into a "real table" for ease of reading

http://www.chaosmint.com/benchmarks/...c-g5-ps7bench/

[skwerrl]

noone has even mentioned the gfx cards being used in this thread...

THAT can impact photoshop by a LOT

[Eug]

2x Xeon 3.06 Ghz 2 GB Ram, 533 Mhz bus,50 MB Test.

Photoshop 7.01Total Time: ~93.4 seconds.

Yeah, I have a tough time believing that a dual 2.0 will reach that Xeon time. I do believe the 119 seconds quoted elsewhere in this thread might be overly conservative though. So, my guess is that dual G5 will perform somewhere in between 93 and 119 seconds. How about a guestimate of 106?

Hi, Photoshop and other application developed with G4 and G4's Altivec in mind, have several coding techniques that are strongly inefficient on G5. If you can follow the program execution at low level, you can see that an high number of DST and ST derivative instructions are dispatched and the same can be said about DCBZ. The problem whith DCBZ is that it works on 32 byte chunks while the G5 operates on the entire 128 bytes cache with as much as four times the effort.

There are a number of other underperformant code lines and several branch misprediction which cause a drop down in performances.

It's not the same as designing filters from the ground up for the G5, but don't forget that the numbers posted are with the G5 optimization plug-in. I do suspect that both X.3 and Photoshop 8 will speed up the G5 even more, but we'll see those speeds when they're out. Right now OS X.2.7 with Photoshop 7.01 and the G5 optimizations will have to do.

One thing I'd be interested in is the numbers with and without the G5 optimizations, on both the G4 and the G5. ie. How much do the optimizations speed up the G5, and how much do they slow down the G4?

[DeathMan]

Originally posted by skwerrl:
noone has even mentioned the gfx cards being used in this thread...

THAT can impact photoshop by a LOT

Really? I didn't realize that the graphics card had much to do with PS performance. Forgive my ignorance, but how does having a better graphics card speed up Phostoshop filters? Does it actually speed things up, or just make them *feel* faster?

This is a serious question.

[ewoh24]

So forgive me if this is a stupid question, but with all this talk about GHz and FSB's etc...how bottlenecked is my poor 12" PB with a measly 867Mhz G4 with a paltry 133MHz FSB? I simply don't get it how a 867 MHz chip can only be fed 133MHz of data and NOT be considered severely strangled! (And that goes for all the 1+ GHz G4's runing 167MHz busses too...)

[3.1416]

Originally posted by Shaktai:
Funny thing is some folks in other forums are already quoting those as truth. I'll wait for actual tests on a real dual 2.0.

Ugh. I guess I should have added giant flashing disclaimers, but I thought it was obvious. Yes, I fully agree that a real dual 2.0 should have better numbers.

[3.1416]

Originally posted by Eug:
Yeah, I have a tough time believing that a dual 2.0 will reach that Xeon time. I do believe the 119 seconds quoted elsewhere in this thread might be overly conservative though.

So do I. Think of the extrapolated numbers as a lower bound rather than a prediction. I expect a real dual G5 to do better, and would be amazed if it did worse. Also, simply adding the times of each test results in a meaningless number because it's heavily skewed towards the more time-consuming tests.

[skwerrl]

a machine with abetter GPU and twice the VRAM would definitely do better in photoshop benchmarks, and i was wondering what gfx cards were used in the testing of the 1.6s becasue the standard config is with an nvidia geforce fx 5200 ultra, while an ATI radeon 9800 with a better chip and 128MB of DDR VRAM would easily blow the nvidia away.

and these cards also enhance 2D performance along with 3D

[Eug]

Originally posted by ewoh24:
So forgive me if this is a stupid question, but with all this talk about GHz and FSB's etc...how bottlenecked is my poor 12" PB with a measly 867Mhz G4 with a paltry 133MHz FSB? I simply don't get it how a 867 MHz chip can only be fed 133MHz of data and NOT be considered severely strangled! (And that goes for all the 1+ GHz G4's runing 167MHz busses too...)

The 12" PB 867 is potentially quite strangled yes. The TiBook 1 GHz (with the same 133 MHz bus) and the 17" PB 1 GHz (with 167 MHz bus), as well as the G4 PowerMacs, are actually less strangled though, because they have quite a bit of L3 cache to feed them. L3 goes a long way to help out, even though it doesn't eliminate the problems with a slow bus. (Note that though none of the G5s have any L3, because the memory is so fast.)

[ewoh24]

quote:

The 12" PB 867 is very strangled yes. The TiBook 1 GHz (with the same 133 MHz bus) and the 17" PB 1 GHz (with 167 MHz bus), as well as the G4 PowerMacs, are actually less strangled though, because they have quite a bit of L3 cache to feed them. L3 goes a long way to help out, even though it doesn't eliminate the problems with a slow bus. (You'll that though none of the G5s have L3, because the memory is so fast.)

I guess Im confused as to why any manufacturer (this is NOT directed just at Apple) could release a computer with those same stats. What percentage of my processor is just sitting there a majority of the time, waiting for data?
That said, I love my PB. Its absolutely perfect. But now I feel I need the speed of at least a Dual 1.42 G4 PowerMac. I can't seem to get the (mistaken, it seems) notion out of my head that a Dual G4 is at least as powerful as a single 1.6 G5. But it dosent make any difference, really, as the price of the G5 is so much more attractive.

[Durandalus]

Originally posted by 3.1416:
Hmm, let's try some extrapolation. Making the conservative assumptions that a single 2.0 G5 would be 10% faster than the 1.6, and that a dual G5 will scale the same as the dual Athlon, we get this:

Not bad, not great. Slightly faster than the P4, although they're up to 3.2GHz now, and we don't have scores for dual Xeons. Still it's good to be in the same ballpark.

Hmm. MHz-wise alone the single 2.0 GHz is 25% faster and not 10%. Plus it has a slighty faster memory system. Should be much faster than your projected numbers.

[littlegreenspud]

my psbench7 results:

0.7
3.0
2.7
0.8
3.0
4.3
1.2
3.4
3.9
0.9
35.3 (?)
0.9
6.2
6.6
7.5
22.0
19.4
49.3
5.3
35.8
3.7

Dual Gig QS 1.5G RAM Geforce 4Ti

[Eug]

Originally posted by littlegreenspud:
my psbench7 results:

35.3 (?)

Dual Gig QS 1.5G RAM Geforce 4Ti

Something seems fubar'd there. A dual 1.42 does it in something like 3 seconds. Even a revision A iMac with G4/500 upgrade card is much faster than your dual 1 GHz.

[gruhead]

Originally posted by stingerman:
[B]The Opteron's would run slower right now since they could only run photoshop in 32-bit compatibility mode.../B]

Excuse me, but Opteron's 32 bit mode is not *much* slower than it's native mode. The Opteron has 3 modes of operation, 32 bit mode, mixed 32/64 bit mode and pure 64 bit mode. All of these modes are in hardware and not emulated, so there is no performance penalty for mixing modes even.

The Opteron can and will run current 32 bit software just fine and also much faster than any current 32 bit CPU on teh market.

Also the $1000/processor price you quoted is just plain wrong. The Opteron 140, 1.4Ghz, is $265 right now and the high end Opteron 246, 2.0Ghz, is $800.

And, Boxx Technologies, a maker of high end gfx/video workstations, has been offering a dual Opteron workstation since June for ~$3000, which *is* within the G5 desktop price range.

And I'd like to add that Boxx Technologies cases are just as fashionable and functional as Apple's, so it looks great oo

If one were to take the time and effort to assemle a dual Opteron system ones self comparable to the high end G5 PM, I'm sure one to undercut Boxx's price by quite a healthy margin.

[littlegreenspud]

----------------------------------
Something seems fubar'd there. A dual 1.42 does it in something like 3 seconds. Even a revision A iMac with G4/500 upgrade card is much faster than your dual 1 GHz.
----------------------------------

Apart from that one strange result I think that the results are not too bad?

[Severed Hand of Skywalker]

Ok everyone who uses their computers for real work. Raise your hands if you spend your day throwing complex filters at giant Photoshop documents all day?

Or do you measure performance by how fast and smoothly you get your work done with tons of necessary apps open at once and switching between them?

[Eug]

Originally posted by Severed Hand of Skywalker:
Ok everyone who uses their computers for real work. Raise your hands if you spend your day throwing complex filters at giant Photoshop documents all day?

Or do you measure performance by how fast and smoothly you get your work done with tons of necessary apps open at once and switching between them?

I was wondering when you'd come in with that comment, since you do the same for every single benchmark thread.

The original poster has already said that performance of some of those filters are extremely important to him, mainly because he does batch actions on hundreds of images. He also says that some images will be several hundred megabytes and thus again, filter performance (eg. unsharp mask or RGB-CMYK) can be very important. I betcha that's one reason why Adobe was so quick to release the G5 plug-in. If performance didn't matter, all the Photoshop pros would still be perfectly happy with G4 450 power Macs and Adobe wouldn't bother with filter optimizations. Oh wait, don't you run a G4 450 Cube, and aren't you going to upgrade to a dual 2.2 GHz G5 rev. B?

Mind you, I agree a lot of the other filters are useless to the majority of the population. Some posters have taken to only reporting the first 12 filters, since everything else is irrelevant to them.

[Severed Hand of Skywalker]

Originally posted by Eug:
. Oh wait, don't you run a G4 450 Cube, and aren't you going to upgrade to a dual 2.2 GHz G5 rev. B?

Bzzzz, wrong. I use my cube for web work which is fine on a 450MHz cube. The problem is running lots of apps at once and switching between them. No filter optimizations will help that.

With the print work I do at work the Dual 1.25 is great for most things as the second prosessor really helps smooth things out. I cannot do most of the print work on the 450 Cube or even the 733 tower is slow for making PDFs.

If someone is doing batch work does it really matter that things are a little faster? I mean you set the machine up and go have lunch, big whoop if it takes an extra 5 minutes.

[action snake]

If someone is doing batch work does it really matter that things are a little faster? I mean you set the machine up and go have lunch, big whoop if it takes an extra 5 minutes. [/B][/QUOTE]


thank you for pointing that out. i don't know why people get all worked up over a couple of seconds or a minute. go smoke a cigarette you
impatient prick, and have a good day.

[Eug]

Well, I don't work with 500 images, but let's say there is a delay of say only 7 seconds per image. That's 3500 seconds, or about a hour. But I agree, for most people a couple of extra seconds per image is no big deal.

In my case it's video not Photoshop. I hate waiting for video to encode (because it takes forever).

[slipjack]

It adds up, snake, it adds up.

Time = Money.

[another_steve]

Originally posted by gruhead:
The Opteron can and will run current 32 bit software just fine and also much faster than any current 32 bit CPU on teh market.

No offense, but instead of adding more conjecture to the Opteron speed debate, how about providing some evidence. From what I've seen so far, the Opteron isn't exactly the steller performer some would like us to believe.

Here's some Opteron Photoshop performance numbers: Not exactly earth shattering, and certain not "faster than any current 32 bit CPU on the market" as you claim.

http://www.xbitlabs.com/articles/cpu...eron-1_15.html

Steve

[1_of_9]

skwerrl wrote....

Originally posted by skwerrl:
a machine with abetter GPU and twice the VRAM would definitely do better in photoshop benchmarks, and i was wondering what gfx cards were used in the testing of the 1.6s becasue the standard config is with an nvidia geforce fx 5200 ultra, while an ATI radeon 9800 with a better chip and 128MB of DDR VRAM would easily blow the nvidia away.

and these cards also enhance 2D performance along with 3D

I've got some interesting news for you. What graphics card that is used makes absolutely no differene in PSBench times...not even a fraction of a second, because re-draw times are not even in the benchmark. Thus, your assertion is flat out wrong.

[idyll]

Originally posted by 1_of_9:
skwerrl wrote....



I've got some interesting news for you. What graphics card that is used makes absolutely no differene in PSBench times...not even a fraction of a second, because re-draw times are not even in the benchmark. Thus, your assertion is flat out wrong.

I second that.

[Lars T.]

Originally posted by nil:
For months Boxx has been selling DP Opterons that run Win2K Pro and XP Pro. Opterons 32 and 64 bit code natively ergo any DP aware Windows OS will run them just dandy. Same as how the G5 runs non 64 bit OSX.

PSBench Scores for an Opteron 144. Total Time taken: 141.6s.

[another_steve]

Originally posted by littlegreenspud:
my psbench7 results:

Dual Gig QS 1.5G RAM Geforce 4Ti

We probably don't want this thread to turn into a PS_Bench thread, but for the sake of reference, here were my scores:

My System: Dual 1 Ghz G4 (Mirror Drive), GeForce 4 Ti (not that the video card matters for this test). My scores were recorded on an earlier build of Jaguar when the system was new.

Code:
                        Mine     Yours
Rotate 90+               0.6       0.7
Rotate 9+                2.7       3
Rotate .9+               2.5       2.7
Gaussian Blur 1+         0.7       0.8
Gaussian Blur 3.7+       2.4       3
Gaussian Blur 85+        3.5       4.3
Unsharp 50/1/0+          1.1       1.2
Unsharp 50/3/7/0+        2.8       3.4
Unsharp 50/10/5+         3.4       3.9
Despeckle+               0.9       0.9
RGB-CMYK+                3.8       3.8 ??
Reduce Size 60%+         0.8       0.9
Lens Flare+              5.6       6.2
Color Halftone-          6.3       6.6
NTSC Colors-             7.6       7.5
Accented Edges-         21.7      22
Pointillize+            18.9      19.4
Water Color-            49.8      49.3
Polar Coordinates+       4.9       5.3
Radial Blur+            35.6      35.8
Lighting Effects+        3.4       3.7
Sum                    179       184.4

Steve

[kupan787]

I threw in the Opteropn numbers taken from http://www.xbitlabs.com/articles/cpu...ron-1_15.html., and bolded fastest results.

Code:
                     Ath   Ath   P4     P4   Ath   2xAth  G5   2xOpt
Test                 3000  2700  3.06   2.8  2200  2200   1.6   244

1 Rotate 90           0.2   0.2   0.2   0.2   0.3   0.3   0.4   0.2
2 Rotate 9            2.8   2.9   2.7   3.3   3.5   2.4   2.7   2.4
3 Rotate .9           2.6   2.6   2.5   3.2   3.3   2.2   2.8   2.2
4 Gaussian Blur 1     0.8   0.8   0.8   0.9   1.1   1.0   0.6   0.7
5 Gaussian Blur 3.7   3.0   2.9   2.0   2.1   3.8   2.4   2.3   2.0
6 Gaussian Blur 85    3.4   3.5   2.3   2.6   4.6   2.9   3.5   2.3
7 Unsharp 50/1/0      1.0   1.1   0.9   1.2   1.3   1.0   1.1   0.8
8 Unsharp 50/3/7/0    3.1   3.1   2.1   2.4   4.0   2.5   2.8   2.2
9 Unsharp 50/10/5     3.1   3.1   2.2   2.4   4.0   2.5   3.4   2.2
10 Despeckle          2.6   2.7   2.2   4.0   3.3   1.9   0.8   2.9
11 RGB-CMYK           8.1   8.2   7.3   9.4  10.0   5.4   4.2   7.4
12 Reduce Size 60%    1.2   1.2   0.9   1.0   1.5   1.1   0.8   1.0
13 Lens Flare         3.9   4.0   2.5   3.8   4.3   3.1   6.0   4.2
14 Color Halftone     2.8   3.3   2.2   2.4   4.3   4.5   4.3   2.6
15 NTSC Colors        2.3   2.4   2.4   2.6   3.1   3.1   4.3   2.4
16 Accented Edges    10.4  10.9  10.9  12.0  13.7  13.9  16.2  12.1
17 Pointillize       17.5  17.7  12.1  18.8  21.3  12.0  25.0  20.1
18 Water Color       22.6  23.6  26.4  29.2  29.4  29.9  35.8  25.7
19 Polar Coordinates  8.0   8.3   7.0   8.5  10.2   6.1   4.9   8.0
20 Radial Blur       46.6  46.9  33.1  43.6  62.7  34.4  54.3  38.2
21 Lighting Effects   2.0   2.0   1.9   2.1   2.5   2.0   3.4   2.0

Total               148.0 151.4 124.6 155.7 192.2 134.6 179.6 141.6

Note I posted the dual scores, as the single was just slightly slower (141.6 for dual vs 143.8 for single)

[RooneyX]

Originally posted by kupan787:
I threw in the Opteropn numbers taken from http://www.xbitlabs.com/articles/cpu...ron-1_15.html., and bolded fastest results.

Code:
                     Ath   Ath   P4     P4   Ath   2xAth  G5   2xOpt
Test                 3000  2700  3.06   2.8  2200  2200   1.6   244

1 Rotate 90           0.2   0.2   0.2   0.2   0.3   0.3   0.4   0.2
2 Rotate 9            2.8   2.9   2.7   3.3   3.5   2.4   2.7   2.4
3 Rotate .9           2.6   2.6   2.5   3.2   3.3   2.2   2.8   2.2
4 Gaussian Blur 1     0.8   0.8   0.8   0.9   1.1   1.0   0.6   0.7
5 Gaussian Blur 3.7   3.0   2.9   2.0   2.1   3.8   2.4   2.3   2.0
6 Gaussian Blur 85    3.4   3.5   2.3   2.6   4.6   2.9   3.5   2.3
7 Unsharp 50/1/0      1.0   1.1   0.9   1.2   1.3   1.0   1.1   0.8
8 Unsharp 50/3/7/0    3.1   3.1   2.1   2.4   4.0   2.5   2.8   2.2
9 Unsharp 50/10/5     3.1   3.1   2.2   2.4   4.0   2.5   3.4   2.2
10 Despeckle          2.6   2.7   2.2   4.0   3.3   1.9   0.8   2.9
11 RGB-CMYK           8.1   8.2   7.3   9.4  10.0   5.4   4.2   7.4
12 Reduce Size 60%    1.2   1.2   0.9   1.0   1.5   1.1   0.8   1.0
13 Lens Flare         3.9   4.0   2.5   3.8   4.3   3.1   6.0   4.2
14 Color Halftone     2.8   3.3   2.2   2.4   4.3   4.5   4.3   2.6
15 NTSC Colors        2.3   2.4   2.4   2.6   3.1   3.1   4.3   2.4
16 Accented Edges    10.4  10.9  10.9  12.0  13.7  13.9  16.2  12.1
17 Pointillize       17.5  17.7  12.1  18.8  21.3  12.0  25.0  20.1
18 Water Color       22.6  23.6  26.4  29.2  29.4  29.9  35.8  25.7
19 Polar Coordinates  8.0   8.3   7.0   8.5  10.2   6.1   4.9   8.0
20 Radial Blur       46.6  46.9  33.1  43.6  62.7  34.4  54.3  38.2
21 Lighting Effects   2.0   2.0   1.9   2.1   2.5   2.0   3.4   2.0

Total               148.0 151.4 124.6 155.7 192.2 134.6 179.6 141.6

Note I posted the dual scores, as the single was just slightly slower (141.6 for dual vs 143.8 for single)

The G5 looks disappointing. Apple have been claiming over twice the speed for the G4 for a few years and the G5 alone can't do much better than a cheapo Peecee.

[3.1416]

Originally posted by RooneyX:
The G5 looks disappointing.

I don't see how the data supports that assessment. The dual Opteron does look disappointing; it loses to the single P4 (faster in 6 of the tests and slower in 10). Meanwhile even my pessimistic extrapolations for the dual G5 2.0 have it significantly faster than both.

[1_of_9]

[I thought a side by side gleaned from a very old thread (about 14 months
old) on Ars would be interesting to post here. By normalizing the scores
no one subset can skew the results. Tests 15 thru 20 are a useless waste of
time because virtually nobody will ever use them professionally. But -
they are a part of PSBench. Total summed times are not a good indicator
of performance IMO, because literally hundreds of ops in Photoshop are
not being used. "Real world" usage could be argued about ad nauseum. The
Dual G5 scores were posted July 26th but are using G4 optimized code (Not
the G5 Adobe plugin).

Ars linkage
http://arstechnica.infopop.net/OpenT...7760969205&p=1


Boxes are Dual Athlon 1700 and 2400 (1.47 ghz & 2.0 ghz) & G4 DP 1.42ghz,
G4 DP 1.25ghz, G5 1.6ghz, DP G5 2.0 ghz, and a DP 2.4ghz Xeons OC'd to
3.06ghz. I just found this forum and want to state some opinion here. I've
noticed many seem to believe Photoshop is a good benchmark app - I don't.
It simply does not stress the amazingly fast machines we have today. The
first two boxes I own - and never does CPU utilization exceed 80%, and
usally is only goes to 20-35% usage. Second, Photoshop does not do SMP
well - IMO. Someone here claimed 70% increases in speed. Perhaps on a
handful of filters that is true, but in my professional 12+ years of
experience - dual processors usually speed things up on average only 15-30%,
and that is from many, many tests I've done over the last 12 years on both
platforms. I left the mac platform in 1995 at both home and work, and just
started using one again about 6 months ago. I think the G5 looks wonderful
on paper, and I hope it is a hit for Apple. What I'm really curious about
is the G5's video capabilites, because that is one area that will really
stress the whole system. I apoligise in advance if this
formatting doesn't work out...

================================================== =====================
|||||||||||||||||||||2xAth||2xAth||2x G4||2x G4||Single ||2x G5||2x P4x
|||||||||||||||||||||1700+||2400+||1.42g||1.25g||G 5 1.6g||2.0g ||oc3.06
================================================== =====================
rotate 90---------------0.3----0.2----0.5----0.5----0.6----0.3----0.3
rotate 9----------------2.6----2.1----2.3----2.4----3.0----2.2----2.1
rotate .9---------------2.4----2.1----2.0----2.2----2.6----5.3----2.0
1 pixel gausian blur----1.0----0.9----0.6----0.7----0.8----3.2----0.6
3.7 gausian blur--------2.7----2.4----2.2----2.1----2.5----5.2----1.3
85 gausian blur---------3.3----2.9----3.1----3.3----3.5----5.1----1.6
50%/1/0 unsharp mask----1.1----1.0----0.9----1.1----1.3----4.1----0.6
50%/3.7/0 unsharp mask--2.9----2.4----2.5----2.6----2.9----4.6----1.3
50%/10/5 unsharp mask---2.9----2.5----2.9----3.1----3.6----6.3----1.3
Despeckle---------------2.2----1.8----0.7----0.7----0.8----4.4----1.3
rgb to cmyk mode change-8.2----7.0----2.6----3.0----4.3----6.4----5.3
60% Reduction-----------1.2----0.9----0.7----0.8----0.8----2.0----0.5
Lens Flare--------------0.6----2.9----4.7----5.1----6.1----6.7----1.4
Color Halftone----------5.0----4.3----4.8----5.4----4.2----6.1----2.2
NTSC Colors-------------2.9----2.3----5.3----6.1----4.2----7.6----2.9
Accent Edges-----------15.6---12.3---15.4---17.5---16.3---15.5---11.8
Pointilize-------------14.7---11.0---13.7---15.4---25.1---15.4----6.7
Watercolor-------------34.2---26.4---35.4---40.1---35.6---30.2---28.1
Polar Coordinates-------7.3----5.8----3.8----4.2----5.1----6.1----3.3
Radial Blur------------38.7---33.2---27.9---32.0---53.8---25.9---17.3
Lighting Effects--------2.2----1.8----2.7----3.0----3.5----6.1----1.3
================================================== ====================
TOTALS in seconds------ 155---126.2--148.4--151.3--180.6--168.7--93.2

Normalized scores.......
--------subset weighted 279----338----346----316----276----189----488

[Eug]

Fixed:

Code:
                   ||2xAth||2xAth||2x G4||2x G4||Single ||2x G5||2xXeon
                   ||1700+||2400+||1.42g||1.25g||G5 1.6g||2.0g ||oc3.06

rotate 90---------------0.3----0.2----0.5----0.5----0.6----0.3----0.3
rotate 9----------------2.6----2.1----2.3----2.4----3.0----2.2----2.1
rotate .9---------------2.4----2.1----2.0----2.2----2.6----5.3----2.0
1 pixel gausian blur----1.0----0.9----0.6----0.7----0.8----3.2----0.6
3.7 gausian blur--------2.7----2.4----2.2----2.1----2.5----5.2----1.3
85 gausian blur---------3.3----2.9----3.1----3.3----3.5----5.1----1.6
50%/1/0 unsharp mask----1.1----1.0----0.9----1.1----1.3----4.1----0.6
50%/3.7/0 unsharp mask--2.9----2.4----2.5----2.6----2.9----4.6----1.3
50%/10/5 unsharp mask---2.9----2.5----2.9----3.1----3.6----6.3----1.3
Despeckle---------------2.2----1.8----0.7----0.7----0.8----4.4----1.3
rgb to cmyk mode change-8.2----7.0----2.6----3.0----4.3----6.4----5.3
60% Reduction-----------1.2----0.9----0.7----0.8----0.8----2.0----0.5
Lens Flare--------------0.6----2.9----4.7----5.1----6.1----6.7----1.4
Color Halftone----------5.0----4.3----4.8----5.4----4.2----6.1----2.2
NTSC Colors-------------2.9----2.3----5.3----6.1----4.2----7.6----2.9
Accent Edges-----------15.6---12.3---15.4---17.5---16.3---15.5---11.8
Pointilize-------------14.7---11.0---13.7---15.4---25.1---15.4----6.7
Watercolor-------------34.2---26.4---35.4---40.1---35.6---30.2---28.1
Polar Coordinates-------7.3----5.8----3.8----4.2----5.1----6.1----3.3
Radial Blur------------38.7---33.2---27.9---32.0---53.8---25.9---17.3
Lighting Effects--------2.2----1.8----2.7----3.0----3.5----6.1----1.3

TOTALS in seconds       155---126.2--148.4--151.3--180.6--168.7--93.2

Normalized scores
(subset weighted)       279----338----346----316----276----189----488

So it's pretty clear that the non-G5-optimized code sucks, considering the single 1.6 with optimized code scores much higher (using the weighted normalized Ars scoring system). It's amazing what a little code tweaking can do.

By the way, your table includes 2xP4, which is of course, impossible. Maybe you mean 2xXeon?

EDIT:

Corrected 2xXeon part in tables. Missed the description in the first read thru.

By the way, I didn't really clean it up. It was properly formatted, but you just omitted the [code] function, which uses a font with fixed spacing.

[1_of_9]

That is what the x after p4 stands for...just was limited on characters. Plus I wrote "2.4ghz Xeons OC'd to
3.06ghz" - can't get much clearer than that. Thanks for cleaning the formatting up!

[Ysean]

Originally posted by skwerrl:
a machine with abetter GPU and twice the VRAM would definitely do better in photoshop benchmarks, and i was wondering what gfx cards were used in the testing of the 1.6s becasue the standard config is with an nvidia geforce fx 5200 ultra, while an ATI radeon 9800 with a better chip and 128MB of DDR VRAM would easily blow the nvidia away.

and these cards also enhance 2D performance along with 3D

I've *NEVER* heard of FPU functions ever being enhanced by a general purpose graphics card. Maybe this is heaven and everything influences performance? haha