Forums | MacLife

gaaame fraaaameraaaaates....
muuuust seeeee mooooore beeeenchmaaaarks...

Got your frame rates here.
How does 268 for Quake 3 grab 'ya?

By the way, those MA Photoshop stats absolutely stink. I just don't know what they are doing, I ran a quick set on my G4 933/1.2 ram and I got:

Rotate - 3.6 secs            (7x faster than the G5 dual)
Convert - 25 secs            (equal to the Dual 1.4 G4)
Unsharp - 19 secs            (almost as fast as the 1.8 G5)
Gaussian blur (25) - 16.5    (neck and neck with the G5s)
Gaussian (1)        - 4.2           (crushed them all)

and i set my Photoshop memory useage low (500 megs) and my history states high (50) abd I ran the test sequentially so that I wasn't working in Ram but off my scratch disk (opposite what they did)

so I don't know what the heck they are doing.

And how big was this file that you used for this test?  The MA test file was 115mb, I believe.

The file was a 115mb psd, just like theirs.

The problem with using a large 115MB file in Photoshop testing, unless you reserve at least 690MB of memory just for Photoshop (6 times file size), you will be hitting the scratch disk.

Exactly. That is why I set my memory setting for 500mb and ran my tests with out a relaunch of photoshop between each test (like MA did). I was hitting the scratch disk and my times were hanging right in there with their results. I think they just have a screwy way of using P'shop for benchmarking.

I would,also, like to see how the results would stack up with a lot of ram in all the systems. That would reveal the processor and bandwidth/memory strengths.

my Celeron 700 spanked the dual 800 Powermac!. Something's terribly wrong with that chart.

I agree. My Pismo 400 shouldn't even be close or even in the same neighborhood.

The speedup attributed to running with 2GB ram vs 512MB doesn't look credible.

Well those are the kinds of speeds that I would expect in the first place, however.

However, the relative Photoshop performance of the 2GHz G5 dual vs the G4 1.42 MP shows about a 29% improvement. This is in good agreement with Barefeats.com and other Photoshop benchmarks posted elsewhere on the web.

I would expect quite a bit more than simply the difference in the clock speed of the two machines (30%) given the vastly improved architecture. MacWorld has posted tests that show a much bigger margin than that. benchmarks
I have a G5 Dual on order and let me tell you, if that box shows up at my house and is only giving me a 29% improvement in performance over a G4 1.4 dual for the $3300. I am spending (G5 dual with 1.5 gig ram), I am going to be seriously pissed off! 

Note to Macaddict: remove that chart before people start to refer to it elswhere...it's flawed.

Amen to that.

Note to Macaddict: remove that chart before people start to refer to it elswhere...it's flawed.

I stand behind the tests -- I performed them myself, using, as stated, a 115MB Photoshop file (a rather complex raw image from a high-end digicam), OS 10.2.7, Photoshop 7.0.1 with the G5 patch, repeating each test over and over and over and over again.  What you see is what happened.

Do keep in mind, however, that in the comparisons of the two G4s and two G5s, I was testing stock systems -- that is, ones with a mere 512MB of DDR SDRAM.  With that amount of memory, Photoshop is, of course, going to disk quite heavily -- which will, of course, depress performance. 

When I bumped the Dual 2GHz G5 up to 2GB of SDRAM, Photoshop flew -- not surprisingly, as Photoshop loves scads of memory.  I look forward to stuffing 2GB into the 1.8GHz G5 and seeing how it performs, and also incrementing the total SDRAM from 512MB to 1GB, to 1.5GB, to 2GB, to 2.5GB, and seeing where the performance boost kicks in.

Also, I look forward to running far more tests on the Dual 2GHz G5.  As I mentioned, we only received it in its stock configuration late Friday morning (and in the tricked-out config on Thursday) -- many more tests are to come, using not only Photoshop (with smaller and larger files), but also with a raft of other apps.  I look forward also to frame-rate comparisons of the Radeon 9600 vs. the 9800 (which, by the way, has its own fan, bringing the total number of fans in the Dual 2GHz G5 to 10).

One anomoly:  For some reason, the QuickTime testing chart in the Web story is wrong -- the performance of the 1.8GHz G5 should be noticeably slower.  Something skanky must've happened between the time I created that chart and when I gave it to Niko to post -- we were flying out the door late Friday afternoon, so I may have given him an incorrect chart.  I'll check it tomorrow at the office.

But as to the Photoshop testing...  Hey, if you can enlighten me about how you think those times might be wrong, I'd be more than happy to listen -- but, as I said, all tests were repeated many times, and all results were within repeatable range.

But can you explain (and make available) the file you used and provide some insight why the photoshop scores are so slow?

I can't make it available 'cause I don't have a copy here (it is, BTW, a single-layer RGB PSD) -- but I doubt it's the file.  It appears that your tests are running in RAM, and my tests on the 512MB-equipped "stock' Macs were definitely hitting the disk.

What OS are you running?  What version of Photoshop?

I removed a stick of 256MB ram and ran the trial with 512MB on the Celeron 700. Photoshop memory was set to use 80% ram and I had 6 history states enabled.

No appreciable change I'm afraid. In fact, it was slightly slightly better! (I shut down Bearshare.)

I would have been very surprised if Win98's memory management would be so poor that a 115MB file would use the scratch disk. I find it hard to believe that jaguar's memory management would be that poor.

It's got to be something else.

Here's the times with your first chart (512.jpg) as the source file and scaled up to 115MB and saved as  .PSD.

Rotate 90: 2.7 sec
RGB>CMYK*: 21.5 sec
Unsharp Mask (200% 5 pixel) : 19 sec
Gaussian Blur 1 pix: 7.0 sec
Gaussian Blur 25 pix: 20 sec

*RGB>CMYK seems to be quite dependent on the actual image content as I learned, and times are dependent on the colour characteristics of the particular file. Actual image content doesn't seem to effect the times of the other actions. But still...

I just realized that above I referred to my Pismo (G3 400/768 ram). I had first emailed MA with my test results and had included results from my Pismo with the same test. On much of the test I was taking twice as much time. But, this is on a G3 powerbook with 8meg Ati Mobility (I think) that should, by all rights be nowhere close. In tests that I have run between it and my G4 933, the G4 is 3x faster overall with a photoshop test suite that I got from OtherWorld Computing.

I find that hard to believe.

Same here, what the hell?

It isn't true, at any rate, as my G4 933 beats the tested rigs on parts of the test.

Unless...Macs really are slow

kidding

Where did you get that Celeron PC?

Here's something else to try. I use a benchmarking action that I got from Other World Computing. You place it in the actions folder for Photoshop and then just run it and time Photoshop until it is done.

I routinely get about 7min 20secs for the whole routine to run on my G4 933/1.2/60/nVidia GeForce 4 MX/SuperDrive/Zip/18gig Cheetah/Miles 2 SCSI host.

This is with Photoshop alone running and Pshops memory use at 500megs.

My Pismo 400 takes 19 minutes

My iMac 350/512 ram takes 21 minutes (how about that?)

All are running Pshop 7, OSX 10.2.6

Get the action and try for yourself.

This is a real world style test that creates its own file and then manipulates it. MA's test with this run a test, quit, relaunch, run another, quit etc just doesn't make any sense. That is not how Photoshop gets used, so what is the point?

This benchmarking action is described here at Other World Computing.

Ribtorus, lets see how the Celeron handles that.

Rik, how about you guys try again with this action? And take that crazy test result down. I swear, if that is even close to the truth, I am going to cancel my G5 dual order.


(let's be serious... )

yeah. I ran the test a couple of times on my 233 mhz G3/768 MB SDRAM and they were nowhere near the 933 mhz, but not too far from the G5. Oh, its so confusing...

Just for grins, I overclocked the celeron 700MHz to 874MHz.

with a 115MB .PSD file;

Rotate 90: 2.1 sec
RGB>CMYK: 16 sec
Unsharp Mask (200% 5 pixel) : 16.2 sec
Gaussian Blur 1 pix: 5 sec
Gaussian Blur 25 pix: 17.2 sec


PS7Bench time: 9 min 7 seconds.