[barbarian]

Here are some xbench numbers for the curious... the only test I couldn't run on panther was the processor test (fails on panther).

--------------
Results 138.11
System Info
Xbench Version 1.1.2
System Version 10.2.7 (G5) (6S80)
Physical RAM 2048 MB
Model PowerMac7,2
Processor PowerPC 970 @ 1.80 GHz
L1 Cache 64K (instruction), 32K (data)
L2 Cache 512K @ 1.80 GHz
Bus Frequency 900 MHz
Video Card GeForce FX 5200
Drive Type ST3160023AS
CPU Test 122.45
GCD Loop 75.07 2.93 Mops/sec
Floating Point Basic 204.79 740.60 Mflop/sec
AltiVec Basic 89.17 2.59 Gflop/sec
vecLib FFT 131.43 2.04 Gflop/sec
Floating Point Library 262.66 10.51 Mops/sec
Thread Test 81.00
Computation 59.78 806.97 Kops/sec, 4 threads
Lock Contention 125.61 1.58 Mlocks/sec, 4 threads
Memory Test 228.51
System 218.90
Allocate 461.76 301.21 Kalloc/sec
Fill 142.84 1136.97 MB/sec
Copy 220.36 1101.78 MB/sec
Stream 239.00
Copy 208.83 1526.57 MB/sec [G5]
Scale 206.49 1523.88 MB/sec [G5]
Add 273.78 1752.21 MB/sec [G5]
Triad 289.62 1769.59 MB/sec [G5]
Quartz Graphics Test 179.25
Line 201.02 5.12 Klines/sec [50% alpha]
Rectangle 177.21 12.47 Krects/sec [50% alpha]
Circle 192.19 4.43 Kcircles/sec [50% alpha]
Bezier 166.53 1.81 Kbeziers/sec [50% alpha]
Text 164.77 2.69 Kchars/sec
OpenGL Graphics Test 198.68
Spinning Squares 198.68 139.04 frames/sec
User Interface Test 166.77
Elements 166.77 53.64 refresh/sec
Disk Test 108.84
Sequential 114.46
Uncached Write 139.27 58.05 MB/sec [4K blocks]
Uncached Write 130.33 53.37 MB/sec [256K blocks]
Uncached Read 80.06 12.67 MB/sec [4K blocks]
Uncached Read 131.52 53.14 MB/sec [256K blocks]
Random 103.75
Uncached Write 95.59 1.43 MB/sec [4K blocks]
Uncached Write 107.57 24.26 MB/sec [256K blocks]
Uncached Read 99.45 0.66 MB/sec [4K blocks]
Uncached Read 114.43 23.55 MB/sec [256K blocks]
------------------------------
Results 149.78
System Info
Xbench Version 1.1.2
System Version 10.3 (Pre-release) (7B74)
Physical RAM 2048 MB
Model PowerMac7,2
Processor PowerPC 970 @ 1.80 GHz
L1 Cache 64K (instruction), 32K (data)
L2 Cache 512K @ 1.80 GHz
Bus Frequency 900 MHz
Video Card GeForce FX 5200
Drive Type ST3160023AS
Thread Test 85.17
Computation 57.98 782.80 Kops/sec, 4 threads
Lock Contention 160.36 2.01 Mlocks/sec, 4 threads
Memory Test 256.93
System 273.08
Allocate 515.04 335.96 Kalloc/sec
Fill 237.52 1890.65 MB/sec
Copy 206.87 1034.34 MB/sec
Stream 242.58
Copy 209.62 1532.35 MB/sec [G5]
Scale 212.78 1570.32 MB/sec [G5]
Add 277.92 1778.70 MB/sec [G5]
Triad 292.30 1785.98 MB/sec [G5]
Quartz Graphics Test 198.11
Line 195.55 4.98 Klines/sec [50% alpha]
Rectangle 178.49 12.56 Krects/sec [50% alpha]
Circle 187.11 4.31 Kcircles/sec [50% alpha]
Bezier 195.18 2.12 Kbeziers/sec [50% alpha]
Text 246.65 4.02 Kchars/sec
OpenGL Graphics Test 170.23
Spinning Squares 170.23 119.13 frames/sec
User Interface Test 245.23
Elements 245.23 78.88 refresh/sec
Disk Test 106.09
Sequential 110.25
Uncached Write 133.07 55.47 MB/sec [4K blocks]
Uncached Write 121.39 49.71 MB/sec [256K blocks]
Uncached Read 80.84 12.80 MB/sec [4K blocks]
Uncached Read 122.59 49.53 MB/sec [256K blocks]
Random 102.22
Uncached Write 93.98 1.41 MB/sec [4K blocks]
Uncached Write 101.77 22.95 MB/sec [256K blocks]
Uncached Read 101.11 0.67 MB/sec [4K blocks]
Uncached Read 113.98 23.46 MB/sec [256K blocks]

[ambush]

Again, why is OpenGL slower, Apple?

The last thing we need on Panzer is a slower OpenGL, really.

But the UI tests are very promising! I can't waitt till Panther!

[SMacTech]

Originally posted by ambush:
Again, why is OpenGL slower, Apple?

The last thing we need on Panzer is a slower OpenGL, really.

But the UI tests are very promising! I can't waitt till Panther!

I have ran several CineBench 2003 tests and OpenGL on Panther has been pretty much the same as 10.2.6

[CheesePuff]

Overall goes from 138 to nearly 150, not bad at all!

[barbarian]

you can't actually compare the big numbers because the panther numbers don't include processor tests...

[nforcer]

Originally posted by ambush:
Again, why is OpenGL slower, Apple?

The last thing we need on Panzer is a slower OpenGL, really.

But the UI tests are very promising! I can't waitt till Panther!

I'm not so sure OpenGL is/will be slower in general for 10.3. Some people have said that certain OGL functions have been very well optimized for 10.3, and I would expect the generally faster feel of 10.3 can be partially attributed to some OGL optimizations.

You have to realize that XBench isn't exactly the most accurate benchmarking tool. You can find some strange results if you look through the database. I seem to remember a report of a G4 generally whooping a G5! This would seem to indicate to me, unusual testing procedures, an XBench bug, or something else.

I'd say that testing procedures are the most important part, and often most overlooked part of the benchmarking process by the average user. Keeping the operating systems in-sync (unless comparing operating systems), the benchmarking software in-sync (unless comparing versions of the software), and keeping the settings as identical as possible (unless comparing the impact of a specific setting) on all of the test systems can make a world of difference when benchmarking. Since many people fail to document how they test, I cannot hold the tests they perform as an accurate representation of performance for many of the test cases involved.

Furthermore, the exact internal processes XBench is executing during testing, and algorithms it uses might be partial to certain systems (Altivec, for example). Although XBench does document things like Altivec, it would be equally useful to know exactly how the tests it does are executed. Since the source code is not available, no one can easily tell how tests are done, and what other variables or internal processes might be affecting things.

Conclusion: XBench is a fun way to see how certain functions generally differ in performance between machines, but I don't consider the results most users submit to be all that accurate. Nor do I consider the limited tests that XBench performs to be representative of overall performance when comparing test results it produces.

[barbarian]

FYI: my process was pretty simple.

Ran XBench under a clean user file (no library additions) and no other apps running except for the dock (finder was quit).

Same version of xbench. Same machine. Same HD, but different partitions.

Anecdotally speaking Panther doesn't feel faster than Jag on the G5 (not that it does feel speedier on my G3 powerbook and on my G4 500mhz). On the G5 don't notice any difference in the major apps I use (Adobe suite, Macromedia suites, Office, filemaker).

Also comparing framerates in OpenGL-heavy apps, Jag actually usually noses out Panther. Again same machine, same app...

Of course Panther isn't finished yet and usually the last round of tweaks that speed things up, but at least on the G5 I think those people expecting a huge speed bump in everyday functions will be disappointed.

[Gul Banana]

On the G5, do they need one? :)

[Catfish_Man]

Originally posted by barbarian:
FYI: my process was pretty simple.

Ran XBench under a clean user file (no library additions) and no other apps running except for the dock (finder was quit).

Same version of xbench. Same machine. Same HD, but different partitions.

Anecdotally speaking Panther doesn't feel faster than Jag on the G5 (not that it does feel speedier on my G3 powerbook and on my G4 500mhz). On the G5 don't notice any difference in the major apps I use (Adobe suite, Macromedia suites, Office, filemaker).

Also comparing framerates in OpenGL-heavy apps, Jag actually usually noses out Panther. Again same machine, same app...

Of course Panther isn't finished yet and usually the last round of tweaks that speed things up, but at least on the G5 I think those people expecting a huge speed bump in everyday functions will be disappointed.

App launch time and general UI responsiveness are massively improved. Posted from Panther.

[barbarian]

> App launch time and general UI responsiveness are massively improve

Well on my G5 app launch and UI responsiveness are about the same...

I do notice a big difference on a G3.

I think perhaps the problem is this.

If something took 10 seconds in Jag G3 it might take 3 seconds in Panther on a G3.

But that same thing that took 10 seconds in in Jag on a G3 might take 2 seconds in Jag on G5. While there might be an improvement in Panther that 2 seconds becomes 1 second which is hard to distinguish from 2 secs... so there is no great perceptual leap in speed. At least not for me.

[barbarian]

> App launch time and general UI responsiveness are massively improved (in Panther)[

Well on my G5 app launch and UI responsiveness are about the same in Panther an Jag at least as far as I can tell.

I do notice a big difference on a G3.

I think perhaps the problem is this.

If something took 10 seconds in Jag G3 it might take 3 seconds in Panther on a G3. Big Perceptual difference.

But that same thing that took 10 seconds in in Jag on a G3 might take 2 seconds in Jag on G5. While there might be an improvement in Panther that 2 seconds becomes 1 second which is hard to distinguish from 2 secs... so there is no great perceptual leap in speed. At least not for me.

[Anand]

Originally posted by barbarian:
> App launch time and general UI responsiveness are massively improved (in Panther)[

Well on my G5 app launch and UI responsiveness are about the same in Panther an Jag at least as far as I can tell.

I do notice a big difference on a G3.

I think perhaps the problem is this.

If something took 10 seconds in Jag G3 it might take 3 seconds in Panther on a G3. Big Perceptual difference.

But that same thing that took 10 seconds in in Jag on a G3 might take 2 seconds in Jag on G5. While there might be an improvement in Panther that 2 seconds becomes 1 second which is hard to distinguish from 2 secs... so there is no great perceptual leap in speed. At least not for me.

I agree with this completely. Even on a 1.6 Ghz G5, MS Powerpoint can be slow in converting metafiles and such - this is in 10.2.7. This is much improved in Panther. Many apps run better and faster in Panther.