What would be a good benchmark of system-wide Quartz performance? Bouncemarks?
<pure_speculation/>
Is a system-wide benchmark even possible for comparing two versions of an API? System-wide hardware benchmarks are somewhat feasible because there are relatively few possible bottlenecks to isolate. However, when benchmarking a huge API, weighting of each type of benchmark becomes a very subjective game.
Is a system-wide benchmark even possible for comparing two versions of an API? System-wide hardware benchmarks are somewhat feasible because there are relatively few possible bottlenecks to isolate. However, when benchmarking a huge API, weighting of each type of benchmark becomes a very subjective game.
This applies to hardware benchmarks as well. You always measure system performance - and this is massively influenced by the API, the machine and the compiler.
Comparing two versions of an API might even be the closest you can get to a meaningful comparison.
Hey. Does that mean us G3´ers (iBook 800 combo) are still very much in the optimation game?
I'd say so since both my 233 rev A iMac and my 400 indigo iMac are much Snappier? with Panther... (although some unfinished parts still take a while, but you can tell there has been a speedup)
G4 dual gig. This speed is not a hallucination. I've booted back into Jag several times and there is no doubt about how fast Panther is. Scrolling in Safari is just grab the thumb and up-down-up-down, no hesitation at all - I can't get the cursor more than 1/4 inch off the scroll thumb no matter how hard I yank it. Preview no longer has ANY hint of that lag when you drag pages around or scroll. It is flat-out instant. Finder windows open very fast. iDisk mounts very fast, no spinlock. Finder is threaded so that if there is a delay, you can still look at files, resize the window, etc.
Spinning cursor does not prevent you grabbing and moving the window. Neat.
Bugs: it can lock up the whole system on occasion; needs a hard reboot.
iChat windows (logs) scroll all the way to the top with no delay. Smooth - no stopping and spinning.
Panther still has some major bugs - mostly spinlocks when you do strange things, or interface glitches like disappearing folders in a list - or it gets confused about unmounting a volume, etc.: the usual alpha software things. I definitely do NOT recommend that anyone install it on their main partition.
Finder windows open very fast. iDisk mounts very fast, no spinlock. Finder is threaded so that if there is a delay, you can still look at files, resize the window, etc.
I take it then that the Finder was really rewritten from the ground up, plumbing and all? I was under the impression that when Steve at WWDC said that the Finder was new, he meant just the UI. Glad that's not the case.
that might be measuring the performance of the AppleScript implementation more than that of Quartz.
Quote:
Originally posted by Xidius
I ran a simple applescript on an 800mhz iBook 12" running Panther, and an 800mhz iBook 14" running jaguar. The apple script basically looked like this:
tell application "finder"
open folder "hard drive"
close window "hard drive"
open folder "hard drive"
close window "hard drive"
open folder "hard drive"
close window "hard drive"
-- repeat 100 times --
open folder "hard drive"
close window "hard drive"
open folder "hard drive"
close window "hard drive"
end tell
Jaguar: 33 seconds
Panther: 46 seconds
What's worse: The Jaguar's window was almost full screen, column view, and the Panther's was not even 1/6.
Both have 32mb vram, quartz extreme enabled, and nothing but script editor and finder running.
That isn't really a true test of Panther. There has been a substantial amount of new code injected into the Finder, so I doubt that the developer preview is indicative of the Finder performance in 10.3 final.
let1kwindowsbloom is a fair test (being plain windows), as is Preview's PDF scrolling etc. An AppleScript to open Finder windows is not, at least in terms of 10.3 final.
There has been a substantial amount of new code injected into the Finder, so I doubt that the developer preview is indicative of the Finder performance in 10.3 final.
How many times have we heard that before (debug code etc.)?
What's worse: The Jaguar's window was almost full screen, column view, and the Panther's was not even 1/6.
It is not an ideal test. Firstly, it involves AppleEvents which may not be optimized in the Developer Preview (who cares?). Secondly, it involves Finder which may still have unoptimized code, too. To correctly benchmark the graphic subsystem you should exclude an application from the test. Crazy? No. Make an Obj-C and Carbon apps which use only one command to show the window and only one command to remove it from the screen. On the same hardware the results of this test show the speed of primitive drawing routines, memory copying/moving efficiency and the window server's robustness. In general, it is these little things which will speed up or slow down every app. In fact, the window bloom test is indicative of low-level stuff optimization.
How many times have we heard that before (debug code etc.)?
He's not talking about debug code. He's talking about that new finder in OSX 'Phanter'. Being new code, it is not optimised for speed. Also the new finder has all sorts of new functionality, and has short cuts to all sorts of folders/idisks, is brushed metal, etc etc.
With time, this new finder will be optimised for speed like quartz has. However, the let 1K windows bloom test is a test of Quartz, the drawing layer, not the new finder.
The Applescript above opens 1000 new finder windows, and is thus a test of opening a new finder window, not just quartz.
Comments
Originally posted by Jonathan
Remember, that Let1KWindowsBloom test is not well optimized, itself..
How UNoptimized can Let1kWindowsBloom be? I would imagine they would have to work to make such a simple task/program be unoptimzed.
Originally posted by BRussell
What would be a good benchmark of system-wide Quartz performance? Bouncemarks?
<pure_speculation/>
Is a system-wide benchmark even possible for comparing two versions of an API? System-wide hardware benchmarks are somewhat feasible because there are relatively few possible bottlenecks to isolate. However, when benchmarking a huge API, weighting of each type of benchmark becomes a very subjective game.
Originally posted by dfiler
<pure_speculation/>
Is a system-wide benchmark even possible for comparing two versions of an API? System-wide hardware benchmarks are somewhat feasible because there are relatively few possible bottlenecks to isolate. However, when benchmarking a huge API, weighting of each type of benchmark becomes a very subjective game.
This applies to hardware benchmarks as well. You always measure system performance - and this is massively influenced by the API, the machine and the compiler.
Comparing two versions of an API might even be the closest you can get to a meaningful comparison.
Originally posted by Anders
Hey. Does that mean us G3´ers (iBook 800 combo) are still very much in the optimation game?
I'd say so since both my 233 rev A iMac and my 400 indigo iMac are much Snappier? with Panther... (although some unfinished parts still take a while, but you can tell there has been a speedup)
Originally posted by applenut
I don't think that's the best idea. OS X will abuse that little poor hard drive
No it won't.
A jump from 57 to 34 seconds to display bloom 1k windows is more like a 70% jump in speed, significantly more than 40%.
1000 windows / 57 seconds = 17,54 windows/sec
1000 windows / 34 seconds = 29,41 windows/sec
29,41 / 17,54 = 68%!
Just great
Originally posted by Anders
What HW are you using?
G4 dual gig. This speed is not a hallucination. I've booted back into Jag several times and there is no doubt about how fast Panther is. Scrolling in Safari is just grab the thumb and up-down-up-down, no hesitation at all - I can't get the cursor more than 1/4 inch off the scroll thumb no matter how hard I yank it. Preview no longer has ANY hint of that lag when you drag pages around or scroll. It is flat-out instant. Finder windows open very fast. iDisk mounts very fast, no spinlock. Finder is threaded so that if there is a delay, you can still look at files, resize the window, etc.
Spinning cursor does not prevent you grabbing and moving the window. Neat.
Bugs: it can lock up the whole system on occasion; needs a hard reboot.
iChat windows (logs) scroll all the way to the top with no delay. Smooth - no stopping and spinning.
Panther still has some major bugs - mostly spinlocks when you do strange things, or interface glitches like disappearing folders in a list - or it gets confused about unmounting a volume, etc.: the usual alpha software things. I definitely do NOT recommend that anyone install it on their main partition.
Finder windows open very fast. iDisk mounts very fast, no spinlock. Finder is threaded so that if there is a delay, you can still look at files, resize the window, etc.
I take it then that the Finder was really rewritten from the ground up, plumbing and all? I was under the impression that when Steve at WWDC said that the Finder was new, he meant just the UI. Glad that's not the case.
Originally posted by Xidius
I ran a simple applescript on an 800mhz iBook 12" running Panther, and an 800mhz iBook 14" running jaguar. The apple script basically looked like this:
tell application "finder"
open folder "hard drive"
close window "hard drive"
open folder "hard drive"
close window "hard drive"
open folder "hard drive"
close window "hard drive"
-- repeat 100 times --
open folder "hard drive"
close window "hard drive"
open folder "hard drive"
close window "hard drive"
end tell
Jaguar: 33 seconds
Panther: 46 seconds
What's worse: The Jaguar's window was almost full screen, column view, and the Panther's was not even 1/6.
Both have 32mb vram, quartz extreme enabled, and nothing but script editor and finder running.
- Xidius
let1kwindowsbloom is a fair test (being plain windows), as is Preview's PDF scrolling etc. An AppleScript to open Finder windows is not, at least in terms of 10.3 final.
Barto
Originally posted by Barto
There has been a substantial amount of new code injected into the Finder, so I doubt that the developer preview is indicative of the Finder performance in 10.3 final.
How many times have we heard that before (debug code etc.)?
Originally posted by Xidius
Jaguar: 33 seconds
Panther: 46 seconds
What's worse: The Jaguar's window was almost full screen, column view, and the Panther's was not even 1/6.
It is not an ideal test. Firstly, it involves AppleEvents which may not be optimized in the Developer Preview (who cares?). Secondly, it involves Finder which may still have unoptimized code, too. To correctly benchmark the graphic subsystem you should exclude an application from the test. Crazy? No. Make an Obj-C and Carbon apps which use only one command to show the window and only one command to remove it from the screen. On the same hardware the results of this test show the speed of primitive drawing routines, memory copying/moving efficiency and the window server's robustness. In general, it is these little things which will speed up or slow down every app. In fact, the window bloom test is indicative of low-level stuff optimization.
How many times have we heard that before (debug code etc.)?
He's not talking about debug code. He's talking about that new finder in OSX 'Phanter'. Being new code, it is not optimised for speed. Also the new finder has all sorts of new functionality, and has short cuts to all sorts of folders/idisks, is brushed metal, etc etc.
With time, this new finder will be optimised for speed like quartz has. However, the let 1K windows bloom test is a test of Quartz, the drawing layer, not the new finder.
The Applescript above opens 1000 new finder windows, and is thus a test of opening a new finder window, not just quartz.