[Hash]

Perhaps, a too simple and basic question.. Yet, its remarkable that despite adding so many new features, Apple managed again to greatly improve OS X. Jaguar, especially 10.2.8, already was very fast, yet Panther is so much faster than Jaguar in almost all respects. Compare to Microsoft - it always managed to do the opposite - think how XP is slower and bloated compared to Windows 98 or even 2000.

For XP, one culprit is usually thought to be its excessive visual features. Yet, OS X still has the Quartz Extreme, Aqua (although subdued greatly) and has hundreds of new visual and other additions and still manages to be so much faster. It would be interesting to know which parts of the OS X were improved most and how.

[Hash]

And one more thought: if this trend of faster OS X releases continues for few years (and why it should not?), then by the time MS releases Longhorn in 2006, we will have a lightning fast Mac OS 10.5 or 10.6 and incredibly slow and bloated Longhorn, resembling 10.0 + all MS crap

[DaGuy]

I wouldn't count on it. There's always room for some squeezing some additional optimization improvements but these will decrease with time and I'd be surprised if Apple continues to significantly speed up the MacOSX after a couple of more releases. Nevertheless, will we beat the crap out of longhorn? Guaranteed!!

[Hash]

Do you imply that after Panther, we should not expect such significant increases in speed and the room for further optimization is much more limited, say for 2004? Or you mean that for further 2005-2006 releases? And, of course, beating Longhorn is guaranteed. Heck, i think that even now XP is slower than OS X 10.3 on comparable hardware and perhaps certainly slower on older machines (with speed of <1 Ghz for macs and pentiums III-original MMX).

[awaspaas]

They've been optimizing the code like mad ever since 10.0 was out. And I'm sure there's a little room to go yet, but probably not much. I think the increases in speed will level out pretty substantially in the next year or two, and then it'll probably start getting slower with new features. Sort of like how every classic OS after 7.5.3 got a little bit slower, if I recall correctly.

Of course this is offset by the inception of faster computers, and those people who get a new one every 3-4 years like me should be just fine.

[Hash]

I am not worried by the possible falling marginal acceleration rate of the OS X after 1-2 releases because already it is very fast for me. Further tweaking is welcome, but is not vital, since hardware improvement will certainly make feel much faster on newer machines even without much further optimization. Still, its impressive how they managed to optimize OS X by such a large factor in just 3 releases, from 10.0 to Panther, even compared to OS 9. What was the secret of the success? What parts were changed most? thats intriguing.

[Love Calm Quiet]

Just contemplate the effect of code optimized for G5! I'll be there's plenty of room for that still.

[Judge_Fire]

Squeezing speed out of apps seems to be a guideline at Apple nowadays.

You know the story of how the Safari team could add any new features, as long as the app speed didn't decrease? If you really wanted to squeeze something in, you'd perhaps need to optimize someplace else

J

[Hash]

Yeah, there always will be plenty of room for optimization for newer hardware..G5 and PPC 980 (?) or G6 and who knows beyond it..

[DaGuy]

Originally posted by Hash:
Do you imply that after Panther, we should not expect such significant increases in speed and the room for further optimization is much more limited, say for 2004? Or you mean that for further 2005-2006 releases? And, of course, beating Longhorn is guaranteed. Heck, i think that even now XP is slower than OS X 10.3 on comparable hardware and perhaps certainly slower on older machines (with speed of <1 Ghz for macs and pentiums III-original MMX).

See Awaspaas post right below yours. It's right on.

[Hash]

Yes, I read it, but you both forget that that hardware will evolve as well.. there always will be possiblity to optimize more because the hardware itself will be changing continously. Its not that we wont see anything new after G5. And it means that limits for optimization will move and move together with new generations of hardware

[wadesworld]

That should be obvious!

They removed the debug code!

[willed]

Who knows, by 10.9, widows might even re-size half-decently...

[Uday's Carcass]

Originally posted by willed:
Who knows, by 10.9, widows might even re-size half-decently...

ever think it might have something to do with the hardware you run it on? if you're using that ibook 500, even with that amount of RAM, you're just begging for reduced performance.

[arekkusu]

There is still a lot of room for optimization. Particularly as programmable GPUs filter down to low end hardware (Radeon 9200 in the new iBook as an example.)

Leveraging more of Quartz _rendering_ of paths, fonts, etc to the GPU could have a substantial effect on overall speed. As a simple example, drawing a group of antialiased line segments on an old Radeon 7500 is about 100x faster than software Quartz.

There are also substantial gains to be made from auto-vectorizing technology going in to new versions of GCC. Won't help G3 machines, but apps recompiled for G4 and G5s will get even faster.

[RooneyX]

Originally posted by willed:
Who knows, by 10.9, widows might even re-size half-decently...

And the Finder might remember some settings too! LOL

[icruise]

Of course if 10.0 (and 10.1 to a lesser extent) hadn't been such a dog, I don't think we would be as impressed with Panther... But, yes, it is very nice.

[Charles Bouldin]

One thing that I think will allow some optimization of future releases is if they start using the IBM xlc compiler instead of gcc. I know that can't be done everywhere...xlc knows nothing of objective C, but there are still many areas where it could be applied (if it isn't already).

[ryaxnb]

Originally posted by wadesworld:
That should be obvious!

They removed the debug code!

Actually the original 10.0 actually HAD debug code!

[nobitacu]

I don't much care about how they got Panther to be this fast of a OS, as long as it is fast, I'm happy.

Ming

[piracy]

Originally posted by ryaxnb:
Actually the original 10.0 actually HAD debug code!

Actually NO, it didn't. This whole f*cking "debug code" thing is a misnomer. Every OS has tools for debugging various pieces; but here, people mean it to be "extra code that slows down the OS". But there is not now, and never has been, any SUPERFLUOUS CODE for debugging in ANY external release of Mac OS X, ever, INCLUDING prerelease builds.

[fireside]

Originally posted by piracy:
Actually NO, it didn't. This whole f*cking "debug code" thing is a misnomer. Every OS has tools for debugging various pieces; but here, people mean it to be "extra code that slows down the OS". But there is not now, and never has been, any SUPERFLUOUS CODE for debugging in ANY external release of Mac OS X, ever, INCLUDING prerelease builds.

its a long running joke idiot. just like the whole Snappy™ thing.

[piracy]

Originally posted by fireside:
its a long running joke idiot. just like the whole Snappy™ thing.

Uh, I know it is, "idiot". I've been here for longer than you. Except this guy wasn't joking. He was referring to 10.0, and I'll guarantee you his post was serious.

[Spheric Harlot]

Originally posted by piracy:
Uh, I know it is, "idiot". I've been here for longer than you. Except this guy wasn't joking. He was referring to 10.0, and I'll guarantee you his post was serious.

So the more to-the-point reply would have been:

"So does every other OS on the planet.

The point is whether it in any way affects performance, which in OS X, it does not, nor has it ever, even in pre-release builds."

No?

-s*

[King Bob On The Cob]

Originally posted by piracy:
Actually NO, it didn't. This whole f*cking "debug code" thing is a misnomer. Every OS has tools for debugging various pieces; but here, people mean it to be "extra code that slows down the OS". But there is not now, and never has been, any SUPERFLUOUS CODE for debugging in ANY external release of Mac OS X, ever, INCLUDING prerelease builds.

So what there is no
if(!root)
NSLog @"Error in this process. You are not logged in as administrator."
?

That would be debug code. If there was no question that the person would be logged in as root this code would not be there and therefor not evaluated saving time but it is. So there IS debug code and there always will be.

[ZackS]

What made 10.0 so slow was mostly that Quartz generated and parsed full XML PDF files for display of every frame. In 10.1 they switched to binary PDF and 10.2 they optimized it and rewrote the window server to composite using Open GL. 10.3 has even more aggressive optimization of this code. That's the main reason why it's faster. In addition, the Finder's networking code seems to be threaded much more cleanly.

That's basically why Mac OS X has been getting faster.

[JB72]

I know it's a joke now, but 10.3 has the best snap factor yet.

[brachiator]

Originally posted by JB72:
I know it's a joke now, but 10.3 has the best snap factor yet.

Has "snap factor" been used before? You might just have resucitated the franchise, if not!

Originally posted by piracy:
Actually NO, it didn't. This whole f*cking "debug code" thing is a misnomer. Every OS has tools for debugging various pieces; but here, people mean it to be "extra code that slows down the OS". But there is not now, and never has been, any SUPERFLUOUS CODE for debugging in ANY external release of Mac OS X, ever, INCLUDING prerelease builds.

Piracy, you've identified a new linguisti-ontological species: the "myth-nomer," if you will. Giving the wrong name to something that doesn't exist!

[Catfish_Man]

Originally posted by ZackS:
What made 10.0 so slow was mostly that Quartz generated and parsed full XML PDF files for display of every frame. In 10.1 they switched to binary PDF and 10.2 they optimized it and rewrote the window server to composite using Open GL. 10.3 has even more aggressive optimization of this code. That's the main reason why it's faster. In addition, the Finder's networking code seems to be threaded much more cleanly.

That's basically why Mac OS X has been getting faster.

The new compiler helps too, as did the new vector math libraries in Jag, and the vector image library in Panther (I have to assume they're using it).

[sodamnregistered2]

Originally posted by JB72:
I know it's a joke now, but 10.3 has the best snap factor yet.

The Snap stops here.

[MadBrowser]

One area that received a lot of attention was text services. They optimized the text layers individually and as a group...

Since pretty much every application uses text, this helped quite a bit.

The WWDC Panther sessions go over this and some of the other stuff.

[ZackS]

Originally posted by Catfish_Man:
The new compiler helps too, as did the new vector math libraries in Jag, and the vector image library in Panther (I have to assume they're using it).

There's also something insane going on with Quartz texture caching. In Jaguar there used to be a tiny delay when opening in which Window server would jump to about 100% CPU. Panther only jumps to about 24% or less. It also rarely rises above 5% under normal conditions on a non-QE machine.

[ngrundy]

Originally posted by King Bob On The Cob:
So what there is no
if(!root)
NSLog @"Error in this process. You are not logged in as administrator."
?

That would be debug code. If there was no question that the person would be logged in as root this code would not be there and therefor not evaluated saving time but it is. So there IS debug code and there always will be.

Actaully no, that would be extremely bad privlige checking code.

something like
if(getuid() == 0)

would be more like what you want for a start, as checking based on username is bad as the username can be changed, the UID is what identifies the user as root not the name. You can rename the root account to billybob and as long as you log into the system as billybob with the correct password and that account has the uid of 0 then you have root privliges.

If you want to see debug code in action grab a copy of freebsd and compile it with WITNESS and INVARIANTS then watch an os crawl to a screaming halt.

[a2daj]

According to an Apple engineer, with 10.3 we finally get true Quartz hardware acceleration on top of OpenGL. So both Quartz and Quartz Extreme get hardware acceleration. That makes a huge difference in GUI responsiveness.

[khufuu]

Originally posted by sodamnregistered2:
The Snap stops here.

Wouldn't that mean that Panther is ... snapilicious!!

[sniffer]

Originally posted by a2daj:
According to an Apple engineer, with 10.3 we finally get true Quartz hardware acceleration on top of OpenGL. So both Quartz and Quartz Extreme get hardware acceleration. That makes a huge difference in GUI responsiveness.

Wow!

I would like to hear more about how acceleration works on none QE hardware.
If anyone could explain how this works I would appreciate it. I know accelerated scrolling have been discussed as a possibility on the web, but full acceleration?

Edit: Or perhaps you only ment Quartz has gotten true acceleration on QE enabled machines?

[Luca Rescigno]

I think one major reason Panther feels faster is because the animations are rendered quicker. Before, even if your computer was far faster than necessary to render a sheet animation at full speed, it would still take a certain amount of time for that sheet to drop down so you could start using it. Now, that animation is programmed to take less time so it's faster on ANY computer. That applies to many animations - drawers, iChat, sheets, and the like.

One reason OS 9 feels so incredibly fast is that there seems to be zero time between when you activate something and when it happens. In OS X, there is often an animation associated with the action. Even if the action is over before you can even react to the change, it still looks slower than OS 9 where closing a window is instant.

[Mr Scruff]

Originally posted by Luca Rescigno:
I think one major reason Panther feels faster is because the animations are rendered quicker. Before, even if your computer was far faster than necessary to render a sheet animation at full speed, it would still take a certain amount of time for that sheet to drop down so you could start using it. Now, that animation is programmed to take less time so it's faster on ANY computer. That applies to many animations - drawers, iChat, sheets, and the like.

One reason OS 9 feels so incredibly fast is that there seems to be zero time between when you activate something and when it happens. In OS X, there is often an animation associated with the action. Even if the action is over before you can even react to the change, it still looks slower than OS 9 where closing a window is instant.

This is very true. A similar effect can be observed on OS 9 by setting the menu blink preference to 1 - instant subjective performance boost.

[moonmonkey]

Originally posted by Luca Rescigno:
I think one major reason Panther feels faster is because the animations are rendered quicker. Before, even if your computer was far faster than necessary to render a sheet animation at full speed, it would still take a certain amount of time for that sheet to drop down so you could start using it. Now, that animation is programmed to take less time so it's faster on ANY computer. That applies to many animations - drawers, iChat, sheets, and the like.

I agree to some extent,
But how come I can now open a folder with 300 items in it and scroll around without lag.
There is no flashy animation involved in this.

Also menus are move responsive.

Booting is faster.

Its much much more than a few frames being dropped in the animation.

Everything is faster.

[moki]

Originally posted by ZackS:
What made 10.0 so slow was mostly that Quartz generated and parsed full XML PDF files for display of every frame.

hrm, I'm not sure where you got this from, but it isn't accurate. XML PDF files were not generated for each frame.

There wasn't any one single bottleneck in any version of OS X, but rather a whole host of things that could be improved. OS X when it was in the planning stages was a technology platform that embraced many interesting architectural choices that were looking several years down the road.

Optimizing parts of the kernel helped. Optimizing parts of the display system helped. Using better compilers helped. Plain old profiling and optimizing bottlenecks across the board helped. Abandoning inefficient algorithms in favor of better ones helped.

There are still plenty of things that can be optimized in one way or another, and I expect that to some extent, they will continue to be.

In any event, many of the forward-looking choices Apple made years ago are beginning to bear fruit, such as the Quartz imaging model. Giving each window its own backing store caused some fairly excessive slowdowns compared to OS 9, because everything had to be effectively drawn twice. Now with hardware acceleration, the overhead of doing this has become fairly small. Apple certainly expected to accelerate this all along, it's just a matter of when.

Still, more things can potentially be accelerated. The text you see on the screen, for instance, is drawn in software, not via the video card. When programmable video cards become mainstream and feature-full enough, it's possible that the code that renders Quartz text can be offloaded.

There are massive inefficiencies in just about every type of software. You could spend decades optimizing anything of reasonable complexity such as an operating system, but you also have to balance adding new features/architectures.

[bonaccij]

Just going over all of these posts, it seems like everyone has forgotten that the reason all of the animations (including scrolling through a list of 300 items in your finder) are faster is because of the new rendering engine from Pixar called Pixlet. I have noticed ridiculous speed gains in simple animations as well as Finder level. In addition, I am sure that the journaled file structure plays an important role as far as caching data.

[moki]

Originally posted by bonaccij:
Just going over all of these posts, it seems like everyone has forgotten that the reason all of the animations (including scrolling through a list of 300 items in your finder) are faster is because of the new rendering engine from Pixar called Pixlet. I have noticed ridiculous speed gains in simple animations as well as Finder level. In addition, I am sure that the journaled file structure plays an important role as far as caching data.

Pixlet isn't a rendering engine, it is a QuickTime codec for compressing high-quality video, and it isn't particularly fast as far as codecs go, but it does some very cool stuff.

The journaled file system doesn't increase performance, indeed there is a slight overhead on write operations. It just keeps a ledger of changes to the disk before they happen, so in the event of a crash, the file system can be returned to a known state.

[Arkham_c]

Another future area of improvement is the compiler used by Apple to compile the OS. Right now the OS is compiled with GCC, which is not a very efficient compiler on PowerPC compared to Metrowerks or The IBM compiler.

When IBM gets their compiler ready for commercial release, it's possible that Apple might compile with it and make the OS significantly faster right there.

It's also likely that Apple might devote resources to making GCC generate faster code on PowerPC/OSX so that we all might benefit from the improvements. I know Apple has submitted some GCC patches in the past that were rejected, so it's apparent that this is an area of interest to them.

[freakboy2]

from using panther on a DA 533 with a geforce 4 ti installed and an imac 800 with a gf2 mx, I'd say that the snappiness is definitely due at least in part to whatever acceleration the video card is lending.

the 533 DA seems faster than the imac. i'd be interested in hearing from apple more details on what htey did to speed things up.

one thing i'll say: my father after getting panther decided he didn't need a dual g5 because of the speed boost.. Oopps.. sorry apple, on less sale. Anyway, panther rocks.

go apple!
fb

also - this post started by bashing the speed of xp. I really don't think you used 98 and xp side by side. In terms of speed on a machine with enough ram > 256 megs, the different in speed is staggering, with XP kicking 98,2000s's respective asses. (my experience is with p3's. I found xp to be like getting a new cpu)

[Drizzt]

Originally posted by freakboy2:
from using panther on a DA 533 with a geforce 4 ti installed and an imac 800 with a gf2 mx, I'd say that the snappiness is definitely due at least in part to whatever acceleration the video card is lending.

the 533 DA seems faster than the imac. i'd be interested in hearing from apple more details on what htey did to speed things up.

one thing i'll say: my father after getting panther decided he didn't need a dual g5 because of the speed boost.. Oopps.. sorry apple, on less sale. Anyway, panther rocks.

go apple!
fb

also - this post started by bashing the speed of xp. I really don't think you used 98 and xp side by side. In terms of speed on a machine with enough ram > 256 megs, the different in speed is staggering, with XP kicking 98,2000s's respective asses. (my experience is with p3's. I found xp to be like getting a new cpu)

Go play with anything lower than a P3, or a P4 lower than 2Ghz.. and you'll think again.

I'm typing this on a P4m 1.8Ghz, and I did some tests with it compared to my 400Mhz PowerBook G4. My PowerBook is faster playing Ghost Recon, encoding AACs with iTunes and generally over the interface. Of course, with well optimised programs the P4 runs circles arround my G4.. but it is still quite surprising to see how difficult it can get to toast it..

[Hash]

Of course, I used XP and 98 side by side, I had Vaio notebook with Win98, and now I am using an Athlon XP with 768 mb of RAM and GeForce Ti 4200, as well as aging Mac G4 with Panther on 640 mb of ram and an 2003 year model of Sharp notebook with Xp and 256 mb of ram and I _can_ judge their comparative performance quite objectively. I did not run any benchmarks but I do real work on them (except Vaio, which I sold) and that is the real benchmark, if you like. Panther on aging Mac G4 is incredible in terms of speed and performance, I am truly impressed and moreover - amazed.

[stew]

Originally posted by Uday's Carcass:
ever think it might have something to do with the hardware you run it on? if you're using that ibook 500, even with that amount of RAM, you're just begging for reduced performance.

Eh..what exactly is so complicated about resizing windows that the ~800MIPS of the CPU are not enough? Considering the frame rate, resizing windows in OS X must be about 6-10 times more complicated than playing a DivX movie.

[Anand]

Has anybody notice that when you use the menu bar the clock stops? Do little things like that save CPU time for other things? How may other "little cheats" did apple do for panther.

[ZackS]

Originally posted by moki:
hrm, I'm not sure where you got this from, but it isn't accurate. XML PDF files were not generated for each frame.

Of course a new one isn't generated for each frame, but that's the way Quartz worked. It would parse XML.

[moki]

Originally posted by ZackS:
Of course a new one isn't generated for each frame, but that's the way Quartz worked. It would parse XML.

So you're saying that it'd take the Quartz routines your application calls, spit out XML, and then parse it to do the drawing operations?

I find no evidence to support this.