Snow Leopard's Grand Central and OpenCL Details?

[pohl]

NO. Please do some research before posting. Snow Leopard is for INTEL CPU'S ONLY. The way it SHOULD be.

AFAIK, the only evidence to this effect is that the Developer Preview is intel-only. But, recall, the first Developer Preview of Leopard was also intel-only, and here I am running it on my PowerMac.

Unless you have some specific, credible statement by Apple that you would like to cite, perhaps you should do some research before posting.

Edit: Here's an interesting tweet while we're on the subject.

 

[proc]

I think X86-only for 10.6 is likely

This is wrong on two accounts:

1. There is no reasonable evidence that Snow Leopard is for Intel CPU's only. Actually, threads discussing this seemed to be evenly divided between people having experience developing software who are quite sure that Snow Leopard will run on PowerPCs, and a fanbois who think it doesn't.

2. If Snow Leopard were to run on Intel CPU's only, then this would cause considerable damage to Apple. First by needlessly giving up > $100 million in software profits, secondly by creating a precedent that would change the value of any new Macintosh in a very strong negative way, thirdly by destroying twenty years of tradition in the NextStep source code base of being processor independent, with lots of invisible cost that are hard to explain to a non-developer.

Obviously that will be hard to understand to someone who feels it necessary to call people "PPC whiners".

So what makes you think that? What experience do you have at developing software? Please tell me. I'd really like to know. The code that I am working on is about 1 million lines of code, of which about 10 are specific to either PowerPC or Intel. That is 0.001%. How much code have you written, and how much could be saved if that code were to run on Intel only instead of Intel + PowerPC?

Expected saving in programming time: Negative. Even without PowerPC, OS X would have to run on Intel 32 bit, Intel 64 bit and ARM. But what non-developers don't realize is that being forced to write portable code keeps the code quality up, which produces huge long-term savings.

I think that the difference between Big Endian (PPC) and Little Endian (X86) would make it a worthwhile difference alone. Adobe, Microsoft and other big software developers seem to agree with me see for example this

Besides these reasons there must be a point where development stops. PPC users are served best with a graceful phasing out, which can be achieved for instance with keeping the APIs between two successive versions the same as much as possible. It would be obvious you can only do such a thing for a limited number of times as an OS developer, before your customers will start complaining because there is no apparent progress being made on the OS, as end-users measure progress with visible feature additions to an OS.

You stated that such a move would not be appreciated by corporate environments: you are right. This only makes it even more important that such a move is accomplished and finished before an OS-developer (such as Apple) starts to direct his products more to corporate environments.

All these reasons combined should give you a better insight what the reason of a drop of PPC compatibility would be.

 

[snberk103]

All big cats gone? How about "NO" for that?
Lynx, Lion, Bobcat, Cougar, Golden Cat, Mountain Lion, Serval.

I think we can remove 2 names from that list.

Lynx: Can you imagine the confusion at the Apple store as they try to explain that you don't need to compile anything to make it work, and that "Lynx" (a text based linux browser) won't actually run on Lynx....

Or Serval: The marketting types would have a hard time explaining OS X Serval is not OS X Server (which is actually OS X Serval Server)

For the record - I don't see OS X becoming OS XI. Perhaps OS 11, but I think we will see something very different when OS X is retired. Or.... OS X will continue for quite a while yet, but the animal category will change. There really isn't any reason Apple couldn't call OS version 11.X.X "OS X". In this case "X" is a letter. Anyone look at which animals Apple has trademarked lately?

I like Bobcat.

 

[DavidCar]

I wonder what the relationship will be between NSOperation and Grand Central Dispatch? If I design an application to use NSOperation, will it be a natural extension to use Grand Central, or will it require a redesign?

 

[Kenny Rullo]

Maybe Grand Central and OpenCL would work under the same layer of abstraction.
Apple have to redesign some core features, such as coreimage or coreanimation, to work seamlessly with this new baby. And let us work without the hassle of manually call them.
Sorta of garbage collector not just for unused object, but for media and threads. Let's call it "media collector", dispatching the right item to the right tecnology: animations to opencl, threads to grand central.

 

[Orng]

All big cats gone? How about "NO" for that?
Lynx, Lion, Bobcat, Cougar, Golden Cat, Mountain Lion, Serval.

"Cougar" will probably be some classic OSX environment within OS 11.
Because while most of us are all hot for the sweet young thing in OS11, OSX will be the mature OS, on the prowl for whatever lovin' it can get.
Otherwise, if Apple has any marketing sense (and they do) they won't use Cougar and only trademarked the name to keep it under their control.

 

[qtx43]

I don't see using the Graphics CPU's to increase work in the general computing environment. If your number crunching you would probably see the big gains.

It depends on what you define as the "general computing environment". Does Photoshop qualify? Or signal processing on sound for e.g. voice synthesis? Cryptography? Physical modeling? Maybe these sound sort of esoteric, but they're not when built into an actual application. As time goes on, people will think of all sorts of uses for your average person who doesn't care much about computers other than as a tool for their real work.

0Almost makes me wish I went to WWDC this year to find out more detail.

Me too! Except, realistically, I doubt there's much real information yet.

 

[ManchesterTrix]

I think we can remove 2 names from that list.

Lynx: Can you imagine the confusion at the Apple store as they try to explain that you don't need to compile anything to make it work, and that "Lynx" (a text based linux browser) won't actually run on Lynx....

Lynx does run on OS X. Also, there would be no confusion because the people who know about the web browser would know the difference between the browser and Apple's operating system without needing a store-employee to hold their hand.

 

[pohl]

I think that the difference between Big Endian (PPC) and Little Endian (X86) would make it a worthwhile difference alone. Adobe, Microsoft and other big software developers seem to agree with me see for example this

Endian-ness is trivial in the context of something like LLVM IR. The link you refer to is a case where the app developers had written a bunch of Intel-specific code. Apple has been, in contrast, endeavoring to write as little architecture-specific code as necessary. (Witness their replacing two architecture-specific fallback-interpreters in the OpenGL stack that were replaced by a unified, architecture-independent, LLVM-based mechanism in Leopard.)

 

[commander.data]

"Reverse Hyperthreading" has been completely debunked. It never existed, not at AMD or anywhere else.

Well, yes. I mentioned that Reverse Hyperthreading was a failed rumour in my original post. I was just pointing out that the concept of breaking an individual thread down into packets for distribution to multiple cores, which is what was used to describe Grand Central at RoughlyDrafted, reminded me of the principle that was associated with Reverse Hyperthreading.

 

[sfh]

everything will run faster ....Sweeeeeet

Basically makes it so the program doesn't have to have multi-thread cpu coding in it to take advantage of the multi cpu cores. This is going to be insane!

 

[A Pittarelli]

this will basically create the difference that there should be between the imac/mac pro and macbook/macbook pro.

 

[CrackedButter]

I think we can remove 2 names from that list.

I don't think Golden Cat is very marketable either. There is of course Sabertooth if the marketing department can get past the prehistoric angle!

Of those names, Apple only need 3 more cats till OS 11 (unless we have a 10.10 release!), Cougar and Lion are the strongest and I'm surprised Lion hasn't been used yet, maybe Apple are saving the King of the Jungle till last!?! Bobcat seems reasonable and while Lynx could cause a slight confusion with Lynx the browser but only the geeks know about it. But Apple have registered it as a trademark anyway and I'm sure they have thought this one through all the way till the end.

My money is on Apple using Lynx, Cougar and Lion.

 

[milo]

yeah for sure. PPC whiners need not post.

Get used to it, pal. You're going to be hearing from us for the next year or more.

This tech is perfectly suited for a machine like the quad G5.

Hi,

Wouldn't Snow Leopard run on multi core G5's?

s.

Absolutely.

NO. Please do some research before posting. Snow Leopard is for INTEL CPU'S ONLY. The way it SHOULD be.

Quit being a jackass. There isn't any official confirmation that it will be intel only (unless I missed an announcment from apple, there wasn't one, was there?). Just speculation based on the initial dev preview. I have to agree with the previous post - trolls like you just give Mac users a bad name.

 

[snberk103]

Lynx does run on OS X. Also, there would be no confusion because the people who know about the web browser would know the difference between the browser and Apple's operating system without needing a store-employee to hold their hand.

All I can say is, - don't confuse me with the facts! Maybe not the best example, but I still think there is room for confusion. The names Linux and Lynx are too similar. Imagine the telephone media interviews that go horribly wrong because the interviewer is not tech savy and confuses the two. Imagine all the media writers who rely on their spellcheckers, and don't notice the one being swapped for the other.

 

[savar]

Wow, actually can't wait for Snow Leopard. Glad that SOMEONE is looking to build a solid, fast foundation for future development rather than adding shiny fluff.

Apple has such a solid foundation now, that I am beginning to think there won't ever be another transition as big as Mac OS 9 -> Mac OS X was. Unless somebody commercializes quantum computing or something equally crazy, OS X is so modular and portable that it should survive really significant changes in architecture.

 

[gnasher729]

I think that the difference between Big Endian (PPC) and Little Endian (X86) would make it a worthwhile difference alone. Adobe, Microsoft and other big software developers seem to agree with me see for example this

If Adobe, Microsoft and other big software developers can't write portable code, that is just too sad for them. MacOS X is based on NextStep, which has been portable for the last twenty years. If that hadn't been the case, we wouldn't _have_ any Intel-based Macs. Let me have a quick look at the CoreFoundation source code: CFApplicationServices.c has no PowerPC/Intel specific code at all. CFArray.c has _one_ place where it adds padding for 64 bit system; that is needed for 64 bit Intel as well as PowerPC. CFBag.c has faster hashing on 64 bit, no other differences. So: No dependencies on PPC vs Intel. Some dependencies on 32 vs. 64 bit. And guess what: There is code there to make CoreFoundation run on Windows!

 

[eastcoastsurfer]

I think that the difference between Big Endian (PPC) and Little Endian (X86) would make it a worthwhile difference alone. Adobe, Microsoft and other big software developers seem to agree with me see for example this

Besides these reasons there must be a point where development stops. PPC users are served best with a graceful phasing out, which can be achieved for instance with keeping the APIs between two successive versions the same as much as possible. It would be obvious you can only do such a thing for a limited number of times as an OS developer, before your customers will start complaining because there is no apparent progress being made on the OS, as end-users measure progress with visible feature additions to an OS.

You stated that such a move would not be appreciated by corporate environments: you are right. This only makes it even more important that such a move is accomplished and finished before an OS-developer (such as Apple) starts to direct his products more to corporate environments.

All these reasons combined should give you a better insight what the reason of a drop of PPC compatibility would be.

Endianness is not an excuse to only support x86. It's a trivial thing to work around as long as you thought about it from the start (which obviously Apple has since they have both code bases). Any data you put out across the network has the same endian issue and must be converted from host -> network and then back network -> host and I don't see people suddenly saying they can't support networking in OS X or write network aware programs.

The other thing that having PPC around does is help make the quality of the software better. Having the software run on multiple architectures can make it easier to find hard to track down bugs.

If Apple drops support for PPC, it's purely a business decision. They want people to move off of PPC and probably have an internal time line as to when they can force the issue.

 

[pilotError]

Endianness is not an excuse to only support x86. It's a trivial thing to work around as long as you thought about it from the start (which obviously Apple has since they have both code bases). Any data you put out across the network has the same endian issue and must be converted from host -> network and then back network -> host and I don't see people suddenly saying they can't support networking in OS X or write network aware programs.

The other thing that having PPC around does is help make the quality of the software better. Having the software run on multiple architectures can make it easier to find hard to track down bugs.

If Apple drops support for PPC, it's purely a business decision. They want people to move off of PPC and probably have an internal time line as to when they can force the issue.

Agreed, the whole point of the Mach kernel was to make it hardware independent.

 

[mdriftmeyer]

Multithreading is when two parts of your program run at the same time. Grand Central doesn't sound like multithreading, it sounds like instead of processes or threads being the thing you schedule on a processor, you instead schedule some kind of work unit.

That in itself wouldn't allow for more parallelism, but would mean you could ensure that processors can be "filled up" regardless of the relative granularities of your tasks or threads.

Correct. It's Parallel Programming via the network packet management model.

 

[Stridder44]

I wish people would stop whining about OS 10.6 working on PowerPC Macs. That kind of attitude is exactly what will keep Macs behind the competition.

No kidding. PPC was an awesome chip in its day, but the sun is starting to set on them (for Apple anyway). As much as it may suck for some, going Intel/64-bit only is the best way. Looking forward is what's important. Hardware is outpacing software.

Now let's just hope Microsoft can do something similar with Windows 7 (yes, I like to game from time to time).

 

[exabytes18]

Have you looked at it. And do you know what they have pland for that kernel.

1) They did not cut there operation system down to 40MB.
They cut ther kernel down.
It can boot up in command promt and only grafik is ACSII grafik.
They can start a webserver, but nothing fancy. it only says I'm here.

And if you get DSL (Damn Small Linux) they you get support for most of you hardware, and grafik interface, and it only 50MB.

So I'm not impresst whit MinWin.

I think you missed the point of my post. Operating system or kernel, whatever, it was running in a barebones scheme. It was to demonstrate that windows is lean and efficient at the core. And bloat people speak of is actually just carrying drivers for everything under the sun and the ability to run legacy programs. MinWin was never intended to be the kernel of Windows 7 or other operating system, it was simply a demonstration.

And wow! If DSL can support most of my hardware, then Microsoft and Apple must be doing something wrong! Geez, multi-gigabyte operating systems condensed down to 50 megabytes!

 

[Gasu E.]

I wish people would stop whining about OS 10.6 working on PowerPC Macs. That kind of attitude is exactly what will keep Macs behind the competition.

Wow, I hadn't realized... so sorry Apple! I vow I will not hold you back again!

Thank you for the attitude adjustment, sir Beric. Your guidance is a boon to us all.

 

[katorga]

GC will rock

In the not too distant future, low cost consumer processors will have 16-32 atom cores. Without an easy to use API and interface, coding an app to take advantage of that many cores will be so difficult that most users will say "WTH am I paying for". This is the same reason that Intel is looking at using lots of cores to do ray-traced 3D games.

GC should make it easier to take advantage of massive multi core systems for consumer activities.

 

[milo]

No kidding. PPC was an awesome chip in its day, but the sun is starting to set on them (for Apple anyway). As much as it may suck for some, going Intel/64-bit only is the best way. Looking forward is what's important. Hardware is outpacing software.

Obviously you're not using a quad G5. It's STILL an awesome chip.

Apple should dump machines when they can't keep up. The high end G5s are more than capable.

Going intel only will only help in that it saves apple resources and money. There's no technical reason that dumping PPC would help on the intel side at all.