Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Snow Leopard's Grand Central and OpenCL Details?
- Thread starter MacRumors
- Start date
- Sort by reaction score
This is most encouraging news. Focus on performance, optimization, hyperthreading, and streamlined code which fully contrasts a certain bloated OS out there.
pthreads are ridiculously heavyweight (at least half a meg of allocation each), as well as being difficult to use safely. Generally it makes a lot more sense to have a pool of threads available and something to manage them and dispatch work to them...
Wow, from 0 to know-it-all in 7.9 seconds flat.
Until you have a final, shipping version of Snow Leopard in your hands, you don't know any more than the rest of us. If believing PPC support will be dropped keeps you all warm and cozy at night, more power to you, but don't assume it gives you the right to lash out at anyone else simply posting a question about it. Dial the attitude back a tad.
So would this all have any benefits for gaming on the Mac???
I hope Apple keeps treading the way with Multi-Touch in OS X too though so it doesn't look like Microsoft is ahead of them when Windows 7 comes out. Even though I think Microsoft has it implemented all wrong I think Apple needs to stay in the lead! I see Apple doing it more with the touch surfaces on the computers though and not in the screens which I think is good. The way Microsoft is trying to do it just looks ridiculous and worthless.
I am all for dropping PPC even though I have only ever owned Intel Macs haha. It just makes sense and is necessary for the future of this great foundation and I am glad Apple is really trying to get things solid!
Last note: Why have we only heard talk of full ZFS support in "Snow Leopard Server" and not "Snow Leopard"??? Would Apple really put it in only server and not both???
I hope Apple keeps treading the way with Multi-Touch in OS X too though so it doesn't look like Microsoft is ahead of them when Windows 7 comes out. Even though I think Microsoft has it implemented all wrong I think Apple needs to stay in the lead! I see Apple doing it more with the touch surfaces on the computers though and not in the screens which I think is good. The way Microsoft is trying to do it just looks ridiculous and worthless.
I am all for dropping PPC even though I have only ever owned Intel Macs haha.
It just makes sense and is necessary for the future of this great foundation and I am glad Apple is really trying to get things solid!Last note: Why have we only heard talk of full ZFS support in "Snow Leopard Server" and not "Snow Leopard"??? Would Apple really put it in only server and not both???
even though i own and love my PPC mac, i think it's time to let go and make it full intel, just imagine the resources saved just programming one! maybe it'll be on time!
Okay, do you honestly think that Microsoft has been sitting around redesigning the user interface since the days of 3.0? They have continued to refine the kernel and are definitely concerned with performance.
http://www.microsoft.com/whdc/system/vista/kernel-en.mspx
Also, ever hear of MinWin? Microsoft cut down their operating system to 40MB and it even had a webserver, just to demonstrate that Windows at the core isn't bloated. The point: if you want to forgo backwards compatibility and drivers for every component under the sun, you'll have an operating system like OSX (tied to hardware and the fear you might not be able to run the next OS on your fancy G5).
I really think this release is going to be huge. Apple has looked to the future and is retooling to be ready for it. If GC lives up to what I'm imagining its promise to be, it will be the first real step I've seen towards tailoring an OS to multicore systems.
Full blown threads make little sense on an 8+ core system. The only time I see all 8 of my CPUs going full tilt is when I'm compiling. It's just too complicated to safely thread out an application that way. Breaking the work into units and sharing it around should give the same benefit to the system that the front end to modern superscalar processors gives-- keep all of the execution units busy.
The real question to me is whether the overhead of farming out these work units is more than just executing them directly. That's always the question with parallel computing, but the smaller the fundamental unit the harder it becomes to maintain efficiency.
Full blown threads make little sense on an 8+ core system. The only time I see all 8 of my CPUs going full tilt is when I'm compiling. It's just too complicated to safely thread out an application that way. Breaking the work into units and sharing it around should give the same benefit to the system that the front end to modern superscalar processors gives-- keep all of the execution units busy.
The real question to me is whether the overhead of farming out these work units is more than just executing them directly. That's always the question with parallel computing, but the smaller the fundamental unit the harder it becomes to maintain efficiency.
That would be fantastic, and not entirely unreasonable to expect. I don't know enough about how Xgrid (or GC) works to know how far apart the technologies are.It would be amazing if Grand Central tied in with xgrid, to allow for applications to easily exploit idle network cpu's if they're listening as xgrid nodes. Wouldn't be practical for most apps, but tasks like video encoding could benefit from being optimized to make use of multiple cpu's on a network. That's likely just a pipe dream, but a guy can dream
There has been no definitive statement on this. Rumors have gone back and forth. Unless you can point to an Apple press release, you need to back off. Your posts are predominantly meaningless attacks.
Ja, that is my concern too. And not only the work unit being too small, but if they don't leave it running long enough they won't be leveraging caches.
The trouble is, it's not one size fits all for the work unit sizes. Perhaps they could dynamically adjust the size to suit the process, but that's more overhead gathering adjustment statistics. Or something simple like breaking work units on CALL instructions.
One can only wonder why then MS doesn't offer 24 additional versions of their OS, one for each level of backward compatibility:
Windows Vista Starter version 3.11, Win CE, Win, 98, Win NT, Win 2000
Ultimate version 3.11, Win CE, Win, 98, Win NT, Win 2000
Home Premium version 3.11, Win CE, Win, 98, Win NT, Win 2000
Home Basic version 3.11, Win CE, Win, 98, Win NT, Win 2000
Business version 3.11, Win CE, Win, 98, Win NT, Win 2000
Enterprise version 3.11, Win CE, Win, 98, Win NT, Win 2000
This would work well within the existing business model, while significantly reducing the bulk of carrying backward compatible code.
So would this all have any benefits for gaming on the Mac???
I am all for dropping PPC even though I have only ever owned Intel Macs haha.
Games tend not to build true native applications for any platform so unless various engines pick up the features then it's not looking good for games.
As it's been said no official drop of PPC yet Indeed no reason to believe just because the Developer Preview doesn't yet support PPC that i won't it the future.
It all depends on what they are really doing under the hood. If is a updater compiler like full use of LLVM as suggested. Maybe the compiler to the virtual machine is Intel only or the only build that is ready the developer use.
Hey the run times engines which take the intermediate representation (IR) on the real hardware are already working in Leopard. So why not keep moving them forward.
Hey if Snow Leopard is really delivering on performance improvements it's just as likely that it will have lower not higher system requirements.
Compile for 10.6?
The biggest question still, for me, is whether applications will need to be compiled explicitly for 10.6 in order to take advantage of the performance improvements, or whether applications will continue to be compiled for 10.5 and effectively get the speed improvements for free when running on 10.6? Leopard delivered updates to Cocoa that made multi-threaded computing easier, but you needed to be compiling your applications for 10.5 only in order to be able to use them. From a developers perspective, if applications running under 10.6 need to be compiled for 10.6 in order to see the improvements, particularly if code changes are necessary, then that's going to be annoying since you might be missing out on the business from PowerPC owners.
This, of course, depends on whether PPC support is indeed being dropped, and I entirely agree with those noting that no explicit confirmation/denial on this subject has been made yet.
The biggest question still, for me, is whether applications will need to be compiled explicitly for 10.6 in order to take advantage of the performance improvements, or whether applications will continue to be compiled for 10.5 and effectively get the speed improvements for free when running on 10.6? Leopard delivered updates to Cocoa that made multi-threaded computing easier, but you needed to be compiling your applications for 10.5 only in order to be able to use them. From a developers perspective, if applications running under 10.6 need to be compiled for 10.6 in order to see the improvements, particularly if code changes are necessary, then that's going to be annoying since you might be missing out on the business from PowerPC owners.
This, of course, depends on whether PPC support is indeed being dropped, and I entirely agree with those noting that no explicit confirmation/denial on this subject has been made yet.
I really hate when people try to talk in the short-sentenced salesman-y Apple speak.
Regarding PowerPC support:
I agree there has been no official announcement, but all signs point to no PowerPC support. The only dissenting rumor was from Gizmodo, and to be frank, it was crap.
1. it was reported in a "someone just sent us this and we don't know them" manner
2. it claimed that the pre-release version of Snow Leopard supported PowerPC... which we know now it does not.
finally, the original rumors that named Snow Leopard and described its "features" also claimed intel-only.
again, no official announcement, but safe bet is no PowerPC support.
arn
I agree there has been no official announcement, but all signs point to no PowerPC support. The only dissenting rumor was from Gizmodo, and to be frank, it was crap.
1. it was reported in a "someone just sent us this and we don't know them" manner
2. it claimed that the pre-release version of Snow Leopard supported PowerPC... which we know now it does not.
finally, the original rumors that named Snow Leopard and described its "features" also claimed intel-only.
again, no official announcement, but safe bet is no PowerPC support.
arn
Think of it as drag and drop multithreading. It lowers the burdens on software developers.
Instead of writing a multithreaded application and then having to jump into bed with pondering the consequences of thread management and how to divide your application, not to mention all the catches and oddities of multiprocessor-aware software, you don't worry so much. All you do is create these blocks, and Grand Central will move them around itself among the cores (and the GPU, if OpenCL-ready), running them in parallel on multiple cores, or best-availability in the same core. Like OS X manages whole applications among cores or CPUs, Grand Central will micromanage within applications.
Multi-core G5's of a similar engineering nature only appeared in the very last Power Mac revisions. Older Dual and Quad systems don't work the same way. Neither configuration is directly compatible with Intel's approach to multicore systems.
That's not logical, but I've been seeing it a lot on here. It's possible that there will be no increase in system requirements, but lowering them does not make sense, as improving performance does not mean that the system requires fewer resources overall. Improving performance has more to do with dealing with resource competition than with improving idle conditions (though the latter may be a way to achieve the former).
There's no way to lower the Intel requirements (all Intel systems are already covered). Lowering PowerPC requirements would mean overcoming the G4's poor bus performance and slow memory speed, which is highly unlikely--even if 10.6 supported PPC.
According to the German news site www.golem.de AMD has joined
Khronos compute working group which has suggested a specification for OpenCL as a platform independent standard.
AMD also offers special "compute accelerator" cards that are based on GPU designs, but not capable of generating a video image. This will also be supported by OpenCL (in addition to AMD's current "Stream SDK").
I think with support from at least one of the two big GPU producers chances are high that we will see OpenCL outside the Apple world.
Microsoft on the other hand has announced some gpu computing technology for Direct X11, but this won't be released before Windows 7 and it looks like this will target GPUs only (not multicore or "manycore" CPUs).
http://www.golem.de/0806/60400.html
Christian
"If the things we value most in an OS are stability, performance and technical advancement, why are those the very things for which we are least willing to pay?" - archdetector
Ada task object as scheduled unit? Hm, "packetising"... so if I call f(g(a), h(b)) will it work out that g,h have no side effects and process g and h in parallel, before calling f with the results? Includes clustering?
Apple's good at taking old research and making it accessible, but bad at proper citation.
Apple's good at taking old research and making it accessible, but bad at proper citation.
What, as opposed to Apple's 31 year old kernel (BSD), which is based on a 37 year old design (UNIX) with a 24 year old overhaul (Mach)?
I'm not saying you're wrong, just that the age of the kernel is not really relevant. It's also worth remembering that the NT kernel was designed by a very good software engineer - Dave Cutler. I believe that, more than anything, it is the need for legacy application and driver support crippling modern NT.
Yea Bittorrent for the processor! Clever move
I'm guessing we'll also see a version of that file system (whatever it was called) that were promised/rumored for Leopard as well.
Feels weird actually anticipating no new features this time around.
/Edit
Moment of inspiration; Variable Core Processors, step into the now. Need more power? Just flop in another core and you're good to go.
I'm guessing we'll also see a version of that file system (whatever it was called) that were promised/rumored for Leopard as well.
Feels weird actually anticipating no new features this time around.
/Edit
Moment of inspiration; Variable Core Processors, step into the now. Need more power? Just flop in another core and you're good to go.