OpenCL 1.0 Specification Completed and Released

[baslotto]

To Winni:

WOW! What a post!
It's the first time that I actually see this from a "corporate" prospective and I have to agree with everything you pointed out.
Reading your post was like watching the aftermath of Pirates of Silicon Valley!
I agree with everything you said and I hope Apple stays the way it is now which is good for a few but not for all of them (for all the reasons we know, plus I'm in the music business).
My greatest happiness is that I can use my 5 year old Powerbook G4 (still fast and ready) and my new iMac, for work and for typing this post on the internet, without being too careful of what I'm doing.

One question about this "OpenCL":

Is this a help to programmers so that ANY applications would find a benefit or only graphics oriented apps are going to be faster?
Is the hardware available now going to get any benefits or only the new MacBook Pros with 2 GPUs are gonna see an improvement?
For instance: my iMac is a 1 year old Core 2 Duo. Are my music-production programs going to be any "faster" in the future with this machine or I'm pretty much stuck with whatever my hardware allows me to do now?

Thank you for any answers,
I'm just trying to bring back the topic to the mac rumors.


Bas.

 

[knightlie]

Y'all are crazy if you think Windows Vista is a failure because people are scared of it. They were scared to death of XP too and stayed with Windows 98 and 2000 for YEARS despite how vastly superiour to 98 and clearly improved from 2000 Vista was.

EVERY Windows release gets the same thing - even Windows 95 way back. How short your memories are. Microsoft cannot release an OS people want to upgrade to. It never has.

In 2009 or 2010, Windows Vista should be much better accepted - the tipping point will be when Windows 7 is released and people want downgrades to Vista...

That's the Microsoft product cycle and always has been. Kinda a sad reflection on their inability to deliver a compelling feature set, eh?

Very true, I was still using Windows 2000 at work until my machine was forcibly removed. Microsoft's biggest problems in this regard are that a) they cannot deliver a seamless, even vaguely hassle-free upgrade, and b) they keep changing the whole OS*, thereby restarting a big chunk of the learning curve for users. Who want's to learn the arrangement of a new operating system every time MS's bank balance needs propping up?

*And some amusing thoughts on Vista's lack of UI consistency here.

 

[mdriftmeyer]

What?

That's one of the silliest responses that I've seen on these fora - "threading" isn't related to "multi-processing"? Are you kidding?

While, of course, you can use threaded programming models on a uni-core machine - it's on a multi-core system that threading is what fundamentally exposes parallelism within a single process context.

And, by the way, none of your arguments address my position that OpenCL/GrandCentral don't bring anything new to the multi-core table - they simplify things, not bring new capabilities.
_____

Please, also note that my last few posts here have not meant to knock OpenCL/GrandCentral per se -- I've been trying to point out that for most people, on most applications, on most systems already sold - ICL/GC won't make a noticeable improvement in the speed of their systems.

Parallel Programming isn't multiple thread programming. I'll let this gentleman explain:

http://www.danielmoth.com/Blog/2008/11/threadingconcurrency-vs-parallelism.html

How GrandCentral and OpenCL Parallel Programming isn't the multithreading model of single core systems of old depends on the model employed:

https://computing.llnl.gov/tutorials/parallel_comp/

With Multiple Processors compounded with Multiple cores, per processor, task segments will have to be methodically broken up to offload linearity of program execution far differently than parallel threading out tasks without much concern for actual performance gains and more for perceived UI responsive behavior, on a traditional single core, ala pthreads and more.

It's all there at the Government site and his blog does a great overview of hands-on work.

This is a nice textbook on the subject matter:

http://www-users.cs.umn.edu/~karypis/parbook/

This one is solid for C++ backgrounds:

http://www.amazon.com/Art-Multiprocessor-Programming-Maurice-Herlihy/dp/0123705916/ref=pd_sim_b_3

 

[AidenShaw]

Parallel Programming isn't multiple thread programming. I'll let this gentleman explain:

http://www.danielmoth.com/Blog/2008/11/threadingconcurrency-vs-parallelism.html

I'm not sure what you're reading, but when I open that first link the first line is:

To take advantage of multiple cores from our software, ultimately threads have to be used.​

I do understand parallelism - in fact I was literally working on SMP operating systems and multi-threaded applications when you were in grade school.


Please understand that I'm not talking about whether Apple's new parallel frameworks are good or bad - I'm addressing the perception that suddenly 10.6 will make older systems much faster.

My comment about threading is that many of the important CPU-intensive applications (video/audio/image) already use multiple threads to exploit multiple cores. GrandCentral isn't going to speed those up by much on current machines -- even when they're re-written to use GrandCentral.

GPGPU programming may make some big improvements when they appear, but the value will be relatively limited. (Picture some GPGPU-enabled Photoshop filters being 10x faster, for example.) The GPGPU suffers from both the issues of parallelism (Amdahl's law and friends) plus the special purpose nature of the GPGPU. Larrabee will be interesting, since it will have dozens of x64 CPUs.

(These two paragraphs assume that major applications will drop support for Leopard and older systems in order to use the new frameworks - that's unlikely for some time.)​

In two or three years, the new frameworks will be a great boost - but not for your Core Duo or GMA900 system.

 

[Eidorian]

I'm not sure what you're reading, but when I open that first link the first line is:

To take advantage of multiple cores from our software, ultimately threads have to be used.​

Please understand that I'm not talking about whether Apple's new parallel frameworks are good or bad - I'm addressing the perception that suddenly 10.6 will make older systems much faster.

My comment about threading is that many of the important CPU-intensive applications (video/audio/image) already use multiple threads to exploit multiple cores. GrandCentral isn't going to speed those up by much on current machines -- even when they're re-written to use GrandCentral. (This is assuming that major applications will drop support for Leopard and older systems in order to use the new frameworks - that's unlikely for some time.)

In two or three years, the new frameworks will be a great boost - but not for your Core Duo or GMA900 system.

Thank you. It gets old when people go on how Snow Leopard is going to suddenly make their older hardware faster.