OpenCL Specification Completed in Record Time for Snow Leopard

[MacRumors]

http://www.macrumors.com/images/macrumorsthreadlogo.gif (http://www.macrumors.com/2008/11/19/opencl-specification-completed-in-record-time-for-snow-leopard/)

Macworld reports (http://www.macworld.com/article/136921/2008/11/opencl.html) from the SuperComputing 08 conference (http://sc08.supercomputing.org/) which is taking place in Austin, Texas this week. The Khronos group (http://www.khronos.org/) was present to celebrate the launch of the upcoming OpenCL specification which Apple is planning to implement in Mac OS X 10.6 Snow Leopard. OpenCL is a new programming interface (API) to allow developers to take advantage of parallel computing across graphics processing units (GPUs) and multi-core CPUs.

While few details were revealed, Khronos’ president Neil Trevett explained how quickly the OpenCL specification had come together. “If you go to some other larger standards bodies, it’s quite normal for a standard to take five years or more,” Trevett said. “That’s quite commonplace. You actually have to really push to get it down to eighteen months. Our record was 12 months, up to now; we’ve done this one in six [OpenCL].”The speed at which they completed the specification was due to Apple's tight schedule to allow it to ship with Snow Leopard. The specification is now being reviewed by lawyers for the next 30 days to make sure no intellectual property has been breached. The specification will presumably be complete once it passes this 30 day inspection.

Trevett was very optimistic about the prospects of OpenCL specifically implemented in Snow Leopard:“If Apple ends up following through on the plans they stated on building this specification into Snow Leopard, I think you could see opportunity for imaging applications vendors, video application vendors to tap into the goodness of GPU hardware,” Trevett said. “Everyone has a supercomputer locked away in their Mac, but it’s hard to get at it. And OpenCL will unlock the potential of that supercomputer.”He also acknowledged that OpenCL would work from the cell phones to high end computers, leaving open the possibility that we could see OpenCL implemented in future iPhones.

Apple's Snow Leopard (OS X 10.6) is expected to ship in 2009 with recent hints (http://www.macrumors.com/2008/11/18/mac-os-x-snow-leopard-10-6-due-in-q1-2009/) suggesting the first quarter of 2009.

Article Link: OpenCL Specification Completed in Record Time for Snow Leopard (http://www.macrumors.com/2008/11/19/opencl-specification-completed-in-record-time-for-snow-leopard/)

[AppleMatt]

Sounds pretty impressive...!

I'd love to see an OpenCL app on an iPhone, purely out of interest as to what it could do.

AppleMatt

[840quadra]

Sounds good. I am glad to see this as an opportunity for applications to really shine. Hopefully Adobe will support this on the OS X side.

http://forums.macrumors.com/image.php?u=47064&dateline=1165207334&type=profile

[xUKHCx]

Seems like apple are really pushing this forward. Will be interesting to see some benchmarks.

Anything to bring some extra speed to a computer is welcome.

Snow Leopard sounds very very interesting, wish apple had given a bit more away but I suppose they were hurt by the Leopard Secret features.

[poundsmack]

the Khronos has never been known for super speedy standard formalization. I am glad that OpenCL got rocketed through by apple. lets just hope that rushing it doesnt lead to issues like OpenGL 3.0 being put out a little before it was ready (many are looking forward to 3.1 to add teh features they wanted to see in 3.0)

[BlueRevolution]

The question that comes to mind for me is whether the quality, stability, or reliability of the interface will have suffered from the tight schedule. Of course, Apple can churn these out much more quickly because it isn't a standard at all, whereas Trevett's comparison is to "larger standards bodies", which bring a number of parties together to work on a standard that will benefit all of them. Apple is doing this on a unilateral basis, so of course they can do it faster.

Yes yes, good job Apple. Just don't waste too much time patting yourselves on the back.

[pgifford]

that's fast. good job Apple!

[m1stake]

Good for them. Now, to fix the mess that is OpenGL 3. :(

[theheadguy]

This is good, really good. Most of all, I like 'tight schedules'... Snow Leapord can't come soon enough.

[koobcamuk]

I hope Apple doesn't give anything away on Snow Leopard.

[rockinrocker]

i just hope it works as advertised...

[amac4me]

Can't wait for Snow Leopard!

[guzhogi]

I'd love to see what OpenCL (along w/ Grand Central) will bring. I wonder how much processing power goes wasted (not having enough to do as well as poorly written software). Plus, it would be good to have a standard way of doing multiprocessing (saying the APIs do it well).

As for OpenGL, I don't know anything about it. But reading these posts, I hope 3.1 gets all the features you want, Apple actually uses compatible graphics cards when they come out (not 2+ years after the fact), and has good drivers to go w/ them. If all this happens, new Macs would absolutely scream! But, if you want to see really great graphics, go outside. :D

[Yuppi]

My guess is that they were able to be so quick because NVidia has lots of experience with CUDA. It doesn't make sense that Apple alone would create such a service. Also both NVidia and AMD (formerly ATI) have big interests in making these technologies available. Especially AMD which plans to built-in GPUs into processors.

[commander.data]

I'm amazed OpenCL was completed so quickly. Unlike say OpenGL 3.0. Apple no doubt had the entire spec laid out already. The delay was just having to shove it down people's throats getting them to agree that it may not be the best in world, but it'll do for a 1.0 version. And it's great that they are going to get it out before DX11.

“The fact that if we could hit this impossible deadline [Apple] would support it in Snow Leopard was a huge plus to us,” said Intel’s Tim Mattson.
...
“The specification was written very carefully not to tell vendors how they implement things. So that gives them flexibility they need to map onto virtually any device they can think up,” Mattson said. “So, yeah, it’ll be beneficial to users across the board. And I can’t emphasize enough that I’ll be able to write one program that I’ll be able to recompile and it will run on my cell phone, it will run on my Powerbook.”
I wonder if Intel's enthusiastic support for OpenCL means that they might add support for say the GMA X3100. It's interesting that he mentioned it might run on a Powerbook. Perhaps he meant a MacBook Pro, but maybe the spec is flexible enough that it can be implemented on older GPUs if Intel, nVidia, and ATI develop the drivers. Certainly unified shaders in DX10 generation hardware isn't necessarily required since the X1900 did GPGPU work just fine. I emailed ATI a few weeks ago putting in a plug for OpenCL support for the X1600 that is in many MacBook Pros and iMacs. A Stream Computing manager there replied that OpenCL was still being defined but they'll keep it in mind. So hopefully there are results. It should be in everyone's interest to support OpenCL on as much hardware as possible.

And hopefully Intel will support OpenCL in Larrabee and not just their own interface implementation.

[rumplestiltskin]

Perhaps people are expecting the usual "standards" jockeying when Microsoft is involved in the process.

You'd be amazed at the progress we can make when we remove Microsoft from the equation. :D

[Cromulent]

Apple is doing this on a unilateral basis, so of course they can do it faster.

No they are not. The Khronos group which is the standards body which made this standard is made up of lots of different companies not just Apple.

This is most definitely a standard from a standards body. It is not Apple just doing it on their own.

[BornAgainMac]

In January I hope Apple will have a demo on how good it could be compared to current technology.

[ChrisA]

The question that comes to mind for me is whether the quality, stability, or reliability of the interface will have suffered from the tight schedule. Of course, Apple can churn these out much more quickly because it isn't a standard at all, whereas Trevett's comparison is to "larger standards bodies", which bring a number of parties together to work on a standard that will benefit all of them. Apple is doing this on a unilateral basis, so of course they can do it faster.

Yes yes, good job Apple. Just don't waste too much time patting yourselves on the back.

No, apaerently Apple didn't do this all by themselves. What they did is work or a long time with Intel, nvidia, AMD and other before going to Kronos with their draft OpenCL spec in hand. They did the work in a small club.

I read only a light weight "technical overview" that had some code examples. I looks to me like the standard itself is small and there is not to much to argue about it it. All the work will be in implementing it. Writing an OpenCL run time system looks to me like a very majoir effort

Also while Applemight use a little of OpenCL in SL. I think before we see it widely used we will need for there to be some higher level libraries. It would a an almost no-brainer touse OpenCL to drive the physics in an OpenGL based game but first you'd have to re-write one of the common physics engines touse OpenCL. My gues is that uses will not see much from OpenCL untill after mid 2010.

[m1stake]

As for OpenGL, I don't know anything about it. But reading these posts, I hope 3.1 gets all the features you want, Apple actually uses compatible graphics cards when they come out (not 2+ years after the fact), and has good drivers to go w/ them. If all this happens, new Macs would absolutely scream! But, if you want to see really great graphics, go outside. :D

It's the competitor for D3D. Unfortunately, it's not going anywhere fast, and certainly will not make an appearance in Snow Leopard. Not that it makes much difference, because anyone who knows anything plays games on Windows anyway.

[Analog Kid]

Yeah, yeah, yeah... It's hard to get excited about how fast it's going through the standards bodies when we know nothing about what it really is. I have high hopes for it, but I have no idea if it's going to do anything I dream it will.

[Analog Kid]

I read only a light weight "technical overview" that had some code examples. I looks to me like the standard itself is small and there is not to much to argue about it it. All the work will be in implementing it. Writing an OpenCL run time system looks to me like a very majoir effort
Is that tech overview public? I'd like to see it if you can point me to it.
[Edit: I found this one (http://s08.idav.ucdavis.edu/munshi-opencl.pdf), but would still be interested in others]
It's the competitor for D3D. Unfortunately, it's not going anywhere fast, and certainly will not make an appearance in Snow Leopard. Not that it makes much difference, because anyone who knows anything plays games on Windows anyway.
And anyone who knows anything else knows that games aren't the primary purpose of OpenGL.

[Chaos123x]

Will Snow leopard support CrossFire and SLI?

Or even better Hybrid versions of each.

[gkarris]

Is that tech overview public? I'd like to see it if you can point me to it.
[Edit: I found this one (http://s08.idav.ucdavis.edu/munshi-opencl.pdf), but would still be interested in others]

And anyone who knows anything else knows that games aren't the primary purpose of OpenGL.

Whoo, hooo! OpenCL!

Going to be great!

It's for computers AND has the word, "Open" followed by two CAPITAL letters!

(goes back to playing Atari "Asteroids"...) :eek:

:D

[guzhogi]

It's the competitor for D3D. Unfortunately, it's not going anywhere fast, and certainly will not make an appearance in Snow Leopard. Not that it makes much difference, because anyone who knows anything plays games on Windows anyway.

I knew what OpenGL is, just meant I didn't know much about version 3 of it. Hopefully everyone who designs & uses OpenGL will make it better. Awful that D3D is a proprietary, Microsoft-only API.

[8CoreWhore]

OpenCL on the iPhone could be put to work for voice recognition. :apple:

[javi.mahai]

No, apaerently Apple didn't do this all by themselves. What they did is work or a long time with Intel, nvidia, AMD and other before going to Kronos with their draft OpenCL spec in hand. They did the work in a small club.

I read only a light weight "technical overview" that had some code examples. I looks to me like the standard itself is small and there is not to much to argue about it it. All the work will be in implementing it. Writing an OpenCL run time system looks to me like a very majoir effort

Also while Applemight use a little of OpenCL in SL. I think before we see it widely used we will need for there to be some higher level libraries. It would a an almost no-brainer touse OpenCL to drive the physics in an OpenGL based game but first you'd have to re-write one of the common physics engines touse OpenCL. My gues is that uses will not see much from OpenCL untill after mid 2010.

Actually, OpenCL is mostly CUDA.

[Neuro]

I can't help but wonder, with all this potential extra and continual use of the graphics chip - won't the system run much hotter?

[shoulin333]

Well it's too bad games arn't still mostly on OpenGL, we could have good ports of games to mac easily, but with everything now on D3D, there will have to be 2 distinctive versions of games, which means no getting rid of bootcamp anytime soon.

[commander.data]

Will Snow leopard support CrossFire and SLI?

Or even better Hybrid versions of each.
I doubt it. The point of OpenCL and Grand Central is that you don't need Crossfire or SLI. These technologies are only useful for graphics rendering in games since games need to see a single combined GPU. GPGPU functions and OpenCL don't care if you have many slow GPUs or 1 big combined GPU. In fact, SLI and Crossfire would probably be detrimental because of wasted overhead. Instead, Grand Central will assign individual tasks to each separate GPU.

Now this won't help raw fps in games, but realistically that's hardly a huge priority. Games however will benefit by having physics or other computation done on 1 GPU, say the slower 9400M, with the graphics rendering done on the other GPU, say the 9600M. This will be faster than trying to SLI the 9400M and 9600M, dealing with the overhead, and having graphics rendering and physics compete for time in the combined GPU and having to deal with memory thrashing.

[shoulin333]

SLI is possible because Nvidia is now producing the chipsets for macs, we could easily see SLI in the new MacPro, wether it has anything to do with OpenCL though is unknown

[ChrisA]

Is that tech overview public?

I followed links from the OP to a slide show in PDF format. One of the sections in the slide show is "technical overview". It's about 15 slides in. There is an example of using a GPU to invert a square matrix. Lots of "set up" involved seting up command queues and creating an "instance" of the computer kernal and so on. So you'd better have quite a few square matrices that need to be inverted or it seems you'd be better off using a conventional library. But if your robot controller needs to solve a set of simultaneous equations 100 times a second OpenCL is the ticket.

I don't have the link but it was only 2 or 3 clicks from the macrumors front page.

[twoodcc]

this sounds great, for macs and iphones alike

[Nano-tube]

I believe that OpenCL will accelerate only specific tasks.
Though I am very curious to see what it will actually achieve, I am holding my fingers crossed for a 10% to 15% increase in OpenCL supported tasks.
I think it will be a respectable boost.

I wonder if OpenCL will be used by OS X to accelerate its GUI... :rolleyes:

[commander.data]

Actually, OpenCL is mostly CUDA.
Has someone actually compared the syntax and implementation between OpenCL and CUDA. It'd be interesting if they really are closely related since I believe the guy that Apple hired to develop the OpenCL spec was actually a former ATI guy. Although, I guess OpenCL, CUDA, CTM, and DX11 would all be fairly closely related since they are running on similar graphics architectures and are doing similar things.

SLI is possible because Nvidia is now producing the chipsets for macs, we could easily see SLI in the new MacPro, wether it has anything to do with OpenCL though is unknown
I don't see how nVidia producing a mobile chipset really has much effect on the Mac Pro. nVidia could only produce a mobile chipset now after having had to cajole a mobile FSB license from Intel. nVidia has tried to get a server FSB license to make a platform for Quadro, but Intel has steadfastly denied them. This isn't likely to change since the server market is a huge money maker and stability is also key so Intel wouldn't want to risk diluting anything.

nVidia themself has also refused to license SLI for use on Intel chipsets up to now. They've only recently allowed SLI on Skulltrail and the X58, because they had no way of developing their own chipset for these platforms. The Skulltrail platform was actually server based which they are locked out of and Intel hasn't given nVidia a QPI license so nVidia can't make high-end desktop chipsets.

If nVidia wants to get better performance on Mac, they really ought to optimize their drivers, since the HD3750 still outperforms the 8800GT in Core Image acceleration, and get the most out of their existing technology before trying to SLI things.

I believe that OpenCL will accelerate only specific tasks.
Though I am very curious to see what it will actually achieve, I am holding my fingers crossed for a 10% to 15% increase in OpenCL supported tasks.
I think it will be a respectable boost.

I wonder if OpenCL will be used by OS X to accelerate its GUI... :rolleyes:
GPGPU and OpenCL is only really effective in highly parallel, floating point tasks. Like encoding video. So we'll see it in all the media apps and the new Quicktime X. It won't likely help the GUI directly, although Apple could program some complicated but questionably useful physics effects. Like dropping stuff in the trash could make the can and the dock shake realistically. :rolleyes:

[Nano-tube]

I can't help but wonder, with all this potential extra and continual use of the graphics chip - won't the system run much hotter?

Excellent point.

[madmaxmedia]

Sounds pretty impressive...!

I'd love to see an OpenCL app on an iPhone, purely out of interest as to what it could do.

AppleMatt

It could copy and paste.

I kid! :p

[madmaxmedia]

I can't help but wonder, with all this potential extra and continual use of the graphics chip - won't the system run much hotter?

Only for very computationally intensive applications that would make use of OpenCL. But then you'd get a corresponding increase in speed.

In general use, it shouldn't make a difference.

[elppa]

Perhaps people are expecting the usual "standards" jockeying when Microsoft is involved in the process.

You'd be amazed at the progress we can make when we remove Microsoft from the equation. :D

Funny, but funny in a sad way for it has so often in the past been true.

[AidenShaw]

The question that comes to mind for me is whether the quality, stability, or reliability of the interface will have suffered from the tight schedule.

On the other hand, if OpenCL turns out to be a steaming pile of fresh dung - they can blame the schedule. ;)

I hope for Apple's sake that the adage "haste makes waste" doesn't become appropriate.

Wide exposure and deliberation on a standard aren't fast, but they help produce a solid spec that doesn't need a lot of revisions as it is implemented and released.

[demiphonic]

wow. I'm actually excited :D ....Apple is really moving at a pace! :eek:

[PeterQC]

Excellent point.

I'm sure it will have the exact opposite effect. Just look at what happened with the Air in video decoding with the new version: Lot of the encoding was transfered to the GPU and now the Air run a lot more cooler when viewing flash videos and movies. Now think about it, but for everything you do.

[Neuro]

Only for very computationally intensive applications that would make use of OpenCL. But then you'd get a corresponding increase in speed.

In general use, it shouldn't make a difference.

I'm just wondering if Photoshop, Final Cut Pro or 3D render apps start integrating with it closely, then surely the GPU will start to be pushed in the same way as when playing say a 3D game? We all know that the fans spin up.

Also, I think the new APIs are likely to inspire developers to come up with new more powerful apps which they may not have thought about before?

[inkswamp]

The more I read about what Snow Leopard is going to be, the more convinced I am that much of this was originally planned to be in Leopard. Maybe even some of this constitutes the "secret features" Steve talked about that failed to materialize. (Complain about that if you must, but I prefer they stay tight-lipped about features if they're not sure they can ship them instead of over-promising and under-delivering.)

I can't help but think that so many of these features sound like things that may have been yanked from Leopard at the last minute to meet their already slipping release date.

[grabberslasher]

Actually, OpenCL is mostly CUDA.

OpenCL is not mostly CUDA, it's a multi-device framework that taps into the programmable GPUs on all sides and other physics processors and CPUs available in the system.

While OpenCL and CUDA do similar things on the same part of silicon on the Nvidia boards, they're not the same code. It's meant to be *much* grander than CUDA alone.

[deannnnn]

that's fast. good job Apple!

It's not out yet...

[m1stake]

And anyone who knows anything else knows that games aren't the primary purpose of OpenGL.

Did I say it was? DirectX has more than its fair share of inadequacies, as does OpenGL. But it works perfectly well for what Apple does with it.

Games are a huge focus of D3D, so of course it will best an API that doesn't give games much importance.

[commander.data]

I'm just wondering if Photoshop, Final Cut Pro or 3D render apps start integrating with it closely, then surely the GPU will start to be pushed in the same way as when playing say a 3D game? We all know that the fans spin up.

Also, I think the new APIs are likely to inspire developers to come up with new more powerful apps which they may not have thought about before?
Well the idea is that for tasks that can be optimized for GPGPU, the GPU would be able to get them done faster than the CPU. So instead of the CPU working all out for half an hour, the GPU works all out for 5 minutes. If these are going to be tasks you were going to do anyways, then there is a net power savings. Of course, it may encourage more use of the GPU as more programs come up, but that's just progress.

[t0mat0]

From what I can see, if they push the timing for the legal stuff, they can go a bit more open by SIGGRAPH Asia in December, leaving Apple a chance to talk about it within Snow Leopard at MWSF. Till the legal stuff is done, no demos or specs.

[inkswamp]

From what I can see, if they push the timing for the legal stuff, they can go a bit more open by SIGGRAPH Asia in December, leaving Apple a chance to talk about it within Snow Leopard at MWSF. Till the legal stuff is done, no demos or specs.

Um, since when does Steve Jobs talk about Macs at Macworld anyway?

(I kid. I kid... sorta.)

[m1stake]

"OpenCL is an amazing new feature of OSX. ...Which is why I'm so excited today to ignore that and show you the watered down iPhone version!"

In seriousness, the ghosts of future past, present, and future, would have to give him one hell of a visit for that to possibly not happen :p

[Analog Kid]

As for OpenGL, I don't know anything about it.It's the competitor for D3D. ... Not that it makes much difference, because anyone who knows anything plays games on Windows anyway.... games aren't the primary purpose of OpenGL.Did I say it was?
yes

[javi.mahai]

OpenCL is not mostly CUDA, it's a multi-device framework that taps into the programmable GPUs on all sides and other physics processors and CPUs available in the system.

While OpenCL and CUDA do similar things on the same part of silicon on the Nvidia boards, they're not the same code. It's meant to be *much* grander than CUDA alone.

Actually, OpenCL is mostly based on CUDA with some Apple people doing some nifty dynamic compilation on top.

CUDA has been ported to the CPU as well to NVIDIA GPUS (and ATIs are getting there). Basically they are going to be doing super-threading algorithms that can run on any streaming device, or on a normal device that is not as fast but does not break the programming model so it can scale across different cores in the system (from different vendors, but it is geared towards NVIDIA initially)

Although, I always get a kick about working in a project and getting corrected from an outside source :-) hint.. hint...

[t0mat0]

Um, since when does Steve Jobs talk about Macs at Macworld anyway?

(I kid. I kid... sorta.)

Introduced the MacBook Air, Time Capsule, Apple TV Take 2, iTunes Movie rentals. iPhone firmware update.

I take your point! But the DC in WWDC hints, just as MW in MWSF. Hopefully this years MWSF can be a bit more World of Mac-esque. (Liveblog (http://gizmodo.com/364209/apple-macworld-2008-liveblog-archive) from Gizmodo - how it wasn't the best of keynotes (partly due to all the rumours being spot on and spoiling the surprise in a way)).

[m1stake]

<snip>

Not really. D3D and OpenGL are the two big standards, they do compete. The same is true of Windows and OSX, which arguably are aimed at different goals.

[Analog Kid]

I'm sure it will have the exact opposite effect. Just look at what happened with the Air in video decoding with the new version: Lot of the encoding was transfered to the GPU and now the Air run a lot more cooler when viewing flash videos and movies. Now think about it, but for everything you do.
It depends on how it's used, but right now power consumption is limited by the fact that the hardware is used very inefficiently. OpenCL and Grand Central will mean that all CPU cores and graphics cores will be able to run at a much higher load-- and that will get hotter. It got cooler doing the video decode because it's running on the GPU instead of the CPU.

If you're doing the same thing more efficiently it will get cooler, if you're doing more things more efficiently it'll get hotter but as someone above said, that's progress.
I believe that OpenCL will accelerate only specific tasks.
Though I am very curious to see what it will actually achieve, I am holding my fingers crossed for a 10% to 15% increase in OpenCL supported tasks.
I think it will be a respectable boost.
You're right that it'll be most effective for certain types of very parallel tasks, but the potential gains are much greater than that:
Mathematica and CUDA (http://www.marketwatch.com/news/story/Mathematica-Users-Get-100X-Performance/story.aspx?guid=%7B0D67D738-95F4-41FE-ABF9-BE858D93AF37%7D)
Wolfram is reporting that Mathematica is seeing a 900-9900% increase from CUDA. Granted Mathematica is essentially the perfect application for GPU acceleration, but that gives you an upper bound at least.
Not really. D3D and OpenGL are the two big standards, they do compete. The same is true of Windows and OSX, which arguably are aimed at different goals.
Snarkiness aside, my point was that by saying the state of OpenGL doesn't matter because everybody plays games on Windows you are saying that OpenGL is only useful for playing games.

[m1stake]

Then we can agree that it's good for making a pretty desktop background and maybe booting as well :p

I suppose I'd have a stronger point if I changed OpenGL to OSX. OSX has obviously not been built with games in mind because of the FPS difference between OSX verions and PC versions. While this is OpenGL's fault, Apple hasn't done a fantastic job with their implementation anyway, or their graphics drivers. Blame for everyone!