Intel working on OpenCL-capable Ivy Bridge chips bound for Apple's MacBook Air

The leader in OpenCL for GPUs and CPUs is actually AMD. Their implementation and Apple's are second to none.

Intel is dragging far behind. Nvidia is still trying to hedge its bets with CUDA but has full OpenCL 1.1 in it's driver stack.

What about the Mac mini?

It uses slightly more powerful chips than the air.

So IB's graphics are about 60% faster than SB, if Intel is to be believed. OpenCL is slower on chips of this calibre than modern processors, I don't think that's a huge deal maker or breaker. Unless for example you could use the iGPU at the same time as the dGPU, while the latter was doing the graphics work the former was helping the processor with texture decompression and such. But for now you can only use one with Intel. AMD lets you use both, but of course we won't see AMD processors in Apple computers at least for another generation, probably much more.

What are you talking about? There is no dedicated GPU in the MBA.

MBA does not have a quad core i7 at 3 GHz with the latest graphics chip. Rather, it's designed to be fast enough for many users with an emphasis on size and power consumption. If it's fast enough today, increasing the speed by 60% is big news to a lot of people - people for whom it was fast enough, but just barely... or people for whom it was not quite fast enough.

Darn - I was getting ready to buy an MBA to replace my MBP later this year. This is a big enough jump (especially when combined with the lower energy usage of Ivy Bridge) that I may have to hold off a while.

I know. I meant for the rest, hence using two chips at once, if Intel ever made that possible. 60% boost is nice, but on a GPU of this calibre OpenCL won't be faster than the same code running on a CPU anyways.

Apple's OpenCL implementation fully allows use of multiple OpenCL-capable GPU's simultaneously. If you have a Mac Pro, you can add additional GPUs to increase your OpenCL capabilities.

On machines with an integrated GPU and a discrete GPU, you have three targets for your OpenCL code, the CPU, the integrated GPU and the discrete GPU.

As for waiting for the new Air, I purchased the 11" Air with i7/256GB, and have been using it for all-day meetings along with programming sessions while connected to a larger monitor. You really don't want to wait unless you're buying it for entertainment value. It's an absolute game changer to have something so light, and yet so powerful.

Lion + 11" display + full-screen apps is much, much better than I imagined.

I don't believe you fully understand OpenCL on OS X.

On Ivy Bridge, OpenCL provides "separate" and "additional" SIMD compute resources that you would otherwise not have access to. Depending on model, you'll have 12 or more execution units, each running at a minimum of 600MHz. That's 4.8GHz that shares the same cache as the processor, so you won't even need to waste time loading the data onto the GPU and flushing the results when you're done. OpenCL will be HUGE for Ivy Bridge.

It's irrelevant.

High end machines will have dedicated GPUs, so the dedicated GPU will provide most of the graphics horsepower.

Low end machines will just use the Intel integrated GPU - so the gains are important.

Only a very small number of machines (such as the MBP) have both - and the ability to switch between them. The faster iGPU will be important when running in 'low power' mode.

You're also wrong about OpenCL not being powerful on a GPU like this. It depends on the task being accomplished. For the right tasks, the GPU (even in iGPU) will be very powerful. Look at the things that your graphics card is currently doing. The GPU (again, even in iGPU) can accomplish many things much faster than even a fast CPU. The same thing could apply to OpenCL if the tasks are properly chosen.

So much for the "Apple dumping Intel" rumor and gossipmongers.

It's fun.
"Apple to sign foundry deal for A6 processors",
"Apple to dump Intel and move to ARM",
"Apple going to use Intel in Macs for another few years and ARM on it's mobile devices".

if we are lucky, but this is Intel we are talking about it is hard to say what is 60% faster.

Where did you get that idea from? Seriously if you look hard enough you can find a benchmark that proves anything. It is not benchmarks that make the difference it is the overall system improvements that make or break the usefulness of OpenCL codes. So if you can do task x on the GPU in parallel with the CPU you win, sometimes the win is huge even if the GPU might be technically slower, in the end you have more cores working for you.

However you basic assumption that the GPU is slower is in many cases nonsense. Even slow GPUs can soundly outperform the CPU given a task that fits the execution resources of a GPU.

The whole point of OpenCL is to leverage the GPU with tasks that make sense to run on the GPU. The GPU is no place (today) to run branchy narrow code. GPUs are optimized for cores that can work on a lot of data in parallel.

On the AIRs and machines like it without a discreet GPU it is huge. It will effectively make the machine much more responsive over a wider array of workloads.

Actually I do wish that Apple would take AMD more seriously. It is good that they are using AMD GPUs but I see their long term Fusion plans as being far more in line with Apples needs than Intels. AMD is going very hard after the idea of heterogeneous computing and has detailed the long term plan to get there. Intel on the other hand seems to be resisting and being dragged into the tech by Apple.

Finally OpenCL does not equal texture computing. That is a gross misunderstanding of the wide array of uses for GPU acceleration.

Just wondering because I never paid attention to this in the past, but why did they move away from nvidia for graphics in general? The two companies don't even seem to be on speaking terms now and I recall reading something about it some time ago. I just can't remember details.

I think some misconceptions here are because some people may be confusing OpenCL and OpenGL.

OpenCL is still alive? It was announced years ago, yet no software as far as I know takes advantage of it. Certainly nothing on OS X I know of.

I'm not sure what the deal was but it apparently involved some arrogance on NVidias part. In any event these days AMD has all around better GPUs so it isn't like Apple is loosing here.

Apple basically set the standard for leveraging GPUs, OpenCL is the most widely used standard for leveraging GPUs these days.


Nope we are talking about Open Cee Ell

Then you don't know what is happening in the developer world. Do a little searching and you will find a surprising amount of support for OpenCL. When I hear statements like this I take it as an indication that the person complaining about OpenCL really doesn't understand what it is.

Yeah, odd that we've never had one of the Firepro options on the mac pro rather than the quadros with all of the ATI/AMD cards currently in macs. The reason I mention this is because the quadro cards have had massive driver, performance, and heat (to the point of where it causes artifacts and kills the card early) complaints.

http://support.apple.com/kb/HT4728

"Final Cut Pro X, Motion 5, and Compressor 4 all require an advanced graphics card (or GPU) for many of the features these application use."

http://www.phaseone.com/en/Software/...ade-Pro-6.aspx

"Interactive Speed. OpenCL support accelerates your editing tasks"

http://blog.cudachess.org/2010/02/ap...-3-and-opencl/

"Apple Aperture 3 is probably the first mainstream application to use OpenCL technology. Its not on the specifications or technical informations, but it use OpenCL for RAW decoding and processing"

OpenCL is going to have a slow uptake because like any optimisation code, developers need to think hard about how to use it best. Combine that with the fact that the vast majority of people have no OpenCL-capabale GPUs.

Ivy Bridge changes this because it means that 100% of all shipping x86 computers are OpenCL-capable. This makes it worthwhile to use it.

The only downside I've seen with OpenCL is that there aren't enough resources reserved for your UI so the UI doesn't refresh as smoothly. This happens if you have Motion and FCPX in the background doing their rendering.

Being able to use the IGP for the UI and the dGPU for OpenCL would be good or even just reserve 25% of the GPU for the UI.

I don't care about the "developer world." What's happening in the consumer world? Not an awful lot, judging by Marvin's post. Just Apple's video and photo apps and a couple of others. Photoshop doesn't use it. Neither does Premiere, Handbrake or any other power-hungry consumer app. So far, it's been a big "Openwhat?" for consumers, including power users.

You can't switch complex apps overnight to use it. They explain on the Handbrake site why they don't use OpenCL yet:

https://trac.handbrake.fr/wiki/SupportFAQ#pixiedust

I think it's best they don't use it because while it might double encoding speed, it would make the UI unusable during that period and it's nice to multi-task while encoding in the background.

For apps like Motion, Aperture, Final Cut etc. the real-time feedback is beneficial so you want to process as quickly as possible.

Adobe already use GPU computation but they went with CUDA (Nvidia-only) with their Mercury engine because OpenCL wasn't ready at the time of development.

It's not meant to be a technology to speed up everything, just things that really benefit from massively parallel processing and this tends to be graphics rendering more than anything because of the way the pixels are produced.

OpenCL was only put into action 3 years ago, which isn't a long time when you have to get a standard approved, a stable specification and production-ready drivers along with hardware support.

There's a physics engine called Bullet which will use OpenCL:

http://www.youtube.com/watch?v=RoIyRoANAy4

This will be used in games, possibly iOS games to get round the issues of PhysX being CUDA-only and Havok being owned by a company who doesn't know how to make good GPUs (Intel). OpenCL is at version 1.1 and this was only approved in June 2010. We'll see support in iOS devices in the coming year or two:

http://www.phoronix.com/scan.php?pag...item&px=ODc3NA

It's very much in the groundwork and experimentation stage just now and it's annoying to have to wait for consumer benefits but at least one day you'll be able to tell your kids or grandkids how lucky they are they didn't have to live without it and tell them stories of the waiting times you had to endure that they have no appreciation for.

Where do you think software comes from, Santas sac maybe?

Frankly there is little pay off for most consumers. However if you where to stop whining for a bit you could spend some time reviewing software that actually uses OpenCL. Of course much of this software is for professionals, so a Mc Donalds employee isn't likely to use these suites.

Well did you really expect technology this powerful and advanced to be implemented in consumer grade software this early? Seriously, the people that are leveraging OpenCL are the ones that can gain a strong competitive advantage from it's use.

Three years is an eternity in the software field. Nobody would want to install Leopard on a new system, and that's less than three years old. The only exception is Windows, where people clung desperately to an aging XP during the Vista fiasco.


Geez. What a nasty attitude. If I had a dollar for every piece of software that never made it off developers' hard drives, I could retire right now. Real Soon Now and vaporware are ridiculously common. Just listen to yourself. I wrote in my first post that OpenCL is still obscure three years after it was announced and most people don't know any software using it. You went on the warpath as if I'd killed your firstborn, despite later admitting "Frankly there is little pay off for most consumers."

An OS is quite different from a programming language. Its rate of adoption also tends to be slower. The appeal of OpenCL is that it's royalty free. Other than that go read wiki or something. The comparison to Leopard is a poor one. Leopard is an OS version, not a programming language.

A large number of video games take 3 years to go from concept to market even using established game engines. EA and Activision only manage yearly releases because they have about 500 people working on them at any given time.

Incremental OS releases are different because they don't have to change a lot of code to make a noticeable change. An incremental OS release is like a DLC for a game.


Vaporware is something that is announced and never released. OpenCL is very much in the open and in use. It's just not widely used - same as Thunderbolt.

CPUs and GPUs are merging and OpenCL gives you architecture independence to be able to run code to fully utilise the hardware you have. In fields of work where the highest levels of computation are required, OpenCL gives massive speedups.

Not really. It took almost two years from the time that the first iPhone beta was released until there where a good quantity of no trivial apps available. For those apps to stabilize it took even longer.

Again you fail to see this from the perspective of the developer which in the case of OpenCL is all that matters. In many cases developers simply aren't going to go out of their way to tell you which libraries they use in their software. For the user it simply doesn't matter. OpenCL is just one of many libs and headers that get folded into an app. For example a developer could make extensive use of BOOST and not tell you.

That is completely true (assuming the software is any good). From the standpoint of a user does it really matter? Another point of view is all the custom software that never leaves the place of development because it is considered strategic to a business. Either way you may never know which libraries are in use.

Again you simply don't know what you are taking about. OpenCL is widely used, the problem is you have to pull your head out of the sand and look around a bit. You say it is hardly used in the Apple whorl yet the facts are very different. For example grab otool and start looking at the apps and libraries that make installed software on your machine. Run it against CoreImage and you get something like:
XXXXXXXXXXXX$ otool -L CoreImage
CoreImage:
\t/System/Library/Frameworks/QuartzCore.framework/Versions/A/Frameworks/CoreImage.framework/Versions/A/CoreImage (compatibility version 1.0.1, current version 1.0.1)
\t/System/Library/Frameworks/CoreVideo.framework/Versions/A/CoreVideo (compatibility version 1.2.0, current version 1.7.0)
\t/System/Library/Frameworks/OpenCL.framework/Versions/A/OpenCL (compatibility version 1.0.0, current version 1.0.0)
\t/System/Library/Frameworks/OpenGL.framework/Versions/A/OpenGL (compatibility version 1.0.0, current version 1.0.0)
\t/System/Library/Frameworks/IOSurface.framework/Versions/A/IOSurface (compatibility version 1.0.0, current version 1.0.0)
\t/System/Library/Frameworks/OpenGL.framework/Versions/A/Libraries/libGLImage.dylib (compatibility version 1.0.0, current version 1.0.0)
\t/System/Library/Frameworks/Accelerate.framework/Versions/A/Accelerate (compatibility version 1.0.0, current version 4.0.0)
\t/System/Library/Frameworks/Foundation.framework/Versions/C/Foundation (compatibility version 300.0.0, current version 833.1.0)
\t/System/Library/Frameworks/ApplicationServices.framework/Versions/A/ApplicationServices (compatibility version 1.0.0, current version 41.0.0)
\t/usr/lib/libcrypto.0.9.8.dylib (compatibility version 0.9.8, current version 0.9.8)
\t/System/Library/PrivateFrameworks/FaceCoreLight.framework/Versions/A/FaceCoreLight (compatibility version 1.0.0, current version 1.4.2)
\t/usr/lib/libstdc++.6.dylib (compatibility version 7.0.0, current version 52.0.0)
\t/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 159.0.0)
\t/usr/lib/libobjc.A.dylib (compatibility version 1.0.0, current version 228.0.0)
\t/System/Library/Frameworks/CoreServices.framework/Versions/A/CoreServices (compatibility version 1.0.0, current version 53.0.0)
\t/System/Library/Frameworks/CoreFoundation.framework/Versions/A/CoreFoundation (compatibility version 150.0.0, current version 633.0.0)

Note the linkage to OpenCL above. Now ask yourself what apps use CoreImage. So you see the reality is that almost every app uses OpenCL to the extent that CoreImage does. This is OpenCL in CoreImage, where else it might be used is up to the person looking.


i've also told you repeatedly that you are wrong and you have no idea what you are talking about. OpenCL has been tremendously successful and is in many ways eclipsing CUDA as the primary way to leverage GPU's for compute tasks. OpenCL has been so successful that CUDA now looks like legacy software.

How is that contradictory, I've gone after your bogus claims because they are totally misleading. The fact is many consumer apps will never see a huge benefit from OpenCL, simply because they don't deal with data in a way that can every be leveraged by the facility. On the flip side Apple does use OpenCL in their libraries, so by default many programs do make use of OpenCL even if the core of the program does not.

The bigger problems I see it is why are you hung up on OpenCL? Seriously Apple supplies many libraries that get very modest use but are used by some apps. Even more importantly do you expect Apple and the various software vendors to always spell out what features they are using. Especially in the case of OpenCL which has fall back mechanisms so that code will execute even if the OpenCL compatible card isn't there. The whole facility is designed to be transparent to the user.

A lesser problem might be the fact that the term GPGPU computing is misleading. What is currently executing on the GPU is not really general purpose code. It certainly isn't normal GPU code either. GPU's display the best advantage on data parallel code, outside of image processing there isn't a lot of consumer apps that are structured such that the GPU makes a difference executing core code. So you see OpenCL used in places where developers can use the strength of the GPU to their advantage.

So am I on a warpath, not really I just get extremely angry when people knock OpenCL without knowing. OpenCL is so important that I don't even recommend computers without OpenCL support unless another feature is of overriding concern. You may call that overboard which is your right to do so, but I believe this board needs to be balanced and reflect reality. That reality is that Apple does use OpenCL and the standard has very wide industry support. I'd even go so far as to call it surprisingly successful.

A phrase I've heard many times and have grown to grasp over the years. Software is a process of development, very few software products are ever completely finished.


Outside of games look at what is or has happened in the internet browser wars. Things like Safari evolve over many years. They will continue to evolve and hopefully be debugged, for the as long as they are relevant. 3 years is really nothing.

More importantly they don't want to change a lot of code. Keeping an OS stable and compatible is a big deal. Apps can be refactored on a daily basis if the developer is up to it, you can not do that with an OS.

Well it is used in CoreImage and that is very widely used. I understand the point outside of that but I think people are extremely confused when it comes to OpenCL, it really isn't the technology to throw at a note taking app.

Now compare the noise we hear on these forums about OpenCL vs vecLib or Acclerate. Those are two of many, of the more obscure libs used by Mac OS/X.

Many are missing this point too. Right now GPU & CPU integration is about at the point where the floating point (FP) processor was first moved on chip. Some time after that Apple began to find that the FP processor was now faster than the older methods using fixed point math. The whole CPU has improved over the years but everybody now a days takes the FP unit for granted. Now we are in a similar state with the GPU, its value is only going to increase as it becomes more integrated in to the CPU. Or maybe I should say the CPU gets integrated into the GPU. A few years down the road and the CPU will start to look like a little logic block stuck in the corner of the SoC. More importantly Apple appears to be at the forefront of driving technology in this direction.