OpenCL ties Apple to NVIDIA

[AppleInsider]

Apple's push to accelerate Mac performance in innovative ways is likely to bind the company even closer to NVIDIA's GPUs, which already support the OpenCL technology Apple will be releasing in Mac OS X 10.6 Snow Leopard.

This fall, Apple passed over Intel's integrated chipset to use the new NVIDIA 9400M controller in its unibody MacBooks. The controller's integrated GPU is significantly more powerful than the Intel GMA X3100 integrated graphics processor Apple had been using, providing as much as 6.2 times the graphics performance of the Intel chip according to benchmarks Apple touted at its launch.

The new NVIDIA GPU-integrated controller, as with all of NVIDIA's GeForce 8 series or better GPUs, supports the company's CUDA design, which makes it capable of running OpenCL tasks that offload processing to the GPU rather than the CPU.

Earlier this month, AMD also announced official support for OpenCL on its ATI GPUs that support Stream technology. It expects to deliver its OpenCL-compliant compiler and runtime early next year as part of its ATI Stream 1.4 SDK. AMD says Stream support is already built into "millions" of the company's Radeon graphics cards, but delivering that latent processing power will require graphics drivers with support for AMD's Stream SDK as well as the OpenCL tools.

Mac users have never been at the top of the list for receiving GPU driver support from ATI or NVIDIA; in many cases, Apple has delivered its own driver software that often does not take full advantage of the hardware features available on other platforms. However, the company's latest MacBook collaboration with NVIDIA demonstrated some of the best graphics hardware support yet on the Mac, suggesting new interest from Apple in pushing its platform's performance via GPUs.

OpenCL helps solve two problems in that regard. First, Apple is delivering it as a vendor neutral, cross platform technology that overlays the proprietary, incompatible efforts of NVIDIA's CUDA and AMD's ATI Stream, making it much easier for third party developers to support both and therefore broadly deliver GPGPU software acceleration. Both NVIDIA and AMD are interested in broadening the use and utility of their GPUs beyond just graphics and gaming; OpenCL solves an key industry interoperability problem in a way that key vendors have indicated they are happy to support.

OpenCL also solves a problem for Apple: the company wants to continue be able to work with both AMD and NVIDA, using whichever GPUs offer the best price and performance. By providing a powerful cross platform parallelism technology that can target both company's products, Apple can deliver the biggest performance leap possible in Snow Leopard without tying itself permanently to one vendor. In the short term however, it appears NVIDIA will help Apple achieve GPU acceleration fastest on the Mac. AMD's support for OpenCL will help broaden the technology's critical mass on other platforms, including Linux, reserving the potential for Apple to use AMD GPUs in the future.

[ View this article at AppleInsider.com ]

[zunx]

Apple and nVidia, THIS IS WHAT IS NEEDED:

Mac OS X support for the nVidia multicore cards (Tesla Personal Supercomputer):
http://www.nvidia.com/object/tesla_c...solutions.html

Practical example for Mathematica:
http://www.wolfram.com/mathematica

being 100 times faster at:

http://www.physorg.com/news146247669.html

nVidia says that it is possible, but only if Apple allows it.

[Johnny Mozzarella]

I was debating if I should upgrade my 1st gen MacPro with an NVIDIA GeForce 8800 GT ($279) or the cheaper ATI Radeon HD 3870 ($199).

After reading this article I'm leaning towards the 8800 GT since it sounds like ATI will be slower in delivering drivers that will take advantage of OpenCL in Snow Leopard.

Now I'm debating if I should just wait till Snow Leopard comes out or the Mac Pro is refreshed and see if any new cards become available.

[stonybrookadam]

You sir, are an idiot.

Let me get this straight:

*OpenCL is open. Anyone can implement. NVIDIA or AMD or Intel
*Apple's driver support from AMD and NVIDIA has historically been lacking (??)
*AMD announces its intention to write OpenCL drivers.

How the HELL does this "tie" Apple to Nvidia?

There is a shipping product from Nvidia, but AMD is trying to catch up. Tying would be availability from only one source for the indeterminate future. Last time I checked, 10.6 isn't shipping. Perhaps the AMD drivers will be shipping before 10.6!

Hardly unbiased or even linguistically correct journalism.

[refulgentis]

Originally Posted by stonybrookadam

You sir, are an idiot.

Let me get this straight:

*OpenCL is open. Anyone can implement. NVIDIA or AMD or Intel
*Apple's driver support from AMD and NVIDIA has historically been lacking (??)
*AMD announces its intention to write OpenCL drivers.

How the HELL does this "tie" Apple to Nvidia?

There is a shipping product from Nvidia, but AMD is trying to catch up. Tying would be availability from only one source for the indeterminate future. Last time I checked, 10.6 isn't shipping. Perhaps the AMD drivers will be shipping before 10.6!

Hardly unbiased or even linguistically correct journalism.

This article has such a disconnect between the title and the body, it makes me think someone wrote a more sensational headline without reading the article. The article is also very unusually poorly written, it's by far the worst I've read in 3 years of reading AI, so much so that I registered just to point it out.

[Johnny Mozzarella]

Originally Posted by zunx

Apple and nVidia, THIS IS WHAT IS NEEDED:

Mac OS X support for the nVidia multicore cards (Tesla Personal Supercomputer):
http://www.nvidia.com/object/tesla_c...solutions.html
...
nVidia says that it is possible, but only if Apple allows it.

Thanks for that link.
It would be nice if OpenCL apps in Snow Leopard could use these cards for parallel processing.

[Rayz]

So when you said 'OpenCL ties Apple to NVDIA'

what you really meant was:

'It doesn't'

That is one desperately cheap trick ....

[John the Geek]

Originally Posted by stonybrookadam

You sir, are an idiot.
How the HELL does this "tie" Apple to Nvidia?

Apple invented OpenCL. NVidia has been working close with Apple. NVidia already have OpenCL compliant hardware, whereas ATI does not.

How doesn't this tie Apple to NVidia?

[John the Geek]

Guys, there's more than one definition of the word "tie."

It also means simply "to connect." You people are reading too much into it, and then getting upset at yourselves for being gullible.

[Ahruman]

Originally Posted by refulgentis

This article has such a disconnect between the title and the body, it makes me think someone wrote a more sensational headline without reading the article. The article is also very unusually poorly written, it's by far the worst I've read in 3 years of reading AI, so much so that I registered just to point it out.

Yes, there’s been a definite drop of quality since AppleInsider tied itself to Daniel Eran Dilger, a.k.a. Prince McLean.

[Techslacker]

Originally Posted by John the Geek

Apple invented OpenCL. NVidia has been working close with Apple. NVidia already have OpenCL compliant hardware, whereas ATI does not.

How doesn't this tie Apple to NVidia?

For the most part stonybrookadam is right but seriously nothing has been released to prove one way or another so it's silly to argue over it.

How many times did we hear that Apple was tied to the ppc? Anything can happen between now and release time.

[Techslacker]

Originally Posted by Ahruman

Yes, there’s been a definite drop of quality since AppleInsider tied itself to Daniel Eran Dilger, a.k.a. Prince McLean.

I don't know about that. It still seems it's a requirement here to write from an extreme biased point of view and most of the time warping reality misleading readers.

[ltcommander.data]

Originally Posted by John the Geek

Apple invented OpenCL. NVidia has been working close with Apple. NVidia already have OpenCL compliant hardware, whereas ATI does not.

How doesn't this tie Apple to NVidia?

Where do you get the information that ATI's hardware is not OpenCL compliant?

http://www.businesswire.com/portal/s...80&newsLang=en

ATI already has OpenCL running in their labs and plans to incorporate it in the Steam SDK for all currently supported Steam processors. What's more, ATI is providing tools for developers to switch from their own proprietary Brook+ language to OpenCL.

http://www.tgdaily.com/content/view/38764/140/

Which brings up the point that ATI has publicly said they are replacing their proprietary CTM implementation with it's Brook+ language in favour of OpenCL. nVidia however is continuing to push their proprietary C for CUDA language in addition to OpenCL. If you are asking who is more committed to OpenCL it's ATI since they aren't doing their own thing on the side.

And ATI hardware has support for 64-bit floats in their HD3xxx and HD4xxx series whereas nVidia only supports 64-bit floats on the GT200 series which isn't available on Mac. The nVidia 8xxx and 9xxx series only support 32-bit floats. 32-bit floats is what is used to process the graphics in games, but 64-bit support can be used in GPGPU programs. So ATI hardware as it stands is generally more full featured for GPGPU as well.

And yes, it does seem strange that there is another nVidia/Apple article only 5 days after the last one, yet this one really doesn't add much new information.

[gescom]

Originally Posted by ltcommander.data

Where do you get the information that ATI's hardware is not OpenCL compliant?

http://www.businesswire.com/portal/s...80&newsLang=en

ATI already has OpenCL running in their labs and plans to incorporate it in the Steam SDK for all currently supported Steam processors. What's more, ATI is providing tools for developers to switch from their own proprietary Brook+ language to OpenCL.

http://www.tgdaily.com/content/view/38764/140/

Which brings up the point that ATI has publicly said they are replacing their proprietary CTM implementation with it's Brook+ language in favour of OpenCL. nVidia however is continuing to push their proprietary C for CUDA language in addition to OpenCL. If you are asking who is more committed to OpenCL it's ATI since they aren't doing their own thing on the side.

And ATI hardware has support for 64-bit floats in their HD3xxx and HD4xxx series whereas nVidia only supports 64-bit floats on the GT200 series which isn't available on Mac. The nVidia 8xxx and 9xxx series only support 32-bit floats. 32-bit floats is what is used to process the graphics in games, but 64-bit support can be used in GPGPU programs. So ATI hardware as it stands is generally more full featured for GPGPU as well.

And yes, it does seem strange that there is another nVidia/Apple article only 5 days after the last one, yet this one really doesn't add much new information.

Read the above once again & again

[Hattig]

Originally Posted by stonybrookadam

You sir, are an idiot.

Let me get this straight:

*OpenCL is open. Anyone can implement. NVIDIA or AMD or Intel
*Apple's driver support from AMD and NVIDIA has historically been lacking (??)
*AMD announces its intention to write OpenCL drivers.

How the HELL does this "tie" Apple to Nvidia?

There is a shipping product from Nvidia, but AMD is trying to catch up. Tying would be availability from only one source for the indeterminate future. Last time I checked, 10.6 isn't shipping. Perhaps the AMD drivers will be shipping before 10.6!

Hardly unbiased or even linguistically correct journalism.

Damn straight.

Indeed everything I had read pointed to AMD getting their OpenCL implementation out before NVIDIA. AMD have been executing excellently with their drivers for a few years now.

[ajmas]

Truth is, if you are have any concern for graphics then Intel graphics chips aren't worth touching with barge pole. From my experience they are usually integrated chips, offering the bare minimum of graphics capability.

[Wiggin]

Originally Posted by Techslacker

I don't know about that. It still seems it's a requirement here to write from an extreme biased point of view and most of the time warping reality misleading readers.

But that didn't used to be the case. Ever since AI joined up with Roughly Drafted we've been seeing more and more of these sensationalist and biased articles. If I wanted that I'd read Roughly Drafted. AI is still one of the better Apple-centric web sites, but they lost much of their edge over the past several months.

[UltimateKylie]

This article is SO Wrong and used as a way to defend Apple choosing faulty nVidia chips.

First of all I will mention that ATI have dropped their standard to focus purely on OpenCL while nVidia will continue to persue CUDA in addition to OpenCL. How that helps an Apple created standard, I do not know.

In addition this article suggest Apple choose the best chips and is willing to work with both ATI/AMD and nVidia. Fact is most people agree that ATI 4000 series line up is the best lineup currently. They have the best benchmarks, especially in the midrange which is what the 9600 on the Macbook Pro is.

Further fact is that ATI has full license to create an Intel Chipset and AMD would gladly accept the business. In addition all Intel Chipsets support ATI Graphics. Either solution would work including possibly a dedicated graphics card on both the macbook and macbook pro (just having a more powerful version on the pro)

[TheNipponese]

The weirdness of this article's title aside, the only thing that can help Apple on the GPU front would be the availability to use cards from third party vendors that also work with Windows machines.

This Mac-only card game has been killing Apple since forever.

[ksec]

OpenCL was only a minor thing. AMD / ATI could have had OpenCL as well. The problem is Intel hate AMD much more then Nvidia.
Therefore ATI will never produce a chipset for Intel, and Apple have been sick and tired of Intel iGFX.

That is the main reason for Nvidia 's tie. Not to mention Nvidia proberly have something very interesting on their Roadmap.....

[mdriftmeyer]

Originally Posted by stonybrookadam

You sir, are an idiot.

Let me get this straight:

*OpenCL is open. Anyone can implement. NVIDIA or AMD or Intel
*Apple's driver support from AMD and NVIDIA has historically been lacking (??)
*AMD announces its intention to write OpenCL drivers.

How the HELL does this "tie" Apple to Nvidia?

There is a shipping product from Nvidia, but AMD is trying to catch up. Tying would be availability from only one source for the indeterminate future. Last time I checked, 10.6 isn't shipping. Perhaps the AMD drivers will be shipping before 10.6!

Hardly unbiased or even linguistically correct journalism.

Here, here!

Did HuffingtonPost just buy out AppleInsider? What a crock.

[mdriftmeyer]

Originally Posted by ksec

OpenCL was only a minor thing. AMD / ATI could have had OpenCL as well. The problem is Intel hate AMD much more then Nvidia.
Therefore ATI will never produce a chipset for Intel, and Apple have been sick and tired of Intel iGFX.

That is the main reason for Nvidia 's tie. Not to mention Nvidia proberly have something very interesting on their Roadmap.....

Horsecrap. Nvidia and AMD have never owned an Operating System and to conclude they could have invented OpenCL is myopic.

OpenGL came from SGI because it was originally designed for IRIX. SGI extended it and opened it up.

OpenCL comes from Apple because it was originally designed for OS X. Get it?

[GregAlexander]

Originally Posted by AppleInsider

The new NVIDIA GPU-integrated controller, as with all of NVIDIA's GeForce 8 series or better GPUs, supports the company's CUDA design, which makes it capable of running OpenCL tasks that offload processing to the GPU rather than the CPU. <snip>
AMD also announced official support for OpenCL on its ATI GPUs that support Stream technology.

So... sorry for the entirely selfish and practical question:

My 2 year old MacBook with an NVidia GeForce 8600 (from memory) ... that would be an "8 series" so will work with OpenCL?

And my wife's year old iMac with the "AMD Radeon HD 2400"... does that support "Stream Technology"?

Both would get a nice boost from Snow Leopard if that's the case.

[Futurist]

Setting aside the semantic wrestling around the word "tie", aren't we missing the point? It seems to me that we are in the midst of one of those rare technology development and adoption convergences that will make computing much the better for all of us.

To wit: (i) Apple/Khronos deliver OpenCL, providing a single API for developers which is compatible with both CUDA and Stream, (ii) Apple punts integrated Intel GPUs, (iii) Apple adopts hybrid integrated/discrete GPU approach with late 2008 MacBook Pros, (iv) Apple releases OSX 10.6 in Q1 (?) 2009, with greatly enhanced multicore and CPU/GPU scheduling via Grand Central, finally transforming (prospectively) GPUs into GPGPUs and accelerating desktop and mobile parallelism, (v) Intel delivers 45 nanometer Penryn multicore CPUs with processing power equal to or greater than the MPUs they replace with lower thermal dissipation profiles, thus enabling (iii) and (iv) without blowing the systems' thermal budgets and, finally, (vi) NVIDIA's integrated and discrete GPUs can (according to NVIDIA) operate simultaneously (even if Apple's current Leopard OS distro does not support this capability).

Add to this the fact that discrete GPUs are available from Apple with up to, for example, 512MB of GDDR3 memory in the MBP and that integrated GPUs will ultimately (Snow Leopard?) be able to take advantage of 8GB of fast DDR3 SDRAM (because Intel's Penryn/Centrino 2 platform is designed to address at least that much main memory) and we have a lot to be thankful for from a graphics and compute performance perspective.

[GregAlexander]

Originally Posted by Wiggin

Ever since AI joined up with Roughly Drafted we've been seeing more and more of these sensationalist and biased articles. If I wanted that I'd read Roughly Drafted. AI is still one of the better Apple-centric web sites, but they lost much of their edge over the past several months.

I enjoy Roughly Drafted articles - the technological understanding seems good & the extrapolations can be quite insightful. Unfortunately they do have a strong bias that can reduce the article quality, making some stretches that are a bit too much - perhaps negating anything useful in an article if you can't pick where these are (which happens sometimes).

The bias is not always obvious so I'll have to get used to looking at the authors on AI. I'd rather see a co-branding (stick up a "roughly drafted" logo so it's clearly different) to make it clear that it's a different class of article.

[winterspan]

Originally Posted by John the Geek

Apple invented OpenCL. NVidia has been working close with Apple. NVidia already have OpenCL compliant hardware, whereas ATI does not.

How doesn't this tie Apple to NVidia?

WRONG. Although they are certainly outdated, The ATI cards used in the iMac and Mac Pro along with any other 2xxx, 3xxxx, or 4xxxx series cards are entirely capable of using ATI's existing GPGPU (Brook+/Stream) SDK and thus almost surely OpenCL.

Originally Posted by stonybrookadam

*snip* How the HELL does this "tie" Apple to Nvidia?
There is a shipping product from Nvidia, but AMD is trying to catch up. Tying would be availability from only one source for the indeterminate future. Last time I checked, 10.6 isn't shipping. Perhaps the AMD drivers will be shipping before 10.6! Hardly unbiased or even linguistically correct journalism.

Originally Posted by refulgentis

This article has such a disconnect between the title and the body, it makes me think someone wrote a more sensational headline without reading the article. The article is also very unusually poorly written, it's by far the worst I've read in 3 years of reading AI, so much so that I registered just to point it out.

I completely agree with both of you. This is a total nonsense article. AMD actually pledged to work with Apple and fully support OpenCL -- and stop marketing their proprietary GPGPU SDK --- BEFORE NVIDIA. Meanwhile, nVidia continues to market CUDA and only said they would be supporting both OpenCL and CUDA. And even more Ironically, OpenCL is the VERY THING that PREVENTS Apple from being tied to a specific GPU vendor in the future for GPGPU processing.

[retroneo]

Stupid article.

[winterspan]

Originally Posted by ltcommander.data

Where do you get the information that ATI's hardware is not OpenCL compliant?

http://www.businesswire.com/portal/s...80&newsLang=en

ATI already has OpenCL running in their labs and plans to incorporate it in the Steam SDK for all currently supported Steam processors. What's more, ATI is providing tools for developers to switch from their own proprietary Brook+ language to OpenCL.

http://www.tgdaily.com/content/view/38764/140/

Which brings up the point that ATI has publicly said they are replacing their proprietary CTM implementation with it's Brook+ language in favour of OpenCL. nVidia however is continuing to push their proprietary C for CUDA language in addition to OpenCL. If you are asking who is more committed to OpenCL it's ATI since they aren't doing their own thing on the side.

And ATI hardware has support for 64-bit floats in their HD3xxx and HD4xxx series whereas nVidia only supports 64-bit floats on the GT200 series which isn't available on Mac. The nVidia 8xxx and 9xxx series only support 32-bit floats. 32-bit floats is what is used to process the graphics in games, but 64-bit support can be used in GPGPU programs. So ATI hardware as it stands is generally more full featured for GPGPU as well.

And yes, it does seem strange that there is another nVidia/Apple article only 5 days after the last one, yet this one really doesn't add much new information.

Indeed, this is all true. And one more thing I wanted to add is that while Nvidia's GTX280 card is indeed powerful in raw processing capability, ATI's 4870 is *much faster* in terms of double-precision/FP64 calculations.
Below are the specs for the dedicated GPGPU/stream processing cards from each respective company and which use the same GPUs as the Nvidia GTX280 and the ATI 4870.

Nvidia Tesla C1060 card (Based on GTX 280)
933 Gigaflops (FP32)
78 Gigaflops (FP64)

AMD FireStream 9270 card (Based on Radeon 4870)
1200 Teraflops (FP32)
240 Gigaflops (FP64)

That is 3X faster!. Also, considering the 4870 is much cheaper (and cheaper to produce) than the GTX280, I wouldn't be at all surprised if that was the case for these dedicated cards. Most scientists, engineers, and other professionals planning to use these GPGPUs for parallel processing will be using double-precision calculations, so unless nVidia is hiding something, I can't see why ATI wouldn't dominate them in this new market. The only other empirical metric to compare them would be power efficiency, and I believe the nVidia card uses about 15% less energy, but with 3X the double precision performance, the performance/watt is still heavily in ATI's favor.

[GregAlexander]

It's interesting how I read your speed stats there and think "Wow, ATI is much faster", gravitating towards the speeds shown - when what you said was actually quite different:

Originally Posted by winterspan

while Nvidia's GTX280 card is indeed powerful in raw processing capability, ATI's 4870 is *much faster* in terms of double-precision/FP64 calculations.

So in terms of OpenCL and Snow Leopard, is it raw processing capability or double-precision calculations that will make the most difference?

[kaiwai]

Originally Posted by stonybrookadam

You sir, are an idiot.

Let me get this straight:

*OpenCL is open. Anyone can implement. NVIDIA or AMD or Intel
*Apple's driver support from AMD and NVIDIA has historically been lacking (??)
*AMD announces its intention to write OpenCL drivers.

How the HELL does this "tie" Apple to Nvidia?

There is a shipping product from Nvidia, but AMD is trying to catch up. Tying would be availability from only one source for the indeterminate future. Last time I checked, 10.6 isn't shipping. Perhaps the AMD drivers will be shipping before 10.6!

Hardly unbiased or even linguistically correct journalism.

I agree - what it sounds like was if all those of us who don't have Nvidia GPU's in our computers, we're SOL.

It would be nice to know what the situation actually is for OpenCL and whose GPU's are actually supported - from the above headlines it sounds as if the only ones supported are Nvidia.

I hope Apple don't rush it; heck, I'd be happy if the didn't release it until the end of next year - beginning of the following year; it isn't as though 10.5 is so desperately out of date that it needs to be scrapped quickly.

[GregAlexander]

Originally Posted by kaiwai

I hope Apple don't rush it; heck, I'd be happy if the didn't release it until the end of next year - beginning of the following year; it isn't as though 10.5 is so desperately out of date that it needs to be scrapped quickly.

Whenever they release it I'm betting it'll be exclusive to new Apple hardware at first, and branch out later. And it will not be pushed as a replacement to 10.5, instead as a functionally identical alternative IF you have the right hardware and peripherals/drivers.

[fuyutsuki]

Guys, anyone who thinks AMD will be releasing OpenCL aware drivers for their older cards is nuts.

I like ATI like the next guy who was 100% Radeon for years. But if the roles were reversed: Nvidia would hold out their labs features for FUTURE products too. In that case, and given the Nvidia chipset now in Apple's ENTIRE portable line (iMac next?) the relationship between Apple and nvidia is clearly closer than ever.

ATI and Nvidia alike have left the Mac until last for new silicon and crucially: new DRIVERS for donkey's years. Nvidia's sigificant partnership on the new portables could well change this.

But let's believe it when we see it. That means 10.6 and not marketing speak about labs.

[mdriftmeyer]

Originally Posted by fuyutsuki

Guys, anyone who thinks AMD will be releasing OpenCL aware drivers for their older cards is nuts.

I like ATI like the next guy who was 100% Radeon for years. But if the roles were reversed: Nvidia would hold out their labs features for FUTURE products too. In that case, and given the Nvidia chipset now in Apple's ENTIRE portable line (iMac next?) the relationship between Apple and nvidia is clearly closer than ever.

ATI and Nvidia alike have left the Mac until last for new silicon and crucially: new DRIVERS for donkey's years. Nvidia's sigificant partnership on the new portables could well change this.

But let's believe it when we see it. That means 10.6 and not marketing speak about labs.

AMD/ATi released support for their Streams API in their mainline driver under Linux, on December 10, 2009 via their Catalyst 8.12 driver. They will shortly be releasing their OpenCL implementation in the mainline driver. I'd expect by the end of March 2009.

[winterspan]

Originally Posted by GregAlexander

It's interesting how I read your speed stats there and think "Wow, ATI is much faster", gravitating towards the speeds shown - when what you said was actually quite different:
So in terms of OpenCL and Snow Leopard, is it raw processing capability or double-precision calculations that will make the most difference?

I apologize for the confusing language. Before looking up the numbers, I was almost certain that the Nvidia GT200 (GTX 280 brand)chip was faster than the RV770 (4870 brand) in single-precision calculations, but those numbers are accurate and straight off each company's website. Perhaps ATI's Stream drivers were improved or the benchmark was updated. Either way, it is complicated because their GPU architectures are very different and which one is faster will most likely depend on the application, especially considering the Nvidia GT200 also has ~25% more memory bandwidth. On double precision calculations, the difference will be much greater as the nVidia card really takes a hit in performance.

Regarding OpenCL, most consumer applications and non-scientific image, video, and audio processing (think Photoshop, Videoencoding, etc) will not need the extra accuracy that double precision floating point (64-bit) provides. But double precision is vital for most scientific research/simulations and engineering applications. In the end, it really won't matter because *ANY* high-end GPU will be so much faster than a modern quad-core CPU it won't even be funny (in tasks that can take advantage of it of course)

Originally Posted by kaiwai

It would be nice to know what the situation actually is for OpenCL and whose GPU's are actually supported - from the above headlines it sounds as if the only ones supported are Nvidia.

Apple's OpenCL, Nvidia's CUDA, and ATI's Stream/Brook+ are all very similar. They take advantage of the flexible processing architecture that was created for DirectX10.
In nVidia's case, supported cards include all of the 8-series, 9-series, and newer, including the GTX280 and GTX260. (codename G80, G84, G86, G92, G94, G96, and GT200)
For ATI, supported cards include all of the 2000-series, 3000-series, and 4000-series. (codename R600 and codename R700)

Originally Posted by fuyutsuki

Guys, anyone who thinks AMD will be releasing OpenCL aware drivers for their older cards is nuts.

I don't see why they wouldn't, as the architecture is pretty much all the same (R600 and R700), and all of the 2000, 3000, and 4000 series cards already support their existing "Stream/Brook+" GPGPU development kit, so it's an easy step to full OpenCL support.

From ATI's webpage:

Q: Will the AMD FireStream SDK work on previous generation hardware?

A: To run the CAL/Brook+ SDK, you need a platform based on the AMD R600 GPU or later. R600 and newer GPUs are found with ATI Radeontm HD2400, HD2600, HD2900 and HD3800 graphics board.

[kaiwai]

Originally Posted by GregAlexander

Whenever they release it I'm betting it'll be exclusive to new Apple hardware at first, and branch out later. And it will not be pushed as a replacement to 10.5, instead as a functionally identical alternative IF you have the right hardware and peripherals/drivers.

I very much doubt that. Moving to 64bit drivers does not mean re-writing whole drivers specially for 64bit. Sure, existing drivers when written which made assumptions of 32bit will need to be tweaked but I'd say that a good number of drivers that were written to be compiled into 32bit or 64bit shouldn't experience problems.

[GregAlexander]

Originally Posted by kaiwai

I very much doubt that. Moving to 64bit drivers does not mean re-writing whole drivers specially for 64bit. Sure, existing drivers when written which made assumptions of 32bit will need to be tweaked but I'd say that a good number of drivers that were written to be compiled into 32bit or 64bit shouldn't experience problems.

I wasn't so much meaning the 64bit issue as other issues and OpenCL drivers. Basically I mean that
* it's a significant underlying rewrite with associated challenges, and
* at the same time Apple doesn't intend for everyone to switch since it'll look identical, and
* IF Apple is trying to push Snow Leopard out the door quickly
then they may choose a launch method different to what they normally do.

Testing on one or 2 products exclusively allows for a quicker launch. It also means they'll get real-life feedback of issues with their new system (in addition to their own internal testing). As they extend support for other systems they get to test them thoroughly and not have a huge number of Mac fans jump aboard as soon as they possibly can.. despite any warnings Apple may give.