Apple's OpenCL standard near complete in just six months
Apple has reportedly set an industry record by moving its OpenCL parallel computing standard from its beginnings to imminent approval in half a year, paving the way for its inclusion in Mac OS X Snow Leopard.
Representatives of the standards overseer Khronos Group and its partners used Austin's SC08 high-performance computing show this week to tout the speed of the draft format's completion to Macworld and gathered members of the press.
Khronos' presentation from the event now shows that the first feature-complete edition of OpenCL (Open Compute Language) was submitted for ratification in October, or just four months after it was first proposed alongside the unveiling of Snow Leopard. The operating system will use the technology to accelerate general-purpose tasks using both individual processor cores as well as video chipsets inside its systems.
The group's chief, NVIDIA executive Neil Trevett, also suggests through statements that the total development time for OpenCL was unusually quick as a whole. A timeline provided by the organization shows Apple having worked on the initial proposal with AMD, Intel and NVIDIA in an undetermined amount of time before the June unveiling, but Trevett himself indicates that the total time involved was just six months -- unprecedented speed for a certification process often known more for the caution involved.
?If you go to some other larger standards bodies, it?s quite normal for a standard to take five years or more,? he explains. ?That?s quite commonplace. You actually have to really push to get it down to eighteen months. Our record was [twelve] months, up to now; we?ve done this one in six.?
Much of the rapidity is directly attributed to Snow Leopard. For Intel, the prospect of seeing OpenCL already in a shipping operating system for 2009 has been a strong lure. Employees have "divorced [their] families" and worked extreme levels of overtime to complete a draft that many said would be "impossible," according to the chipmaker's Tim Mattson.
But while congratulating themselves for their accelerated work, neither Apple nor the other standard developers have yet to outline how OpenCL will specifically benefit Macs.
At present, NVIDIA's vendor-specific equivalent language, CUDA, is used primarily to speed up data and scientific calculations by using the generalized nature of GeForce video hardware alongside the main CPU. OpenCL will at a minimum open this to AMD's ATI Radeon hardware, but isn't yet known to be addressing any specific roles in Mac OS X. Mattson notes that it can often be used for processing physics, such as those found in games.
It's also said that OpenCL can scale down to smaller devices, including smartphones, as long as the processor or graphics chipset are capable of supporting this kind of acceleration.
As quickly as the standard has gone through industry checks, however, Apple and Khronos both have a pair of additional steps to undertake before their format is entirely ready. The individual vendors will first have to sign off on the completed draft and suggest changes as necessary, after which a specification can reach the public; later, Khronos will also introduce some tests to ensure that hardware and software developers properly follow the standard.
No timeline has been supplied for when these would take place, though the ratification will take a minimum of 30 days to finish. This would nonetheless give Apple enough time to include OpenCL for both the official mid-2009 release window for Snow Leopard as well as the rumored early schedule.
Representatives of the standards overseer Khronos Group and its partners used Austin's SC08 high-performance computing show this week to tout the speed of the draft format's completion to Macworld and gathered members of the press.
Khronos' presentation from the event now shows that the first feature-complete edition of OpenCL (Open Compute Language) was submitted for ratification in October, or just four months after it was first proposed alongside the unveiling of Snow Leopard. The operating system will use the technology to accelerate general-purpose tasks using both individual processor cores as well as video chipsets inside its systems.
The group's chief, NVIDIA executive Neil Trevett, also suggests through statements that the total development time for OpenCL was unusually quick as a whole. A timeline provided by the organization shows Apple having worked on the initial proposal with AMD, Intel and NVIDIA in an undetermined amount of time before the June unveiling, but Trevett himself indicates that the total time involved was just six months -- unprecedented speed for a certification process often known more for the caution involved.
?If you go to some other larger standards bodies, it?s quite normal for a standard to take five years or more,? he explains. ?That?s quite commonplace. You actually have to really push to get it down to eighteen months. Our record was [twelve] months, up to now; we?ve done this one in six.?
Much of the rapidity is directly attributed to Snow Leopard. For Intel, the prospect of seeing OpenCL already in a shipping operating system for 2009 has been a strong lure. Employees have "divorced [their] families" and worked extreme levels of overtime to complete a draft that many said would be "impossible," according to the chipmaker's Tim Mattson.
But while congratulating themselves for their accelerated work, neither Apple nor the other standard developers have yet to outline how OpenCL will specifically benefit Macs.
At present, NVIDIA's vendor-specific equivalent language, CUDA, is used primarily to speed up data and scientific calculations by using the generalized nature of GeForce video hardware alongside the main CPU. OpenCL will at a minimum open this to AMD's ATI Radeon hardware, but isn't yet known to be addressing any specific roles in Mac OS X. Mattson notes that it can often be used for processing physics, such as those found in games.
It's also said that OpenCL can scale down to smaller devices, including smartphones, as long as the processor or graphics chipset are capable of supporting this kind of acceleration.
As quickly as the standard has gone through industry checks, however, Apple and Khronos both have a pair of additional steps to undertake before their format is entirely ready. The individual vendors will first have to sign off on the completed draft and suggest changes as necessary, after which a specification can reach the public; later, Khronos will also introduce some tests to ensure that hardware and software developers properly follow the standard.
No timeline has been supplied for when these would take place, though the ratification will take a minimum of 30 days to finish. This would nonetheless give Apple enough time to include OpenCL for both the official mid-2009 release window for Snow Leopard as well as the rumored early schedule.
Comments
...but Trevett himself indicates that the total time involved was just six months -- unprecedented speed for a certification process often known more for the caution involved.
guess in that certification process they were playing Steve's favorite song by Devo, "Whip It"... "Crack that whip..."
The group's chief, NVIDIA executive Neil Trevett, also suggests through statements that the total development time for OpenCL was unusually quick as a whole.
It's all done with GPUs now so it's about 10 times quicker.
A timeline provided by the organization shows Apple having worked on the initial proposal with AMD, Intel and NVIDIA in an undetermined amount of time before the June unveiling, but Trevett himself indicates that the total time involved was just six months -- unprecedented speed for a certification process often known more for the caution involved.
This is what happens when you deal with industry pushing companies like GPU companies. Maybe Intel should get them to handle all the USB 3 stuff too.
Reasons that I think the developments needs to get out sooner rather than later is that apps need to be changed to take advantage of the improvements and these improvements are huge. Being able to accelerate video encoding by a factor of 2, some things by a factor of 10. This is just an unprecedented performance jump and would take well over 2 years to do in a normal hardware cycle.
If it's going to take a while before developers know how to develop the apps properly then the APIs need to be in place ASAP. WWDC can focus on teaching devs how to use it.
As the presentation from the unix guy had said, 'forget everything you know about multi-threaded computing (turns out developers didn't know that much anyway)' - funny but also quite true. Parallel computing is the way forward. Hardware has dictated that and developers need to make the transition.
I think it's interesting the way he put to forget about multi-threading. I figured they would have a lot in common. I guess we'll have to wait and see?
But while congratulating themselves for their accelerated work, neither Apple nor the other standard developers have yet to outline how OpenCL will specifically benefit Macs.
So, the rush, money, and stress to the devs all have no apparent ROI at this point?
Laughable at best.
Thank goodness I own a SR MBP, it will be able to take advantage of OpenCL, I'm not sure if intel GMAs will be able to harness the power of OpenCL?
No, Intel graphics chips won't be able to do anything - if they can, it won't be significant, likewise with any ATI chips Apple used like the Radeon X1600, 2600 Pro etc.
No, Intel graphics chips won't be able to do anything - if they can, it won't be significant, likewise with any ATI chips Apple used like the Radeon X1600, 2600 Pro etc.
FWIW, OpenCL accelerates things across CPUs too, not just GPUs. So you won't get a 120x speedup, but you'll easily get 6x.
This is what happens when you deal with industry pushing companies like GPU companies. Maybe Intel should get them to handle all the USB 3 stuff too.
Good point.
And, for me, the whole Open CL married with the i7 and prospect of a 4800 series GPU/Nv 280 card allows for not only a hardware bump but a major software bump to allow the hardware to perform even faster.
Nice one, Apple...nice to see them pushing software efficiency and not just the hardware equation.
Snow Leopard is shaping up to the one of the most exciting releases ever in this respect.
Basically, Apple pushes 'Cuda' or something like it or better forward because they drive the platform. So, owning a Mac isn't and never has been, merely a hardware equation vs PC. It's the software, stupid.
Lemon Bon Bon.
It's nice to finally see computing evolve into two new areas that
are delivering big speed improvements.
GPU currently with drawing screen elements and processing small video or image files and the prospects of leveraging more of the GPU's programmability for general purpose apps.
SSD- Finally a laptop can have a storage system that is faster than the typical 7200/10k rpm HDD drives. It removes a bottleneck that has hobble mobile performance.
Snow Leopard certainly addresses the GPU via OpenCL and hopefully the much smaller application and library files of SL will help with enabling small and fast SSD to be the primary OS drive with HDD being the large storage that you need.
These and the improvements in multithreading the cores is going to certainly change how people design fast computers. I know of one company selling Mac Pros with SSD drives in RAID configuration for the OS and then utilizing massive external storage. This makes so much sense and it also makes sense to view the GPU as an accelerator of everything now rather than just graphics.
I'm certainly geeked for Snow Leopard.
It's nice to finally see computing evolve into two new areas that
are delivering big speed improvements.
GPU currently with drawing screen elements and processing small video or image files and the prospects of leveraging more of the GPU's programmability for general purpose apps.
SSD- Finally a laptop can have a storage system that is faster than the typical 7200/10k rpm HDD drives. It removes a bottleneck that has hobble mobile performance.
Snow Leopard certainly addresses the GPU via OpenCL and hopefully the much smaller application and library files of SL will help with enabling small and fast SSD to be the primary OS drive with HDD being the large storage that you need.
These and the improvements in multithreading the cores is going to certainly change how people design fast computers. I know of one company selling Mac Pros with SSD drives in RAID configuration for the OS and then utilizing massive external storage. This makes so much sense and it also makes sense to view the GPU as an accelerator of everything now rather than just graphics.
I'm still not sold on SSD.
It's still too low of storage, still extremely expensive, still hasn't been proven for longevity (hasn't been out long enough). I'm sure all of this will change eventually, but for now it's hard to make these your main internal drives. I have 4 terabytes in my linux server... it cost a grand total of $440 for the 4 1 terabyte drives. Keep in mind that is all internal. The speed improvements aren't significant enough to justify the price yet.
I see it coming down though... so don't attack too hard!!!
I'm still not sold on SSD.
It's still too low of storage, still extremely expensive, still hasn't been proven for longevity (hasn't been out long enough). I'm sure all of this will change eventually, but for now it's hard to make these your main internal drives. I have 4 terabytes in my linux server... it cost a grand total of $440 for the 4 1 terabyte drives. Keep in mind that is all internal. The speed improvements aren't significant enough to justify the price yet.
I see it coming down though... so don't attack too hard!!!
No arguments will come from me buddy, "pragmatic" is my middle name. I still think the SSD market is a bit like the wild wild west. The cheap stuff doesn't perform and the stuff that does perform is so expensive it's ridiculous.
When newegg is selling 80GB 72k laptop drives for $60 and the SSD drives are 10x that price I'm a bit nonplussed about the hype. I'm willing to pay 2-3x the price of HDD but faster launch and boot times don't affect me that much. I rarely reboot my computer unless i've updated something and launch times don't bother me all that much. I do want to see more competitors for SSD though to reduce the price because they are not worth the premium that i'm seeing right now.
OpenCL is not appropriate for the majority of tasks but can help a lot in, say, Apple's pro applications. Grand Central has a wider use. As far as the Cocoa Frameworks are optimized using this two technologies, all Cocoa applications will benefit. But to take the most of it they need to use Grand Central and/or OpenCL explicitly. This will take some time.
Also, Apple will need to concentrate in providing efficient implementations and good documentation on the new technologies, finish the new Finder, recompile all other bundled apps for 64 bit, finish the new 64 bit kernel etc. This leaves no time for full Framework optimization across the board, new versions of Aperture, final Cut and other pro apps (some of them are still Carbon and need a significant work to move to Cocoa). I doubt they will have resources for iPhoto, iMovie, iTunes etc. before SL is released. ZFS is a no-go for the client and may be not ready for the prime time on the server side as well.
We have to look at it this way: Snow Leopard is a solid foundation for the future. It marks a strategic move for Mac OS and may turn out to be a showcase for the entire industry. But the changes will take several years to complete and will take more than one 10.x releases to reach maturity.
No, Intel graphics chips won't be able to do anything - if they can, it won't be significant, likewise with any ATI chips Apple used like the Radeon X1600, 2600 Pro etc.
Why not ATI? What's wrong with X1600/2600 (apart from being 2-3 years old now and not top of the line)? They still do OK in my MBP.
They weren't designed for general purpose computing. AFAIK, ATI's Stream SDK compatible cards are:
AMD FireStream 9250, AMD FireStream 9270, ATI FirePro V5700
and ATI FirePro V8700, and ATI Radeon HD 4000 series graphics cards.
Nvidia on the other hand have pretty much all their GPUs fully supporting CUDA and by extension OpenCL since the Geforce 8 series:
http://www.tgdaily.com/content/view/37612/139/
70 million+ GPGPUs in the market. All of Apple's notebook lineup will contribute another 15 million or so. Unless Apple plan on using the higher-end ATI cards, I think Nvidia is the best way forward ATM.
We know Apple are planning on leveraging GPU computing because Jobs described the Nvidia additions in terms of floating point performance. This is not generally a measure of graphics capability but raw computation.
They weren't designed for general purpose computing. AFAIK, ATI's Stream SDK compatible cards are:
We know Apple are planning on leveraging GPU computing because Jobs described the Nvidia additions in terms of floating point performance. This is not generally a measure of graphics capability but raw computation.