RSSSo what is 'OpenCL,' Apple's next enhancement to Mac OS X 10.6?
by Scott M. Fulton, III
On Monday, Apple made mention of a curious new technology it said would help accelerate the development of CPU-to-GPU process sharing, calling it OpenCL. But the lack of information about what it is makes us all the more curious.
In its press materials released Monday, Apple made mention of a technology it called OpenCL, whose purpose was reportedly to enable so-called GPGPU functionality -- the ability for graphics processors to handle some of the heavy computing tasks normally threaded to CPUs. Since Apple is already involved with a project that's part of Khronos Group's OpenGL, specifically to enable GPGPU functionality, our first reaction was that this must be a typo.
But from Apple's standpoint, the "C" was not a typo, although at first we received two competing definitions of it: Open Compute Library or Open Computing Language.
Here's how Apple's Web site currently defines it: "Another powerful Snow Leopard technology, OpenCL (Open Compute Library), makes it possible for developers to efficiently tap the vast gigaflops of computing power currently locked up in the graphics processing unit (GPU). With GPUs approaching processing speeds of a trillion operations a second, they're capable of considerably more than just drawing pictures. OpenCL takes that power and redirects it for use in high-performance computing applications like genomics, video encoding, signal processing, and simulations of physical and financial models."
Apparently yesterday, Apple began issuing corrections to press sources that made reference to "Open Compute Library," telling them the definition should actually read "Open Computing Language." That memo apparently has yet to reach its own webmaster.
Late yesterday, CEO Steve Jobs himself issued one of those corrections to The New York Times, telling reporter John Markoff that OpenCL will indeed be something big and new -- something "way beyond what Nvidia or anyone else has, and it's really simple."
That's curious, because Nvidia had been Apple's partner in the Khronos Group, although arguably Nvidia members have taken the lead in that group. As far back as 2004, Apple had been tasked with building developer tools for that group's GPGPU standards, which were -- and are -- being built around the OpenGL umbrella.
Jobs stopped short of revealing any more details, beyond merely dissing his company's partner -- or maybe, former partner -- in GPGPU technologies.
Though more information could naturally be expected to have been revealed at the company's developers' conference, it was not. That led AnandTech's Ryan Smith to resort to a seldom used tool in modern technology marketing -- logic -- to piece together what Apple may have meant.
"We have heard that Apple has wanted to add full GPGPU support to the Mac for some time now (having been one of the first companies to embrace early GPGPU usage for their video editing applications)," Smith wrote, "but we have also heard that they are unhappy about the incompatibility between the GPGPU languages. They don't want to have to write two of everything, nor do they want their developers doing so."
Smith went on to speculate that Nvidia and AMD (the parent of ATI) would have to be on board with this idea in order for it to work in the first place. But that was before Jobs made his Nvidia comment to the Times; and Apple's relationship with AMD up to this point has historically been non-existent.
So an entire WWDC goes by with Apple having introduced us to the notion of the prospect of the possibility of "one more thing," without any clear guidance as to how it intends to pull it off.
16 Comments
To me, OpenCL is a LLVM abstraction spanning on both x86 CPU, NVidia CUDA and AMD close to the metal API.
You write your computing intensive task using Open Computing LANGUAGE, then compile it to OpenCL bytecode. And, at runtime, a JIT compiler specific to your own hardware (which CPU do you have, which GPU) compiles that byte into actual machine code optimized for your actual hardware.
Apple already does this in Leopard for OpenGL, balancing the graphic computation between the CPU and the GPU depending of the hardware available and the current execution context using LLVM.
Score: 0
I think at the very most this will give Intel and AMD time to work with its vendors to develop a hardware based solution.
Do I think this is a little to late for the times, actually I do.
I think this will allow Apple users to get some additional graphical power out of their Mac(s) for the time being.
This will give Apple an advantage since they will have software and hardware based features once Intel provides them with hardware.
I suspect we won't see GPU's interacting with CPU for another 3 years. I just hope Microsoft is smart and works on something similar within their next Direct X update.
A combination of Software and Hardware to get rid of the faults of one another is most likely the best solution.
Score: 0
You do realize that everything Apple is showing developers is NOT revealed in the keynote and PRs? OpenCL is definitely being discussed with (at least some) devs under NDA.
And it's kind of hard to be a "one more thing" when it's one of the five or so features already announced.
Score: 0
Sounds awesome. The software IMO is not Apple's problem. If they would just get rid of the whole proprietary hardware bit, I think they could really do well. I personally think this is one of the main factors holding them back. I know I would at least be dual booting it if I could install it on my PCs.
Score: 0
Indeed, would prob shift a few tasks to Apple OS if I could run it on a decent laptop but as it stands XP/Vista it is...
Score: 0
Is your inability to load MacOS on your PC based upon ethics or a lack of ability? I loaded v10.4.8 onto my DELL Laptop and while I did have to invest quite a bit of time into the endeavor, I was very pleased with the results. Or - does your PC have a hardware issue? I was lucky that my DELL laptop was compatible.
Score: 0
I wasn't aware that it would function at all with PC hardware...You have peeked my curiosity now though!
Score: 0
Really, this does make sence considering how close Apple has gotten with Intel.
Intel is moving into the GPU business, and with a closely intergrated CPU system, then it only make sence that Apple would want to come up with a system to let programmers be able to use both. No? Larrabee is Intel's move into this space. Intergrated GPU with CPU. AMD's Fusion is headed that way to.
Score: 0
Apple wants to help the CPU and GPU work together with software while AMD wants to do it with hardware. Hardware will always win over software. Its better to have a hardware firewall, its better to have a hardware physics card, its better to do hardware decoding, so AMD/ATI's hardware CPU/GPU should be faster than Apples way of doing things.
Score: 0
This is fairly basic material...
You can do a specified task 'better' (i.e. faster) in hardware but it costs a ton of money to dev the circuits (and way more to integrate down to chips) whereas software is adaptable - and typically kills off specialized hardware gains within 1 year on generic silicon.
Score: 0
"software is adaptable - and typically kills off specialized hardware gains within 1 year on generic silicon"
What is your basis for that statement? If you really meant that what I think you meant, which is, software developers can implement an arbitrary algorithm in software that will be run on a generic CPU that will run as fast as the same algorithm implemented in hardware you obviously don't know what your talking about. If you would like explanation of why this is true then I will provide, but I suspect you not interested anyway.
Score: 0
Well, thats kind it... Intel is looking to do the same. It's nice to see all the companies getting behind this sort of thinking. Use both the GPU and CPU as much as possible. =)
Score: 0
His point is probably the universal irrational fear of hardware that high-level "programmers" (script kiddies) have.
When they start talking "abstraction" like it's an advance over "writing code" I say bye byyye... bye bye liberals. :)
They "invent" shiney new paradigms per second, connectivity solutions, ASSOCIATIVITY, (I s***e you not) infrastructure, virtualization and other M$ spoon-fed esoteric clueless middleman hyperbola, I can't wait until theyre all driven into space by an enormous mutant star goat.
Or a dot-com purge.. I mean bust. That usually works.
Score: 0
Wow a passionate damning, always good when founded and I absolutely agree - with the exception of associativity (can't actually find associativity even in the oxtail english dictionary!)- being an 'Enterprise' scale BI architect long, long before the liberals (or Dr Kimball come to that) had devised their 'best practice' labels, it is data association (or associativity...perhaps ) tricks that fuel my interest in using the number crunching abilities of GPUs to solve a problem that will potentially change BI architectural approaches at their very very grass roots. On that basis Steve better be on the ball or hope that not too many people actually know why they would want this type of technology ;-)
Score: 0
phew, yet another crap coming from rotten apple
Score: 0
"way beyond what Nvidia or anyone else has, and it's really simple" - this guy bores me so...
Score: 0