Oh my god, i am so late to this news, and yet no one in the discussion has mention it yet! I have mention in another post about the possibility of apple building its own chipset as well. I am so happy for a rumor to surface on similar ideas.
If it was really Graphics inside Intel chipset that makes apple unhappy then this should be it.
Everyone seems to think about the P.A Semi buy out. But no P.A Semi is not the expert in Video Processing field, Apple is using other technology that it has licensed, and that is
PowerVR!
Long time ago PowerVR issued a press release stating "someone" has acquired the sole right for using its SGX and VX technology in specific area.
Now i know this is vague and it seems in many instance and clues we know it was Apple. But we have yet to see product using it. Many thought iPhone chipset was it but it turns out they are using older MBX core from PowerVR.
So may be the deal was for apple to use SGX and VX on Desktop and Laptop Computers?
For those who dont know, SGX have some have decent performance for its size and power usage. And While SGX may not be the best GPU in its class, VX has score the lowest power usage for decoding 1080p H.264 Video. It is one of the best if not the best video decoder on the market.
Apple could use ARM core, SGX and VX Core inside its Mac Chipset.
This way we would have common platform top to bottom. And Quicktime X to support hardware acceleration of H.264 video on all apple products, Mac, iPod and iPhone.
Last of all, Apple is not developing anything. I'm pretty sure with all the talk lately about system-on-a-chip (SoC) from Intel and the press, that Apple is using a new chip from Intel that is a SoC. That means the CPU, GPU, southbridge, northbridge are all on one chip. This makes sense for laptop redesigns since you only need one main chip for the motherboard, therefore changing size and thermal considerations and simplifying the overall layout. This chip is probably very customizable to fit new notebook designs and only available for Apple from Intel.
An SoC from Intel makes the most sense to me.
You were doing well until the last paragraph. SoC chips are good for things like iPhone but not your average notebook/laptop. Atom 1.6Ghz anyone ?. Please... that is turning the clock back another 4 years or so.
SoC will not likely get MBP type power/performance anytime soon, if ever.
PA Semi doing a specialised SoC makes a lot of sense. The MIPS based Samsung chip in the iPhone would be a nice one to go. PA Semi also has experience and talent to do system Chipsets.
Look at what ServerWorks chipset did for Xeon...
I figured CUDA or OpenCL type technologies coming with Snow Leopard has a lot to do with chipset redesign. A great chipset will make the cores sing like what Serverworks did for X86 servers those years ago. Plausible ?. I think so.
they already have to write code in universal binary to be used on intel and powerpc chips... are they going to now need to write code in universal tertiary?
I'm sure the changes will just be upgraded motherboards of some sort, but after juse getting intel processor on board in the last few years, it makes absolutely zero sense to abandon them this quickly for a new type processor.
I think major overhaul is unlikely, more like Apple wants to have more control over design. I don't see OS compatibility being an issue Apple hasn't thought about, more likely the chips will deliver more features that you just don't find anywhere else & once you have them you will wonder how you lived without.
Might be nice to have some chips more properly designed to take advantage of all aspects of EFI. Such a thing would actually make VMWare & Parallels more happy as it might boost their performance.
Don't forget that EFI is the key in compatibility. Windows isn't compatible with EFI already, but EFI is able to emulate BIOS to Windows. EFI makes a lot more possible than people realize.
I think that if apple goes back to powerpc cpu's (thats what this pa semi company made)then they will alienate a lot of developers who had just finished the transition to intel cpu's. I think if they did that the developers would walk away from apple and say its just not worthit to have to learn everything again.
I think it would slow down apples gain in marketshare.
You were doing well until the last paragraph. SoC chips are good for things like iPhone but not your average notebook/laptop. Atom 1.6Ghz anyone ?. Please... that is turning the clock back another 4 years or so.
SoC will not likely get MBP type power/performance anytime soon, if ever.
The guy you responded to did accurately describe what a SoC is. S0C do not imply any one performance point, in fact a SoC can often be had that is higher performance than the equivalent assemblage of discrete parts. One huge factor in SoC performance is the reduction in power usage due to the reduction in the number of off chip drivers used in an implementation.
With respect to mainstream computing only recently has the amount of die space exceed what is required to make the CPU itself. This means that it is possible to go the SoC route for high performance computing. Possible but this doesn't mean that Intel has publicly announced such a chip. In any event I do not believe that any of Intels announced or un-announced S0C have anything at all to do with this rumour.
If the new Macs are to have something special then by definition it can't be publicly available Intel stock.
Quote:
PA Semi doing a specialised SoC makes a lot of sense. The MIPS based Samsung chip in the iPhone would be a nice one to go. PA Semi also has experience and talent to do system Chipsets.
Look at what ServerWorks chipset did for Xeon...
Again MIPS has nothing to do with iPhone. That is ARM based technology.
As to PA Semi and its ability to do a chip set ala ServerWorks that is a possibility. The problem is there is no advantage for Apple in a straight up chip set implementation. PA Semi would have to be adding something to the mix that would not be expected in a system chip set. So we come back to just what is that something special given that it would need to be more than just a run of the mill chip set.
Quote:
I figured CUDA or OpenCL type technologies coming with Snow Leopard has a lot to do with chipset redesign. A great chipset will make the cores sing like what Serverworks did for X86 servers those years ago. Plausible ?. I think so.
Not unless this supposedly new chip set has the facility to actually accelerate the types of codes that OpenCL is thought to support. It still comes down to, will the chip set have some sort of built in acceleration and what exactly will that acceleration be. In could be general video decoding hardware to something like a new run at a vector engine.
Quote:
PS: SoC from Intel still sucks in perf/watt.
Are you sure you are up to speed on Intels latest offerings? It is pretty hard for a S0C not to lead in a performance per watt metric.
Am I the only person who has noticed that the Apple logo is positioned incorrectly, or at least differently, in the purported photo of the new Mackbook Pros?
Custom Apple designed chipsets for product classes that do not exist yet. Come on people; use your brains.
I fully agree with your statement.
Indeed, the purchase of P.A. Semi seems to indicate potential products in areas Apple does not currently address.
For example, Microsoft seems to be making great strides in automotive media control/entertainment systems (e.g. Ford's 'InSync' and in other markets known as 'Blue and Me'). I see no reason why this type of application shouldn't be part of Apple's evolutionary development.
Based on what technology is already employed in their existing products (e.g. GPS in the iPhone 3G), the likes of Apple, P.A. Semi and a partner like Bose, should be able to bring to market sophisticated, yet elegant, and desirable "appliances" for automobiles, aircraft and trains in a very short amount of time.
And speaking of Apple appliances, I envision P.A. Semi has the potential to help Apple in the future differentiate the iPod, iPod Touch, iPhone and the much rumored tablet PDA class devices.
I agree with most others that Apple's desktop and laptop systems will continue to use Intel CPUs and chipsets for the near future, but may employ P.A. Semi or other third-party technologies to enhance differentiation from its competitors.
For example, Apple has applied for patents on multi-gesture controls. Then witness HP's "all-in-one" TouchSmart touch-enabled PC. Yawn...
Personally, fingerprints on screens is a "pet peeve" of mine, so I think Apple is going to go one better where you do gestures in front of a more sophisticated iSight camera and "touch" without physically touching the screen. Better yet, "type" without touching a physical keyboard. It would all be done by tracking the hands and eyes.
You may disagree with me on the above, but I truly believe it is not a matter of if, it's when. And as we all know, Apple is the kind of company to bring this type of technology to the market first and (mostly) succeed at it.
My guess for Snow Leopard is not just the CUDO/OpenCL to leverage GPGPU codes, but to optimise multi-core and potentially heterogeneous multicore CPU systems in future. Many would like to see some form of advanced AltiVec which X86 tries to play catch up.
In portable systems, advanced power management on the chipset is going to be important and Intel is really behind on this tech.
Wow, this thread is an incredible demonstration of how people can't read! And on top of that it displays the staggering levels of technical ignorance that make do in the place of actual knowledge.
Quote:
Originally Posted by majortom1981
I think that if apple goes back to powerpc cpu's (thats what this pa semi company made)then they will alienate a lot of developers who had just finished the transition to intel cpu's. I think if they did that the developers would walk away from apple and say its just not worthit to have to learn everything again.
Apple isn't going back to PPC and nobody is suggesting that they will. PA Semi licenses core designs they need, revises them to incorporate their advanced power saving design techniques, and then integrates them into a system-on-chip. Apple has likely directed them to work on ARM-based designs for iPod/iPhone, and (as speculated in this thread) may also be using them to design northbridge/southbridge chips for their Intel-processor-based Macs.
Since the 80s Apple was doing its own chipset and motherboard design which included memory controller, I/O interfaces, DSPs, and processor interconnects. The G5's chipset was quite an impressive piece of tech. In the last few years Apple has acquired Racer and now PA Semi, among others. They will leverage Intel's processors, but they are going back into the motherboard design business... which these days means the big northbridge/southbridge chip (or chips) since they are so heavily integrated. We'll have to wait and see whether this will include a GPU. I find it unlikely since laptop/desktop class GPUs are still pretty sophisticated and competing with ATI/nVidia will be challenging... but who knows what they've licensed for inclusion on their SoC. I still remember rumours from a couple years ago of vector processing cores (like DSPs, the Cell processor's SPEs, or the processing elements of modern GPUs) coming to Mac chipsets, but they never materialized. Still, something like Grand Central and OpenCL is precisely what Apple would need to put into place to allow 3rd party developers to leverage this kind of technology directly. This kind of tech would obviate the need for specialized sound cards or H.264 decoder chips, and it would be far more flexible.
If this is true, it is interesting timing. Intel's Penyrn chips (like all predecessors) have a northbridge that includes a memory controller. The next chips, coming starting this fall, will for the first time (for Intel) include a memory controller on the processor die and will switch to Intel's Quick Interconnect technology. This is a pretty major change in the northbridge and thus would mean a significant piece of work on Apple's part for their chipset which lasts only a single generation of machines. Perhaps they licensed the bus interface rather than designing it themselves, but this seems like unfortunate timing. Why not just wait until Nehalem has arrived and design around the new interconnect which is (hopefully) going to be standard going forward, much like HyperTransport on AMD. Speaking of which, Apple used HyperTransport and was on the design board... I guess they'll eschew that in favour of Intel's new equivalent tech.
My previous post above has me thinking again about this issue. (yes I know it is early in the morning)
Think of it this way Apple develops a chip set with a Vector Engine facility that it owns out right. A vector engine that is effectively decoupled from what ever CPU they are running. Think of it as an Alt Vec (AV2)for the new century.
This would give Apple some serious advantages that we can highlight below.
1.
One the AV2 facility would be a programmable unit suitable for vector computations. Hopefully as such it would handle both long and short vectors and be fully 64 bit. This would give Apple a common platform for application acceleration no matter what platform, be it AMD, Intel, PPC, ARM or whatever. All Apple would need is its own Vector Engine with interchangeable front side buses.
2.
Lets face it Intels vector implementation sucks and really hasn't caught up with Alt Vec of old. Even if current implementations suck less there is still the issue of running vector code through your main CPU complex. Vector code just sucks up a lot of bandwidth and CPU resources leading to laggy systems. Further even with all of Intels latest improvements decoding all that Blue-Ray can offer up is not possible. So such hardware may very well mean the implementation of BlueRay on Apple hardware.
3.
Owning your own Vector Engine means owing the instruction set and the implementation of those instructions. This means hardware acceleration of certain codes could be nothing more than a single instruction in the vector engine. Things like dot products, security codes and the like come to mind. While Apple might call the unit a vector engine in reality it is a location where anything not handled well by the main CPU can be accelerated. Since Apple owns the hardware if it wants to implement an MD5 instruction or some sort of convolution instruction it can. Of course there is always the issue of trade offs but this being a specialized computational unit Apple would be free to accelerate what they need accelerated.
4.
One can only imagine what an Alt Vec like engine could do on modern processes and with a whole different set of design parameters. They would have at their disposal much more chip real estate, far lower power usage, and on chip connectivity to supporting units. It would not be impossible to imagine an implementation that results in a multi threaded hardware platform. Lets say they are able to implement 4 or 8 Alt Vec equivalent cores in such an implementation would that excite people? How about 64? Since each would be its own computational unit the number of cores could be adjusted to suit the hardware. Thus an iPod might get one core while a Mac Pro might get 32.
5.
Having an optimised Alt Vec core available, that can do decent video acceleration, means that Apple can program to one acceleration framework across all devices. When Apple goes to a full SoC for the iPods (they will) it means that they control the video accel hardware (as a general purpose vector engine) thus reducing licensing costs and limiting the diversity of components. Since the unit is a vector engine that means not only do video apps have the potential for acceleration but so do any other apps that can map code onto vector instructions.
6.
While the acceleration of certain codes by moving the code to a GPU works and sometimes amazingly well it is not without problems. One issue being the constantly changing nature of GPU's. A generalized vector facility can make up for many of a GPU's short comings when working effectively with the CPU. Especially if that vector unit is now multi threaded and can more easily handle the things that GPU's do well. I realize that modern GPU have a lot of cores running in parallel but I see something as modest as an 8 core AV2 as offering up a huge number of advantages that would make such preferable for many developers. In a way this would be something like Sony's cell implementation. The main difference is that it would reside in Apple controlled hardware decoupled from whatever is the main CPU.
All in all I see this as a real potential. Such an approach would eliminate the weaknesses of a GPU only approach to acceleration and could ideally be maintainable across multiple generations of hardware and multiple ISA of hardware. In the case of cell each vecotr processor there is rather modest in its power usage though still unacceptable for an iPod. This is where PA Semi and having control of the hardware implementation come into their own. PA knows low power and combining that with an optimised instruction set could yield a very nicely performing Alt Vec like unit that sips power.
Well it is another idea to offer up. WE might all be getting excited about a rumour that has no basis in fact, but it is nice to imagine the evolution of Apple hardware.
Intel has been pushing SSE forward since it came out, each CPU revision improving it. Still not as good as AltiVec, but getting closer. AVX will push it beyond, but won't arrive until 2010.
The alternative is adding separate vector processing cores and using the GPU's processing elements. This isn't the same thing as adding SIMD to the main processors -- different advantages and disadvantages. Really a system should have both.
That is a specific implementation to the general idea that SoC lower power usage for a given performance metric.
Quote:
They are pretty late in the game even for STBs.
My guess for Snow Leopard is not just the CUDO/OpenCL to leverage GPGPU codes, but to optimise multi-core and potentially heterogeneous multicore CPU systems in future. Many would like to see some form of advanced AltiVec which X86 tries to play catch up.
Interesting that you should bring up Alt Vec as that is exactly what i was thinking about after responding to you. See the post above.
I see an Alt Vec implementation, that Apple owns, as having huge potential. Especially if that Alt Vec can be implemented as multiple cores to compete effectively against the GPU accel crowd.
Quote:
In portable systems, advanced power management on the chipset is going to be important and Intel is really behind on this tech.
Behind relative to whom? Sure ARM has a huge advantage here but is Intel that far behind AMD?
Well it is another idea to offer up. WE might all be getting excited about a rumour that has no basis in fact, but it is nice to imagine the evolution of Apple hardware.
Nice to see somebody else thinking along these lines. OpenCL seems like clear evidence that Apple is setting up for precisely this. Crazy powerful vector engines have become fairly common (GPUs, Cell, PhysX, etc.) and will become more common in the future. As you point out, they scale well. The biggest problem is how to program them, and Apple has told us they are working on that... OpenCL to write the programs, and GrandCentral to share the processors between tasks.
Seems like the obvious thing to depress margins across the board and give them technology that their competition can't touch. You'll note in the other thread this was my suggestion but it was summarily ignored (http://forums.appleinsider.com/showt...t=89272&page=5). I stand by my bet.
I see an Alt Vec implementation, that Apple owns, as having huge potential. Especially if that Alt Vec can be implemented as multiple cores to compete effectively against the GPU accel crowd.
Forget "AltiVec". It is a set of instructions and registers attached to the PowerPC. The PowerPC is in Apple's history. Going forward they are free to design vector processors however they want and label them "Velocity Engine", which is why they created their own brand for that in the first place. If you want to have a better guess as to what these kind of specialized processors might look like then the Cell's SPUs are more indicative. The SPUs are still relatively AltiVec-like though, and there are a lot of ways that Apple could choose to go with their vector processing hardware design. I'm looking forward to finding out what kind of cool stuff they've done.
All in all I see this as a real potential. Such an approach would eliminate the weaknesses of a GPU only approach to acceleration and could ideally be maintainable across multiple generations of hardware and multiple ISA of hardware. In the case of cell each vector processor there is rather modest in its power usage though still unacceptable for an iPod. This is where PA Semi and having control of the hardware implementation come into their own. PA knows low power and combining that with an optimised instruction set could yield a very nicely performing Alt Vec like unit that sips power.
Interesting conjecture, Dave.
I mentioned in my thread "touchless" gesture control/typing and was wondering if an Alt Vec would be beneficial for this type of application? Indeed, based on your description (and ultimately my grasp of the technology you mentioned) it would.
Great, just what we need: a more proprietary Apple.
Hooray, let's cheer for giving users LESS choice!
And just how does this make them "more proprietary"? If running on a machine not blessed with wonderous new hardware it just doesn't run as fast. The whole point behind OpenCL (the little we know of it) is to provide an open standard way to program all these various kinds of processors (CPUs, CPUs w/ SIMD, GPUs, and stuff like we're speculating about). Developers write code, code takes advantage of whatever it is you happen to have bought.
With this kind of tech in place I could actually see that Apple might license the OS because now there really is a compelling reason to buy their hardware... its 10x faster!
PS: And do you really want Apple to stop innovating just to avoid being proprietary? I'd rather them push forward to the future, thank you very much.
Intel has been pushing SSE forward since it came out, each CPU revision improving it. Still not as good as AltiVec, but getting closer. AVX will push it beyond, but won't arrive until 2010.
I'd really like to see Apple offer up a complex of Alt Vec cores modelled somewhat along the lines of Cell. The differences would be in the approach to the coupling to the main CPU, access to memory and communications. Done right a successful implementation across all of Apples products is possible.
Quote:
The alternative is adding separate vector processing cores and using the GPU's processing elements. This isn't the same thing as adding SIMD to the main processors -- different advantages and disadvantages. Really a system should have both.
While I don't disagree that a system should have both once it does have a separate vector facility I can see the use of SIMD instructions on the main CPU dropping dramatically. After all if you can dramatically off load the CPU and its buses due to a vector unit handling the high bandwidth, high CPU usage code, why wouldn't you?
The trick of course is that the bandwidth has to go somewhere but this is where having the Alt Vec complex in the chip set might be an advantage. Especially if those Alt Vecs are intelligent enough to handle their own real world interfacing. For example handling the interface to a BlueRay disk player so that the Alt VEC units not only decode the media they handle as much of the bandwidth as possible out of the normal path ways that a system uses.
Of course we are all excited here in relation to this rumour sadly it might be nothing more than Apple adding a new real time clock to the chip set. Sorry for the negativity but there have been to many examples of things getting hyped out of proportion to what actually gets delivered. One has to admit though that Apple admittance that PS is there to design special chip sets (supposedly for Touch/iPhone) does make for a lot of speculation. Even at the iPod level I see a huge advantage for them if they can figure out a way to do video acceleration on a a general purpose vector processor.
Forget "AltiVec". It is a set of instructions and registers attached to the PowerPC.
Well yeah in the implementation that was done on Power PC that is the case. But one could make the statement that Cells SPU's are very ALT-VEC like.
Quote:
The PowerPC is in Apple's history. Going forward they are free to design vector processors however they want and label them "Velocity Engine", which is why they created their own brand for that in the first place.
Very true indeed plus they have the huge advantage of hindsight right now. Thus the vector facility can be designed as multiple cores to support multiple threads or they can take the approach that Cell did. Personally I don't like many of the SPU's design trade offs in Cell so that does not excite me as much as more alternative approaches.
Quote:
If you want to have a better guess as to what these kind of specialized processors might look like then the Cell's SPUs are more indicative. The SPUs are still relatively AltiVec-like though, and there are a lot of ways that Apple could choose to go with their vector processing hardware design.
Certainly they could go with the approach taken with the SPU's but I'd hope that they would address some of the weaknesses there also. Especially with respect to memory accessible by the SPU's, the SPU's independence and access to system resources.
Quote:
I'm looking forward to finding out what kind of cool stuff they've done.
This should be very exciting if it isn't being blown out of proportion already. I'm especially interested in how much performance they can push down to hand held devices. If anybody can place a vector processor into an iPod SoC it would be PA Semi. It ought to result in a very compelling platform. I'm impressed with my 3G iPhone but it doesn't mean I wouldn't want more performance in a hand held. The key in this domain is arriving at a SoC that uses even less power so my battery last longer. I still wonder if a vector processor would hurt more than it helps here. On the desk top though it can't help but to open up the platform to new apps especially if the X in 'X number' of processors ends up being really large.
Comments
If it was really Graphics inside Intel chipset that makes apple unhappy then this should be it.
Everyone seems to think about the P.A Semi buy out. But no P.A Semi is not the expert in Video Processing field, Apple is using other technology that it has licensed, and that is
PowerVR!
Long time ago PowerVR issued a press release stating "someone" has acquired the sole right for using its SGX and VX technology in specific area.
Now i know this is vague and it seems in many instance and clues we know it was Apple. But we have yet to see product using it. Many thought iPhone chipset was it but it turns out they are using older MBX core from PowerVR.
So may be the deal was for apple to use SGX and VX on Desktop and Laptop Computers?
For those who dont know, SGX have some have decent performance for its size and power usage. And While SGX may not be the best GPU in its class, VX has score the lowest power usage for decoding 1080p H.264 Video. It is one of the best if not the best video decoder on the market.
Apple could use ARM core, SGX and VX Core inside its Mac Chipset.
This way we would have common platform top to bottom. And Quicktime X to support hardware acceleration of H.264 video on all apple products, Mac, iPod and iPhone.
Last of all, Apple is not developing anything. I'm pretty sure with all the talk lately about system-on-a-chip (SoC) from Intel and the press, that Apple is using a new chip from Intel that is a SoC. That means the CPU, GPU, southbridge, northbridge are all on one chip. This makes sense for laptop redesigns since you only need one main chip for the motherboard, therefore changing size and thermal considerations and simplifying the overall layout. This chip is probably very customizable to fit new notebook designs and only available for Apple from Intel.
An SoC from Intel makes the most sense to me.
You were doing well until the last paragraph. SoC chips are good for things like iPhone but not your average notebook/laptop. Atom 1.6Ghz anyone ?. Please... that is turning the clock back another 4 years or so.
SoC will not likely get MBP type power/performance anytime soon, if ever.
PA Semi doing a specialised SoC makes a lot of sense. The MIPS based Samsung chip in the iPhone would be a nice one to go. PA Semi also has experience and talent to do system Chipsets.
Look at what ServerWorks chipset did for Xeon...
I figured CUDA or OpenCL type technologies coming with Snow Leopard has a lot to do with chipset redesign. A great chipset will make the cores sing like what Serverworks did for X86 servers those years ago. Plausible ?. I think so.
PS: SoC from Intel still sucks in perf/watt.
doesn't make sense.
they already have to write code in universal binary to be used on intel and powerpc chips... are they going to now need to write code in universal tertiary?
I'm sure the changes will just be upgraded motherboards of some sort, but after juse getting intel processor on board in the last few years, it makes absolutely zero sense to abandon them this quickly for a new type processor.
I think major overhaul is unlikely, more like Apple wants to have more control over design. I don't see OS compatibility being an issue Apple hasn't thought about, more likely the chips will deliver more features that you just don't find anywhere else & once you have them you will wonder how you lived without.
Might be nice to have some chips more properly designed to take advantage of all aspects of EFI. Such a thing would actually make VMWare & Parallels more happy as it might boost their performance.
Don't forget that EFI is the key in compatibility. Windows isn't compatible with EFI already, but EFI is able to emulate BIOS to Windows. EFI makes a lot more possible than people realize.
I think it would slow down apples gain in marketshare.
You were doing well until the last paragraph. SoC chips are good for things like iPhone but not your average notebook/laptop. Atom 1.6Ghz anyone ?. Please... that is turning the clock back another 4 years or so.
SoC will not likely get MBP type power/performance anytime soon, if ever.
The guy you responded to did accurately describe what a SoC is. S0C do not imply any one performance point, in fact a SoC can often be had that is higher performance than the equivalent assemblage of discrete parts. One huge factor in SoC performance is the reduction in power usage due to the reduction in the number of off chip drivers used in an implementation.
With respect to mainstream computing only recently has the amount of die space exceed what is required to make the CPU itself. This means that it is possible to go the SoC route for high performance computing. Possible but this doesn't mean that Intel has publicly announced such a chip. In any event I do not believe that any of Intels announced or un-announced S0C have anything at all to do with this rumour.
If the new Macs are to have something special then by definition it can't be publicly available Intel stock.
PA Semi doing a specialised SoC makes a lot of sense. The MIPS based Samsung chip in the iPhone would be a nice one to go. PA Semi also has experience and talent to do system Chipsets.
Look at what ServerWorks chipset did for Xeon...
Again MIPS has nothing to do with iPhone. That is ARM based technology.
As to PA Semi and its ability to do a chip set ala ServerWorks that is a possibility. The problem is there is no advantage for Apple in a straight up chip set implementation. PA Semi would have to be adding something to the mix that would not be expected in a system chip set. So we come back to just what is that something special given that it would need to be more than just a run of the mill chip set.
I figured CUDA or OpenCL type technologies coming with Snow Leopard has a lot to do with chipset redesign. A great chipset will make the cores sing like what Serverworks did for X86 servers those years ago. Plausible ?. I think so.
Not unless this supposedly new chip set has the facility to actually accelerate the types of codes that OpenCL is thought to support. It still comes down to, will the chip set have some sort of built in acceleration and what exactly will that acceleration be. In could be general video decoding hardware to something like a new run at a vector engine.
PS: SoC from Intel still sucks in perf/watt.
Are you sure you are up to speed on Intels latest offerings? It is pretty hard for a S0C not to lead in a performance per watt metric.
Dave
Custom Apple designed chipsets for product classes that do not exist yet. Come on people; use your brains.
I fully agree with your statement.
Indeed, the purchase of P.A. Semi seems to indicate potential products in areas Apple does not currently address.
For example, Microsoft seems to be making great strides in automotive media control/entertainment systems (e.g. Ford's 'InSync' and in other markets known as 'Blue and Me'). I see no reason why this type of application shouldn't be part of Apple's evolutionary development.
Based on what technology is already employed in their existing products (e.g. GPS in the iPhone 3G), the likes of Apple, P.A. Semi and a partner like Bose, should be able to bring to market sophisticated, yet elegant, and desirable "appliances" for automobiles, aircraft and trains in a very short amount of time.
And speaking of Apple appliances, I envision P.A. Semi has the potential to help Apple in the future differentiate the iPod, iPod Touch, iPhone and the much rumored tablet PDA class devices.
I agree with most others that Apple's desktop and laptop systems will continue to use Intel CPUs and chipsets for the near future, but may employ P.A. Semi or other third-party technologies to enhance differentiation from its competitors.
For example, Apple has applied for patents on multi-gesture controls. Then witness HP's "all-in-one" TouchSmart touch-enabled PC. Yawn...
Personally, fingerprints on screens is a "pet peeve" of mine, so I think Apple is going to go one better where you do gestures in front of a more sophisticated iSight camera and "touch" without physically touching the screen. Better yet, "type" without touching a physical keyboard. It would all be done by tracking the hands and eyes.
You may disagree with me on the above, but I truly believe it is not a matter of if, it's when. And as we all know, Apple is the kind of company to bring this type of technology to the market first and (mostly) succeed at it.
-YipYipYipee
Are you sure you are up to speed on Intels latest offerings? It is pretty hard for a S0C not to lead in a performance per watt metric.
Dave
Not quite. But this one ?.
http://www.dailytech.com/New+Intel+E...ticle12484.htm
They are pretty late in the game even for STBs.
My guess for Snow Leopard is not just the CUDO/OpenCL to leverage GPGPU codes, but to optimise multi-core and potentially heterogeneous multicore CPU systems in future. Many would like to see some form of advanced AltiVec which X86 tries to play catch up.
In portable systems, advanced power management on the chipset is going to be important and Intel is really behind on this tech.
I think that if apple goes back to powerpc cpu's (thats what this pa semi company made)then they will alienate a lot of developers who had just finished the transition to intel cpu's. I think if they did that the developers would walk away from apple and say its just not worthit to have to learn everything again.
Apple isn't going back to PPC and nobody is suggesting that they will. PA Semi licenses core designs they need, revises them to incorporate their advanced power saving design techniques, and then integrates them into a system-on-chip. Apple has likely directed them to work on ARM-based designs for iPod/iPhone, and (as speculated in this thread) may also be using them to design northbridge/southbridge chips for their Intel-processor-based Macs.
Since the 80s Apple was doing its own chipset and motherboard design which included memory controller, I/O interfaces, DSPs, and processor interconnects. The G5's chipset was quite an impressive piece of tech. In the last few years Apple has acquired Racer and now PA Semi, among others. They will leverage Intel's processors, but they are going back into the motherboard design business... which these days means the big northbridge/southbridge chip (or chips) since they are so heavily integrated. We'll have to wait and see whether this will include a GPU. I find it unlikely since laptop/desktop class GPUs are still pretty sophisticated and competing with ATI/nVidia will be challenging... but who knows what they've licensed for inclusion on their SoC. I still remember rumours from a couple years ago of vector processing cores (like DSPs, the Cell processor's SPEs, or the processing elements of modern GPUs) coming to Mac chipsets, but they never materialized. Still, something like Grand Central and OpenCL is precisely what Apple would need to put into place to allow 3rd party developers to leverage this kind of technology directly. This kind of tech would obviate the need for specialized sound cards or H.264 decoder chips, and it would be far more flexible.
If this is true, it is interesting timing. Intel's Penyrn chips (like all predecessors) have a northbridge that includes a memory controller. The next chips, coming starting this fall, will for the first time (for Intel) include a memory controller on the processor die and will switch to Intel's Quick Interconnect technology. This is a pretty major change in the northbridge and thus would mean a significant piece of work on Apple's part for their chipset which lasts only a single generation of machines. Perhaps they licensed the bus interface rather than designing it themselves, but this seems like unfortunate timing. Why not just wait until Nehalem has arrived and design around the new interconnect which is (hopefully) going to be standard going forward, much like HyperTransport on AMD. Speaking of which, Apple used HyperTransport and was on the design board... I guess they'll eschew that in favour of Intel's new equivalent tech.
Think of it this way Apple develops a chip set with a Vector Engine facility that it owns out right. A vector engine that is effectively decoupled from what ever CPU they are running. Think of it as an Alt Vec (AV2)for the new century.
This would give Apple some serious advantages that we can highlight below.
1.
One the AV2 facility would be a programmable unit suitable for vector computations. Hopefully as such it would handle both long and short vectors and be fully 64 bit. This would give Apple a common platform for application acceleration no matter what platform, be it AMD, Intel, PPC, ARM or whatever. All Apple would need is its own Vector Engine with interchangeable front side buses.
2.
Lets face it Intels vector implementation sucks and really hasn't caught up with Alt Vec of old. Even if current implementations suck less there is still the issue of running vector code through your main CPU complex. Vector code just sucks up a lot of bandwidth and CPU resources leading to laggy systems. Further even with all of Intels latest improvements decoding all that Blue-Ray can offer up is not possible. So such hardware may very well mean the implementation of BlueRay on Apple hardware.
3.
Owning your own Vector Engine means owing the instruction set and the implementation of those instructions. This means hardware acceleration of certain codes could be nothing more than a single instruction in the vector engine. Things like dot products, security codes and the like come to mind. While Apple might call the unit a vector engine in reality it is a location where anything not handled well by the main CPU can be accelerated. Since Apple owns the hardware if it wants to implement an MD5 instruction or some sort of convolution instruction it can. Of course there is always the issue of trade offs but this being a specialized computational unit Apple would be free to accelerate what they need accelerated.
4.
One can only imagine what an Alt Vec like engine could do on modern processes and with a whole different set of design parameters. They would have at their disposal much more chip real estate, far lower power usage, and on chip connectivity to supporting units. It would not be impossible to imagine an implementation that results in a multi threaded hardware platform. Lets say they are able to implement 4 or 8 Alt Vec equivalent cores in such an implementation would that excite people? How about 64? Since each would be its own computational unit the number of cores could be adjusted to suit the hardware. Thus an iPod might get one core while a Mac Pro might get 32.
5.
Having an optimised Alt Vec core available, that can do decent video acceleration, means that Apple can program to one acceleration framework across all devices. When Apple goes to a full SoC for the iPods (they will) it means that they control the video accel hardware (as a general purpose vector engine) thus reducing licensing costs and limiting the diversity of components. Since the unit is a vector engine that means not only do video apps have the potential for acceleration but so do any other apps that can map code onto vector instructions.
6.
While the acceleration of certain codes by moving the code to a GPU works and sometimes amazingly well it is not without problems. One issue being the constantly changing nature of GPU's. A generalized vector facility can make up for many of a GPU's short comings when working effectively with the CPU. Especially if that vector unit is now multi threaded and can more easily handle the things that GPU's do well. I realize that modern GPU have a lot of cores running in parallel but I see something as modest as an 8 core AV2 as offering up a huge number of advantages that would make such preferable for many developers. In a way this would be something like Sony's cell implementation. The main difference is that it would reside in Apple controlled hardware decoupled from whatever is the main CPU.
All in all I see this as a real potential. Such an approach would eliminate the weaknesses of a GPU only approach to acceleration and could ideally be maintainable across multiple generations of hardware and multiple ISA of hardware. In the case of cell each vecotr processor there is rather modest in its power usage though still unacceptable for an iPod. This is where PA Semi and having control of the hardware implementation come into their own. PA knows low power and combining that with an optimised instruction set could yield a very nicely performing Alt Vec like unit that sips power.
Well it is another idea to offer up. WE might all be getting excited about a rumour that has no basis in fact, but it is nice to imagine the evolution of Apple hardware.
Dave
Many would like to see some form of advanced AltiVec which X86 tries to play catch up.
http://softwareprojects.intel.com/avx/
Intel has been pushing SSE forward since it came out, each CPU revision improving it. Still not as good as AltiVec, but getting closer. AVX will push it beyond, but won't arrive until 2010.
The alternative is adding separate vector processing cores and using the GPU's processing elements. This isn't the same thing as adding SIMD to the main processors -- different advantages and disadvantages. Really a system should have both.
Not quite. But this one ?.
http://www.dailytech.com/New+Intel+E...ticle12484.htm
That is a specific implementation to the general idea that SoC lower power usage for a given performance metric.
They are pretty late in the game even for STBs.
My guess for Snow Leopard is not just the CUDO/OpenCL to leverage GPGPU codes, but to optimise multi-core and potentially heterogeneous multicore CPU systems in future. Many would like to see some form of advanced AltiVec which X86 tries to play catch up.
Interesting that you should bring up Alt Vec as that is exactly what i was thinking about after responding to you. See the post above.
I see an Alt Vec implementation, that Apple owns, as having huge potential. Especially if that Alt Vec can be implemented as multiple cores to compete effectively against the GPU accel crowd.
In portable systems, advanced power management on the chipset is going to be important and Intel is really behind on this tech.
Behind relative to whom? Sure ARM has a huge advantage here but is Intel that far behind AMD?
Dave
Well it is another idea to offer up. WE might all be getting excited about a rumour that has no basis in fact, but it is nice to imagine the evolution of Apple hardware.
Nice to see somebody else thinking along these lines. OpenCL seems like clear evidence that Apple is setting up for precisely this. Crazy powerful vector engines have become fairly common (GPUs, Cell, PhysX, etc.) and will become more common in the future. As you point out, they scale well. The biggest problem is how to program them, and Apple has told us they are working on that... OpenCL to write the programs, and GrandCentral to share the processors between tasks.
Seems like the obvious thing to depress margins across the board and give them technology that their competition can't touch. You'll note in the other thread this was my suggestion but it was summarily ignored (http://forums.appleinsider.com/showt...t=89272&page=5). I stand by my bet.
I see an Alt Vec implementation, that Apple owns, as having huge potential. Especially if that Alt Vec can be implemented as multiple cores to compete effectively against the GPU accel crowd.
Forget "AltiVec". It is a set of instructions and registers attached to the PowerPC. The PowerPC is in Apple's history. Going forward they are free to design vector processors however they want and label them "Velocity Engine", which is why they created their own brand for that in the first place. If you want to have a better guess as to what these kind of specialized processors might look like then the Cell's SPUs are more indicative. The SPUs are still relatively AltiVec-like though, and there are a lot of ways that Apple could choose to go with their vector processing hardware design. I'm looking forward to finding out what kind of cool stuff they've done.
Hooray, let's cheer for giving users LESS choice!
-Clive
All in all I see this as a real potential. Such an approach would eliminate the weaknesses of a GPU only approach to acceleration and could ideally be maintainable across multiple generations of hardware and multiple ISA of hardware. In the case of cell each vector processor there is rather modest in its power usage though still unacceptable for an iPod. This is where PA Semi and having control of the hardware implementation come into their own. PA knows low power and combining that with an optimised instruction set could yield a very nicely performing Alt Vec like unit that sips power.
Interesting conjecture, Dave.
I mentioned in my thread "touchless" gesture control/typing and was wondering if an Alt Vec would be beneficial for this type of application? Indeed, based on your description (and ultimately my grasp of the technology you mentioned) it would.
-YipYipYipee
Great, just what we need: a more proprietary Apple.
Hooray, let's cheer for giving users LESS choice!
And just how does this make them "more proprietary"? If running on a machine not blessed with wonderous new hardware it just doesn't run as fast. The whole point behind OpenCL (the little we know of it) is to provide an open standard way to program all these various kinds of processors (CPUs, CPUs w/ SIMD, GPUs, and stuff like we're speculating about). Developers write code, code takes advantage of whatever it is you happen to have bought.
With this kind of tech in place I could actually see that Apple might license the OS because now there really is a compelling reason to buy their hardware... its 10x faster!
PS: And do you really want Apple to stop innovating just to avoid being proprietary? I'd rather them push forward to the future, thank you very much.
IMO.
http://softwareprojects.intel.com/avx/
Intel has been pushing SSE forward since it came out, each CPU revision improving it. Still not as good as AltiVec, but getting closer. AVX will push it beyond, but won't arrive until 2010.
I'd really like to see Apple offer up a complex of Alt Vec cores modelled somewhat along the lines of Cell. The differences would be in the approach to the coupling to the main CPU, access to memory and communications. Done right a successful implementation across all of Apples products is possible.
The alternative is adding separate vector processing cores and using the GPU's processing elements. This isn't the same thing as adding SIMD to the main processors -- different advantages and disadvantages. Really a system should have both.
While I don't disagree that a system should have both once it does have a separate vector facility I can see the use of SIMD instructions on the main CPU dropping dramatically. After all if you can dramatically off load the CPU and its buses due to a vector unit handling the high bandwidth, high CPU usage code, why wouldn't you?
The trick of course is that the bandwidth has to go somewhere but this is where having the Alt Vec complex in the chip set might be an advantage. Especially if those Alt Vecs are intelligent enough to handle their own real world interfacing. For example handling the interface to a BlueRay disk player so that the Alt VEC units not only decode the media they handle as much of the bandwidth as possible out of the normal path ways that a system uses.
Of course we are all excited here in relation to this rumour sadly it might be nothing more than Apple adding a new real time clock to the chip set. Sorry for the negativity but there have been to many examples of things getting hyped out of proportion to what actually gets delivered. One has to admit though that Apple admittance that PS is there to design special chip sets (supposedly for Touch/iPhone) does make for a lot of speculation. Even at the iPod level I see a huge advantage for them if they can figure out a way to do video acceleration on a a general purpose vector processor.
Dave
Forget "AltiVec". It is a set of instructions and registers attached to the PowerPC.
Well yeah in the implementation that was done on Power PC that is the case. But one could make the statement that Cells SPU's are very ALT-VEC like.
The PowerPC is in Apple's history. Going forward they are free to design vector processors however they want and label them "Velocity Engine", which is why they created their own brand for that in the first place.
Very true indeed plus they have the huge advantage of hindsight right now. Thus the vector facility can be designed as multiple cores to support multiple threads or they can take the approach that Cell did. Personally I don't like many of the SPU's design trade offs in Cell so that does not excite me as much as more alternative approaches.
If you want to have a better guess as to what these kind of specialized processors might look like then the Cell's SPUs are more indicative. The SPUs are still relatively AltiVec-like though, and there are a lot of ways that Apple could choose to go with their vector processing hardware design.
Certainly they could go with the approach taken with the SPU's but I'd hope that they would address some of the weaknesses there also. Especially with respect to memory accessible by the SPU's, the SPU's independence and access to system resources.
I'm looking forward to finding out what kind of cool stuff they've done.
This should be very exciting if it isn't being blown out of proportion already. I'm especially interested in how much performance they can push down to hand held devices. If anybody can place a vector processor into an iPod SoC it would be PA Semi. It ought to result in a very compelling platform. I'm impressed with my 3G iPhone but it doesn't mean I wouldn't want more performance in a hand held. The key in this domain is arriving at a SoC that uses even less power so my battery last longer. I still wonder if a vector processor would hurt more than it helps here. On the desk top though it can't help but to open up the platform to new apps especially if the X in 'X number' of processors ends up being really large.
DAVE