Intel to sell remaining stake in Apple partner & mobile GPU maker Imagination Technologies

2»

Comments

  • Reply 21 of 31
    MarvinMarvin Posts: 15,309moderator
    macarena wrote: »
    Apple has taken a totally different route and is now so far ahead of Intel, it will crush Intel.

    Apple isn't in competition with Intel. They originally wanted to use Intel for iOS devices but it was Tony Fadell who convinced them otherwise:

    http://blogs.wsj.com/digits/2011/10/26/steve-jobs-wanted-intel-chips-for-the-ipad/

    It was the right decision to go with ARM for iOS as they control the launch times, the chip features, the power draw without giving any of the advantages away to their competition. The PC market has run its course and breaking legacy for speed improvements would do more harm than good.

    If they could maintain some binary compatibility and the performance-per-watt and performance-per-dollar improvements were huge then it would be worth trying but still, it wouldn't be out of any desire to beat Intel.

    If they introduced low-end laptop and iMac models with ARM chips then they'd have a better chance at taking on their PC competitors: HP, Google Chromebooks, Lenovo, Dell etc. They could leave the higher-end models on Intel. If the entry models maintain x86 binary compatibility at the hardware level then there would be almost no difference between them and normal Intel laptops, just the ability to run Windows and possibly some games.
  • Reply 22 of 31
    solipsismy wrote: »
    It's interesting that some of the greatest advancements in CPUs and GPUs are coming from the UK. It's kind of like with music, relatively few people but so much talent.

    Yep.

    There's a reason why it's called Great Britain.
  • Reply 23 of 31
    srams123 wrote: »
    Apple will announce shortly that they are buying out Imagination Tech ... This is a shot in the arm they need to reduce / eliminate reliance on Nvidia !!!

    I'm surprised they haven't already.
  • Reply 24 of 31
    Marvin wrote: »
    Apple isn't in competition with Intel. They originally wanted to use Intel for iOS devices but it was Tony Fadell who convinced them otherwise:

    http://blogs.wsj.com/digits/2011/10/26/steve-jobs-wanted-intel-chips-for-the-ipad/

    It was the right decision to go with ARM for iOS as they control the launch times, the chip features, the power draw without giving any of the advantages away to their competition. The PC market has run its course and breaking legacy for speed improvements would do more harm than good.

    If they could maintain some binary compatibility and the performance-per-watt and performance-per-dollar improvements were huge then it would be worth trying but still, it wouldn't be out of any desire to beat Intel.

    If they introduced low-end laptop and iMac models with ARM chips then they'd have a better chance at taking on their PC competitors: HP, Google Chromebooks, Lenovo, Dell etc. They could leave the higher-end models on Intel. If the entry models maintain x86 binary compatibility at the hardware level then there would be almost no difference between them and normal Intel laptops, just the ability to run Windows and possibly some games.

    Marvin - let me ask you a few questions - then you will realize the direction Apple is taking.

    - have you ever wondered why Apple gave up on XServe? Despite being a small part of sales, it was an important part, because that was the best option for a Mac Based server. I don't see the Mac Mini as a real server - it is just a basic alternative to one.

    At that point, people expected Apple to replace this line with something else - and it didn't happen. Today, Apple is forcing customers to pick Linux servers for even Apple tasks. Why?

    Do you really think binary compatibility is such a big deal? When Apple has perfect Fat Binaries across all its development tools, that can contain all platform versions - and then each platform can throw away the ones it doesn't need?

    Do you really believe Apple will attempt to cripple itself with attempting binary compatibility between a CISC and a RISC based architecture?

    How difficult is it for the Mac App Store to find out the architecture of the requesting machine, and show software that is compatible?

    How difficult is it for Apple to simply enforce a rule - that all software on Mac App Store must be recompiled and delivered for all platforms by some date? They have done this for 64 bit on iOS. They have done this several times. As long as Apple does the hard work and ensures that it is simple recompile, what's the problem?

    How difficult is it to come up with a Server Store - that offers Servers and utilities that can be easily plugged in. Probably even with a seamless management interface?

    Apple has been taking many steps in this direction - just that you haven't seen the real background for these steps yet!
  • Reply 25 of 31
    MarvinMarvin Posts: 15,309moderator
    macarena wrote: »
    have you ever wondered why Apple gave up on XServe?

    Not really:

    http://appleinsider.com/articles/10/11/08/alleged_steve_jobs_e_mail_says_hardly_anyone_was_buying_apples_xserves
    macarena wrote: »
    Despite being a small part of sales, it was an important part. Today, Apple is forcing customers to pick Linux servers for even Apple tasks. Why?

    Why not? They do the job ok. It's just not an area where Apple can add enough value to persuade people to buy the hardware.
    macarena wrote: »
    Do you really think binary compatibility is such a big deal?

    I think it's important. The transition to Intel would have been awful without it. I think it would be easier in future but there's also more software available that needs to be updated.
    macarena wrote: »
    Do you really believe Apple will attempt to cripple itself with attempting binary compatibility between a CISC and a RISC based architecture?

    They managed Rosetta ok.
    macarena wrote: »
    As long as Apple does the hard work and ensures that it is simple recompile, what's the problem?

    They can do both. As we've seen before, not everyone deploys on the App Store so they can't enforce the recompilation and some software doesn't get recompiled soon enough.

    None of your post answers why Apple would have a motivation to beat Intel. They are partners. They have co-developed good technology, there's no reason to abandon that partnership.

    If they can hit a higher PC volume with ARM on the low-end, that's an option and it wouldn't cannibalize their x86 line entirely.
  • Reply 26 of 31
    Rosetta was probably the worst transition methodology used by Apple - it was slow, and buggy. I hope they never ever try to use an emulator! Is that what you mean by binary compatibility? Thank God Apple has better ideas - like what they used for the Intel Transition.

    The Mac transition to Intel is NOT about binary compatibility. They used Universal binaries for that.

    With time a lot of software is getting on to Mac App Store. While there will always be software that's not on the store, I think that will be ignorable - there will be enough on the App Store to meet the needs of most users.

    XServe sold less - but there was still enough demand that people actually went out and bought XServes before they were phased out! Not the typical response to a product that was discontinued! Try running a XCode Continuous Integration Server for a large project on a Mac Mini and you will see why XServes are sorely missed. You really don't want a Mac Pro with its expensive Graphics hardware for most compute only tasks!

    Apple isn't on a mission to save Intel - and where did you get the idea that ARM will be about cheaper Macs? ARM will be about faster, thinner, lighter Macs! Not cheaper. And yes - reread that - I did say faster.

    The problem with your thinking is that you are skating to where the puck was - Apple moved the puck - so the puck is now where Apple is! Just wait till you see a version of OS X that's compiled with Metal. Just wait till GPU based acceleration is a reality for anything and everything. With so low overheads that you dont even notice it!
  • Reply 27 of 31
    MarvinMarvin Posts: 15,309moderator
    macarena wrote: »
    Rosetta was probably the worst transition methodology used by Apple - it was slow, and buggy. I hope they never ever try to use an emulator! Is that what you mean by binary compatibility? Thank God Apple has better ideas - like what they used for the Intel Transition.

    The Mac transition to Intel is NOT about binary compatibility. They used Universal binaries for that.

    Rosetta was binary translation between PPC and Intel:

    http://en.wikipedia.org/wiki/Rosetta_(software)

    The slow, buggy thing was the Classic Environment, which was used to ease the transition between OS 9 and OS X:

    http://en.wikipedia.org/wiki/Classic_Environment
    macarena wrote: »
    XServes are sorely missed. You really don't want a Mac Pro with its expensive Graphics hardware for most compute only tasks!

    The Mac Pro GPUs don't add much cost to the Mac Pro. It's $400 to upgrade both to the D500. These machines just have very high gross margins. XServes may well be missed but if those people who miss them were large in number and had bought them regularly then Apple wouldn't have discontinued them.
    macarena wrote: »
    Apple isn't on a mission to save Intel - and where did you get the idea that ARM will be about cheaper Macs? ARM will be about faster, thinner, lighter Macs! Not cheaper. And yes - reread that - I did say faster.

    The reason that Microsoft was able to price Windows RT devices close to iOS was ARM chips. The Intel models cost much more. I expect they'd be faster in the same power profile too though, I don't think it would be enough faster to warrant the software upheaval.
    macarena wrote: »
    The problem with your thinking is that you are skating to where the puck was

    Says the guy who wants the XServe back.
    macarena wrote: »
    Just wait till you see a version of OS X that's compiled with Metal. Just wait till GPU based acceleration is a reality for anything and everything. With so low overheads that you dont even notice it!

    Metal is about draw calls and it's a programming API. This can be done on Intel, NVidia, AMD GPUs but the reason it isn't is because software has to cater for far more GPUs, which is why there is an abstraction layer (OpenGL). iOS devices have the luxury of using very few GPUs that are completely controlled by Apple and shipping in crazy high numbers. They ship 10x more iOS devices than Macs. Metal helps in games, it would do very little for OS X.
  • Reply 28 of 31
    I know what Rosetta is - and that's what I mean by binary compatibility - ensuring that things run fine at the binary level. If you mean that ARM should somehow support the Intel Instruction Set without a translation layer, I don't think that's possible.

    Rosetta was needed for the transition - but that doesn't make it great. Secondly, Rosetta was just a simple translation layer that allowed the old binary format of the Classic to run on from the new Mac OS X. The Chips and the actual instructions itself were same. Think of it as something like the WoW solution MS came up with to support old Windows apps in Windows NT. Between ARM and Intel the instruction set itself is different. Translation at that level is a tougher ball game and would be a much bigger impact on performance.

    As for XServe - I like XServe but I don't want it back. At least not in its old form. I believe Apple has a huge market for servers that are extremely low power consumption, which can run software that is largely the same as what Linux runs, with just a recompile. Think of an Apple TV as a server - but with much more RAM and more flash. That's what I mean.

    In that space, there isn't any existing market for Apple to worry about compatibility, etc.

    As for Metal, there's two things - it is eliminating abstraction, plus even more importantly, Metal works so well because Apple has integrated the GPU into the SoC itself. And the GPU uses exact same memory as what the CPU uses.

    This is kind of similar to Intel's HD 5000 and all the Integrated Graphics chips - except that Intel has always been crap at Graphics. If Intel was good at Graphics, the world would probably not need External Graphics cards!

    A-series and Metal is a unique model. They have taken the best of class mobile GPU, integrated in the SoC, and then eliminated all overheads and abstraction. And now that best of class GPU has become completely transparent and available for all code.

    The level of efficiency that Apple has achieved with Metal is simply not possible for Intel to achieve - unless Intel dramatically ups it's GPU game. For all the progress Intel has made on Atom as a credible mobile platform, it still sucks for graphics.

    But there is a more critical dynamic. Metal doesn't make much sense if it is used only for iOS devices. The performance of A8 class processors, as well as their ability to handle graphics intensive games, is already so good, it doesn't warrant the effort to come up with a separate API! But if you think of using these chips as regular Desktop chips - replacing Intel, then Metal plays a crucial role. A8 is considered to be at a level where top of line Intel chips were in 2011. A9 with Metal should be faster than most of Intel's current chips.
  • Reply 29 of 31
    Quote:
    Originally Posted by macarena View Post





    This is kind of similar to Intel's HD 5000 and all the Integrated Graphics chips - except that Intel has always been crap at Graphics. If Intel was good at Graphics, the world would probably not need External Graphics cards!



    A-series and Metal is a unique model. They have taken the best of class mobile GPU, integrated in the SoC, and then eliminated all overheads and abstraction. And now that best of class GPU has become completely transparent and available for all code.


    Of course they integrated it into the chip, that is the definition of a SoC. It is like ATM machine. Actually, having a low level API make the GPU LESS transparent and more directly accessible?




    The level of efficiency that Apple has achieved with Metal is simply not possible for Intel to achieve - unless Intel dramatically ups it's GPU game. For all the progress Intel has made on Atom as a credible mobile platform, it still sucks for graphics.


    Why is it not possible for intel to do it? AMD's Mantle and MS's DirectX 12 aim to do the same thing as Metal. Why would Apple not expose the Intel or Nvidia GPU used in a Mac the same way they do the PowerVR GPU in iOS?



    But there is a more critical dynamic. Metal doesn't make much sense if it is used only for iOS devices. The performance of A8 class processors, as well as their ability to handle graphics intensive games, is already so good, it doesn't warrant the effort to come up with a separate API! But if you think of using these chips as regular Desktop chips - replacing Intel, then Metal plays a crucial role. A8 is considered to be at a level where top of line Intel chips were in 2011. A9 with Metal should be faster than most of Intel's current chips.


    Maybe if you only count Macbook Air chips then you might have a point, but as it stands the iPad air 2 is still a bit slower than the first Intel Mac Pro. You also forget that Metal only helps the GPU, it does nothing for the CPU. Will an A9 be good enough for a Macbook Air? Probably, but it won't replace a high end machine anytime soon. Not without scaling power consumption. Remember, it is also easier to cool a seperate massive CPU and GPU than a giant chip.

  • Reply 30 of 31
    Why is it not possible for Intel to achieve what Apple did? For that you need to delve a little bit deeper into the technicals of the GPU and CPU and understand the difference between what Intel means by Shared Graphics Memory and what Apple means by Shared Graphics Memory (in iOS).

    When you have physical memory chips, say of 4GB, Intel partitions that chip and uses a small part of it, say 256MB as Graphics memory. That memory is not available to the CPU, and the rest of the memory is not available to the GPU. When you want to create any GPGPU code, you have to copy memory from CPU area to GPU area, do the calculations, and then copy the result memory back. These steps involve significant costs in terms of machine cycles.

    In iOS, Apple uses a common address space for both CPU and GPU. With just 1GB RAM, this was the only option - but more importantly, since Apple has full control over all the components - hardware, OS, this was possible to do.

    If you look at Metal, there is no need to do this copying of buffers back and forth. That removes a lot of the overhead in making the call.

    The A-series architecture is the only architecture where this is possible. No one else is using such an architecture.

    Pls understand one thing - when we say there is a lot of Driver overhead in OpenGL - it is not a wasted driver overhead. There are lot of additional things that need to be done, so the GPU Can be used as GPGPU. You cannot wish away those things.

    Except if you are Apple!

    When you say Metal only helps GPU, not CPU, you don't understand Metal. The whole point of eliminating all overheads, is that Metal can now be used even for very small tasks - and the parallel processing that is possible with additional cores in the GPU will make even compute tasks much faster. With OpenGL, the overhead was high - so unless you have intensive long running tasks, it didn't make sense to use the GPU.

    I expect A9 with a full implementation of Metal based OS and apps to be way faster than 80-90% of Intel's chip line-up. Of course, the serious work horses - like the quad core i7, or the 6/8/12 core monsters are still way faster. But with no limits on power usage, Apple should be able to put in a few more cores, and get into a zone where the difference doesn't matter. Apple should be able to put in a lot of functionality as hardware implemented, so that things become even more smoother.

    PS: there are also commercial reasons why Intel cannot do something like what Apple did with Metal - if Intel does this, then even cheap processors will be more than enough for any one! No one will have a need to buy the monster work horses. Already Intel is feeling a lot of pain because customers think their old computers are "good enough" and not upgrading! You think they want to shoot themselves in the foot by offering something like this? Apple is still just a small business for Intel - if they offer something like this, they will get killed on their much larger Windows sales! Losing a lot of Apple's business is probably a smaller problem for Intel!

    iOS 9 will be the second OS upgrade when the older iPhone will get faster! This happened with iOS 5 as well (but in gradual steps from 4.2) when OS and apps were recompiled to take advantage of ARC.
  • Reply 31 of 31
    There is another elephant in the room regarding the A series processors. And it is a giant one more the size of a mastodon. It is price. The ARM processors are like an order of magnitude lower in cost.

    An A series machine would allow Apple to lower prices and increase their margins all at the same time.

    Apple's ARM processors offer so many advantages over x86, it is inconceivable that they will stay married to Intel for long.

    The only thing x86 offers is backwards compatibility but it really isn't much of a factor for Apple. It is much different for Microsoft, however. But 80% of all software development is now being performed on the ARM platform and Apple is a huge part of that market. That percentage is only going to increase going forward. I use my iPhone for the vast majority of things I do. I have also been able to leave the work Dell Inspiron, now using an iPad. I used to see a large number of laptops in places like Starbucks. They have virtually disappeared. The vast majority of folks are on a smartphone. Backwards compatibility is not as important as it used to be and as things continue to progress it loses more of its importance. Even Microsoft themselves see this and released an ARM based tablet.

    ARM performance is increasing rapidly and will eventually overtake x86. The ARM marketplace dwarfs x86 and the competition is very intense. Samsung has virtually caught up with Intel's manufacturing process and it's really only a matter of time before they overtake Intel.

    So perhaps the moderator can explain to me why Apple needs to stay with Intel for any length of time? What does Intel really have to offer going forward? Excellent performance for now but at significant cost, both in price and power. They can't design competitive GPUs and the vast majority of software development is now on ARM. Even if ARM requires the same power consumption as x86 to achieve performance parity, the processors are one tenth the cost. Why wouldn't Apple make the switch? Especially when Apple would be no longer tied to Intel's development schedule and are able to produce unique SOCs that would be the first to market?

    Intel's in trouble, no amount of wishful thinking is going to change that. Unless they can figure out a way to dramatically lower the cost of their x86 products and still make a profit.

    I have no desire to purchase the new 1299 MacBook. However, I am anxiously waiting for an iPad pro with the A9. The cost difference is a deal breaker for me with the Intel based laptop.

    Intel should have used their stake in Imagination to obtain additional expertise in making a decent GPU. It doesn't really matter any longer, however, as they won't be able to compete for much longer in any case.

    Interesting how Apple produced a proprietary product and Microsoft licensed their OS to anyone. Intel now has the proprietary product and ARM licenses their technology to anyone. Intel needs to find a new product for a new market that no one is currently involved in or a market producing shoddy goods. They can get back to relevance much like Apple did with their mobile platforms. Or Intel can succumb to ARM like Apple almost did to Microsoft.
Sign In or Register to comment.