By: Doug S (foo.delete@this.bar.bar), October 26, 2016 8:06 am
Room: Moderated Discussions
Andreas (kingmouf.delete@this.gmail.com) on October 26, 2016 2:48 am wrote:
> For a company the size of Apple, I believe that it can command any supplier for specific features and be the
> first to use (and maybe the only). This is custom design services. What you are saying about a benefit for
> Apple could then become a benefit for competitors is a possibility but there are two factors to consider: (i)
> when will this happen? If Apple benefits in generation n and the competitors in generation (n+1) then Apple
> still is far ahead, (ii) Apple from your writing will benefit the most from the combination of software + hardware,
> so even if the hardware benefits trickle down to competitors, I am not sure they can benefit.
It isn't quite that simple, because improvements are cumulative. If Apple comes up with an advance that improves performance by 10%, all their future designs are 10% faster than they otherwise would be. If they keep it to themselves, that's a long term 10% advantage over the competition. If they spec out what they want to Imagination, and that advance makes it into Imagination's products a year later, then Apple only has that 10% advantage for one year.
Given how secretive the GPU world is, it would seem that both NVidia and ATI feel the same way, wanting to keep their advances secret (or at least make it difficult for the other guys to find out about them via expensive reverse engineering)
> Further more, one of your arguments about debugging and reacting to bugs, I think that it is far easier to
> locate and correct bugs when multiple companies are looking at the same problem (from different perspectives
> and use cases) than a single company can do. In the case of debugging and using an error-free design (if
> such a thing exists), the more the better. I am not sure it works with the timings - you may be right that
> it will take more to correct a third-party IP than an in-house, but again I am not sure about that.
Obviously that's true, but if that benefit was the most important factor, all hardware and software would be open source. It is easier for Apple to use a buggy GPU than for Imagination to sell one, since Apple controls the whole stack. They can deliver fixes quickly, whereas if a buggy GPU is used in an Android phone it will take much longer for end users to get the fix - if they ever do!
> Now concerning time to market, I think that traditionally the argument about using third-party
> IP is exactly that: you can design faster by using the work that someone else has already completed
> and verified. With Apple commanding such a high volume of Imagination's business, do you think
> that if Apple sends them a request to fix a bug, it will be priority #1 or something else?
>
> There is also another thing that you need to consider. And this is risk. What if Apple discovers
> that its new design has a fault, is inadequate, bad manufacturability etc? This is the big
> risk with doing custom work and statistically checking the competitive landscape, all manufacturers
> have been hit by it. Apple can manage to hide it under the carpet since it controls all
> the stack and that is a big bonus. But the risk is still there.
Apple takes the same risk by designing custom CPUs instead of using ARM cores, or buying an off the shelf Qualcomm SoC. It is a lot easier to work around a GPU bug in software than it is to work around a CPU bug, so the added risk of a custom GPU is pretty tiny.
Given that they were able to front run ARM's own 64 bit cores, time to market is obviously not a problem on the CPU side. If they consider the GPU as important, and devote a similar amount of resources to it, they could do the same. Right now they are in a sort of halfway state where they have a partially custom design, but the recent transfer of 25 engineers from Imagination mentioned in the article suggests we haven't reached the end game yet.
The hiring is interesting because typically you'd expect a supplier contract to have a clause that prevented such hires. It may be that Apple plans to go full custom, has shared those plans with Imagination, and is performing the transition according to some mutually agreed plan. This is probably related to the retirement of the long-time CEO, divestiture unrelated product lines, etc. that happened last spring. Imagination is either preparing for life after Apple, or for being acquired by Apple.
> For a company the size of Apple, I believe that it can command any supplier for specific features and be the
> first to use (and maybe the only). This is custom design services. What you are saying about a benefit for
> Apple could then become a benefit for competitors is a possibility but there are two factors to consider: (i)
> when will this happen? If Apple benefits in generation n and the competitors in generation (n+1) then Apple
> still is far ahead, (ii) Apple from your writing will benefit the most from the combination of software + hardware,
> so even if the hardware benefits trickle down to competitors, I am not sure they can benefit.
It isn't quite that simple, because improvements are cumulative. If Apple comes up with an advance that improves performance by 10%, all their future designs are 10% faster than they otherwise would be. If they keep it to themselves, that's a long term 10% advantage over the competition. If they spec out what they want to Imagination, and that advance makes it into Imagination's products a year later, then Apple only has that 10% advantage for one year.
Given how secretive the GPU world is, it would seem that both NVidia and ATI feel the same way, wanting to keep their advances secret (or at least make it difficult for the other guys to find out about them via expensive reverse engineering)
> Further more, one of your arguments about debugging and reacting to bugs, I think that it is far easier to
> locate and correct bugs when multiple companies are looking at the same problem (from different perspectives
> and use cases) than a single company can do. In the case of debugging and using an error-free design (if
> such a thing exists), the more the better. I am not sure it works with the timings - you may be right that
> it will take more to correct a third-party IP than an in-house, but again I am not sure about that.
Obviously that's true, but if that benefit was the most important factor, all hardware and software would be open source. It is easier for Apple to use a buggy GPU than for Imagination to sell one, since Apple controls the whole stack. They can deliver fixes quickly, whereas if a buggy GPU is used in an Android phone it will take much longer for end users to get the fix - if they ever do!
> Now concerning time to market, I think that traditionally the argument about using third-party
> IP is exactly that: you can design faster by using the work that someone else has already completed
> and verified. With Apple commanding such a high volume of Imagination's business, do you think
> that if Apple sends them a request to fix a bug, it will be priority #1 or something else?
>
> There is also another thing that you need to consider. And this is risk. What if Apple discovers
> that its new design has a fault, is inadequate, bad manufacturability etc? This is the big
> risk with doing custom work and statistically checking the competitive landscape, all manufacturers
> have been hit by it. Apple can manage to hide it under the carpet since it controls all
> the stack and that is a big bonus. But the risk is still there.
Apple takes the same risk by designing custom CPUs instead of using ARM cores, or buying an off the shelf Qualcomm SoC. It is a lot easier to work around a GPU bug in software than it is to work around a CPU bug, so the added risk of a custom GPU is pretty tiny.
Given that they were able to front run ARM's own 64 bit cores, time to market is obviously not a problem on the CPU side. If they consider the GPU as important, and devote a similar amount of resources to it, they could do the same. Right now they are in a sort of halfway state where they have a partially custom design, but the recent transfer of 25 engineers from Imagination mentioned in the article suggests we haven't reached the end game yet.
The hiring is interesting because typically you'd expect a supplier contract to have a clause that prevented such hires. It may be that Apple plans to go full custom, has shared those plans with Imagination, and is performing the transition according to some mutually agreed plan. This is probably related to the retirement of the long-time CEO, divestiture unrelated product lines, etc. that happened last spring. Imagination is either preparing for life after Apple, or for being acquired by Apple.