Imagination Technologies IP licensing and execution

I'm willing to bet that that "multi-year" license includes also double digit SeriesX GPU IP generations

 

It could also be that Apple and IMG are really "into bed together" so it's a deeper cooperation than normal licensees incl. enhancements / secret sauce.

 

LOL it was pure sarcasm from my side because it seems that from now IMG will call each new fart in future hw a new "generation"

That doesn't of course mean that the above isn't true; I'd say that Apple should have a "John Carmack effect" on IMG; Kristof said once that when Carmack shouted "jump" IHVs just asked "how high?"

 

There is nothing mythical about Apples in-house graphics effort. They have quite clearly spelled out what the facility and its engineers are doing.

Having options is good. Since part of their payment is in per SoC royalties, they can cut costs if they shift to in-house solutions early. If, however, their own effort doesn't produce good enough results, they have their needs covered. And it sounds as if future development at ImgTech is ensured as well.
Seems like a welcome deal all around.

 

Is anybody else using PowerVR now?

Seems like a lot of makers went with Snapdragon last year.

 

Qualcomm always had & continues to have a sizeable market share. According to JPRs last report in terms of GPU integration QCOM &: IMG were both around 30+% each, ARM & Vivante around 10% each.

Things have apparently slowed down for them after TI jumped ship, but given the amount of recent license deals they should recover probably next year. To avoid misunderstandings their market share shrunk by a bit and not their sales volume. They sold something over 100Mio non-MIPS IP cores more then the year before.

 

I don't see why Apple doesn't just cough up the billion it would take to buy IMG if they want to handle everything in house ... between only just hiring a team and still having to handle all the IP issues it's going to take fucking ages before they get anything ready, this isn't like with processors where Intrinsity got them up and running fast.

 

They do a lot of 8 and some 9 figure acquisitions but I can't recall them doing a billion dollar one.

 

And you'd think no one would object?

 

http://www.macrumors.com/2014/02/06/apple-repurchases-fourteen-billion/

Coincidentally it's just being reported that Cook doesn't mind large 10 figure acquisitions, if it's the right fit of course. Beyond fit, I thought Intel's stake in Imagination is similar in size to Apple, no doubt to prevent this very thing. There'd probably need to be government competition approval as well.

 

By the way since there's nothing else but public records to go by, the G6430 had been announced from IMG in June 2012. Now assuming that it wasn't available for licensing earlier, it took Apple exactly a year to integrate it in the A7 SoC. That's not what I'd call "f***ing ages" rather the contrary and no IP announcement to integration isn't anywhere shorter for ARM A5x 64bit cores and Apple Cyclone.

Entropy,

No there isn't. However can you guarantee what exactly it is for? Ask anyone in the background how many billions Samsung had spent recently on a flopped own GPU design and yet they're still using ARM Mali GPU IP. Obviously I'd trust Apple engineering far better success than at Samsung LSI, but there's no guarantee for anything yet either. There's no necessity for expanding a long term commitment (which lasted for 8 years) with another long term commitment. You just extend a licensing deal and that's about it.

Do you know how many in the background are sure that Apple is also developing it's own CPU? If yes then for what (same question is above). In the meantime others as above are asking why Apple isn't buying IMG. Why not buy ARM then with the same reasoning? Why would Apple even bother spending a billion for IMG if it's so guaranteed that they are developing the ultimate GPU that will render anything else useless? Does Apple really need to worry about a few bucks in royalties both for CPU and GPU IP with their profit margins?

Yes they are collecting engineers left and right, but who's to convince me that for the time being their role is not amongst others to integrate any IP with a record pace into their SoCs. Even if Apple is just by say 6 months ahead their competitors for integration does it come for free or do they have a magic wand for it?

 

To address the last point first, I'll quote an Apple job posting from this autumn.

A manager position, obviously. It is backed up by a number of other job ads before and after. Seriously, why would you doubt that they are doing what they say they are - "Driving, tracking and executing GFX IP development owned by site from inception to retirement"?

However, if you are developing new IP, then it also stands to reason that you can't really say at what state it will be at any given time, nor what your suppliers and competitors may have come up with in the meantime. If I were in Apples shoes, I would realize that the cat would be out of the bag at some point, and at a point which is way to early to commit to an in-house solution only. And that this may make the relationship with my supplier a bit strained. For Apple developing in-house IP is a way of ensuring that you don't get your legs cut off from underneath you by a supplier. Apple have no wish to do it to themselves by saying to their supplier that from now on they have to target the needs of other customers because Apple won't use their services in the future - this would only put Apple at risk! So a deal like the above referenced makes good sense. Apple promises ImgTech a steady stream of revenue, coupled with a load of cash up front, demonstrating that they are still interested in IMGTech products going forward, and ensuring that ImgTech has a vested interest in putting their best foot forward for their largest customer. It also allows them breathing room for their own efforts that may or may not pay off in the long run, but which regardless of competitiveness always ensures a level of safety and independence from IP supplier woes.

I hope that ImgTech will work hard to ensure that their products provide competitive advantages over any in-house solution, Apples team will work hard to come up with something insanely great, and both teams will work their butts off to ensure competitiveness with the real enemies at Qualcomm, ARM, Intel, and others.
This is a great time to have an interest in mobile technology.

 

Is it possible that Apple are developping their own GPU for a different class of SoC? Or even for discrete parts?

 

With Apple's obnoxious tendency to secrecy it's hardly ever easy to find out what they're really cooking. However Apple doesn't serve only iOS so yes any educated guess can work but isn't a guarantee for anything either.

A couple of more things:

* I wasn't even aware that Caustic was a small startup from ex-Apple veterans. It could be a pure coincidence that IMG bought them, but I have at the same time severe doubts that Apple has no interest at all in RT. Next best question would be why Apple didn't buy them in the first place?

* Has Apple really a high interest in 3D after all because of X future plans or not? If they don't care, why get your hands dirty and waste billions on GPU development. If they do is it better to have for the ULP SoC market something completely unique or a more wide spread architecture to ensure a wee bit more developer support?

* Let's have a quick stop gap at competing GPU solutions:
ARM Mali: a whole damn lot of low end stuff and their only rather sizeable volume deal is with Samsung. Are the repeated power problems Samsung faced just a myth or is there a bit more than just smoke behind it? Either way any so far T6xx variant doesn't strike me as to have a perf/mW advantage over Rogue GPUs but I'd love to be convinced of the opposite.

Vivante GC: on a recent rise but from what I hear mostly due to them selling 2D GPU IP like hot cakes as amongst others for TI OMAP4470; they've for sure their own share of GPU IP sales but we'll see next year how they're sales volumes shape up. Yes their IP cores don't consume much die area, however they're not any 3D race winners either. I wouldn't underestimate them however.

NVIDIA ULP GeForce: along with AMD they're the master of the notebook, desktop and workstation GPU markets. Now the GK20A might sound impressive on paper with a whole damn lot of colourful promises and expectations, but at the end of the day I'm as "prejudiced" (or not) to not expect any perf/mW advantage against the future Apple A8 GPU.

Qualcomm Adrenos: when I see in the initial Galaxy S4 variants as just one example out of many the 320 GPU to clock in 3D games at 320MHz and in benchmarks at 450MHz do I really need to expand on that one? I'll get to the point and it for sure isn't to undermine the competition.

The point here is if someone like NVIDIA has a hard time to actually beat on any metric so far any PowerVR GPU IP why should I be convinced that Apple with no GPU experience, no related patent portofolio whatsoever can overnight create something that would beat its current solutions by a sizeable margin.

There's already enough material in this post to get potentially lynched by others for my thoughts so let's add another one: Apple should actually license Cyclone right back to ARM in order to show them how much the A15 stinks for smartphone SoCs. Level with me for a second: here you have quite a big efficiency advantage for Apple's custom CPU vs. ARM's own CPU IP offerings. If Apple now has managed or will manage to create an own GPU with such a difference in efficiency compared to any GPU IP out there in the ULP SoC market, by all means let them bring it on asap.

 

It is possible but unlikely to be their primary goal.
Overwhelmingly, their revenue comes from, let's call it, "high-end mobile" solutions. So if they want to ensure IP supply line independence, that area is the bread and butter they need to protect. Also, at this point in time, it is where the money is. The interval from phones to mobile computers is where there is a combination of volume and margins that makes it interesting to compete. Graphics is very scalable due to the workload being trivially parallel, so covering that span with a single architecture is definitely possible. Hell nVidia claims to be doing phones to supercomputing with the same architecture!
I don't know if, uhm, wearable displays or whatever would require a fundamentally different approach than for instance phones. But I wouldn't assume that Apple builds a Gfx R&D center to explore that question alone.

 

Note that there's a huge difference between iOS and MacOS GPU requirements. I'd dare to say that the latter is a lot easier up to a point; while a hw designer cannot obviously be reckless with power consumption it's by far not such a headache as with anything ULP.

And yes of course is it possible to cover with one architecture a wider field, but ask Intel why they aren't using GenX GPUs for smartphone SoCs or how many chances Tegra K1 will have in smartphone design wins.

 

I doubt you will see anything in the public domain at all, until it ships in physical form. If then.
Patent trawling may be the only source of information that ever gets publicized, and in my own field I've never seen a patent application being submitted other than at the absolutely last instant. (And sometimes things are just kept secret. Safer than patents, since patents ensure that your competition know what you're up to and can copy it, and depending on what the product is, you may still have a devil of a time protecting your patent. I have no idea how for instance ImgTech can be absolutely sure there is nothing in nVidia K1 that violates any of their patents. It has to be awfully difficult, if all you ever see is a chip that receives API calls, and spits out pixels. Is disassembling drivers enough? Is it done? It may be simpler and safer to just keep quiet.)