Apple Fires Back at Supplier Imagination in Contract Dispute!

[Daniel Nenni]

Just when you think things could not get worse for Imagination Technologies, Apple is now disputing the timing of the public statements made by IMGTECH regarding the termination of their partnership. Apple says IMGTECH knew back in 2015 but did not release that information until 2017. This makes complete sense to me based on my experience in IP licensing. IMGTECH had to know it was coming based on both direct and indirect information. I certainly did when I did contracts and you can bet there is a paper trail so this will surely go legal...

View attachment 20090

Apple first informed Imagination in late 2015 that it would no longer be buying the U.K. company’s latest technology, Apple said in a statement to Bloomberg. It continued using its older systems.

By 2016, Apple said it told Imagination it was further diminishing the relationship by initiating a clause in its contact that allows Apple to pay a lower royalty rate for using a smaller amount of intellectual property. By February of this year, Apple said it told Imagination it was ending the relationship

Apple’s statement clashes with Imagination’s time line of events. On a conference call with investors this week, Imagination CEO Andrew Heath said the company was informed by Apple at the end of March "that they were certain" that products to be released in 2018 or early 2019 will no longer use Imagination’s intellectual property.

Apple Fires Back at Supplier Imagination in Contract Dispute - Bloomberg

I'm still wondering why Synopsys has not acquired them to complete their IP monopoly. Maybe the legal dust needs to settle? I was told that Cadence and Imagination do not get along at all so they are not a likely suitor.

 

[een5afr]

Well, around late 2015, Apple did put up job vacancies on it's careers page related to UK based GPU HW/SW roles (in London). After which the top brass of HW/SW architects started to leave the company one by one and head towards London. Hence, it's fair to say that Imagination had known this had to happen someday.

Note that ARM was in a similar situation when Apple decided to build it's own in-house ARM arch based mobile SoC. But Apple still retained the ARM architecture license through which ARM can still receive royalty money from iPhone sales, giving Apple the complete freedom to tweak around the underlying u-arch to squeeze every ounce of performance from their custom silicon.

This time around, Apple has decided not to retain the architecture license for PowerVR GPU technology i.e. no royalty payment for IMGTEC from future iPhone products. This is where the problem lies! IMGTEC is not complaining about Apple not using their GPU IP. Rather, they're protesting about Apple's decision to move away from the PowerVR arch completely - since, according to them, it's not possible for Apple to (a) define a new arch, (b) design and verify the core and (c) build up a stable and optimized codebase in just 2 years time.

There's some rationale behind this claim. Low Power GPU architecture is quite a mature domain dominated by IMGTEC, ARM, Adreno and Vivente (and to some extent Nvidia). There's very little room for something radically different without violating any of the existing patents of the aforementioned companies. IMGTEC has been operating in this domain since 1985 and have amassed huge number of patents in this domain.

 

[name99]

@een5afr
You're assuming Apple is interested in building yet another GPU. This strikes me as unlikely; if ALL you care about is GPU's, PowerVR is about as good as it gets.
The issue, IMHO, is that, just like with ARM, Apple wants to run at its own pace, and lost patience with PowerVR precisely when it saw that Furian was GRAPHICS-optimized (as opposed to general-compute optimized) and area-optimized.

Apple doesn't want a GPU, they want a throughput engine to complement their existing latency engine. This is obvious when you see everything they were pushing at WWDC --- all about computation photography, machine learning, pouring data through image filtering to recognizers to renderers that display the data that's been acquired in new ways and aggregated with streams of data acquired from other sources like the camera.
They're way beyond interest in engines whose primary skill is in processing OpenGL!

A throughput engine done right, and that they control, means among other things:

- code can be shifted from the latency engine to the throughput engine and back as easily as it is shifted from the performance to the low-energy core. This would, in the best case, require the throughput engine to implement some (subset of) AArch64, but could be done at a little more hassle with a different instruction set and just sharing address space and coherency.

- being able to ramp the frequency of the throughput core up and down, like that of the latency core, so that power can be balanced across both the way Intel now does.

 

[een5afr]

@name99 ... very informative post ... thank you for sharing your insight.