Latest Mesa AGX Work Points To More Apple M1/M2 Similarities With PowerVR Graphics

[phoronix]

Phoronix: Latest Mesa AGX Work Points To More Apple M1/M2 Similarities With PowerVR Graphics

Since Apple introduced the M1 two years ago as their in-house Apple Silicon for laptops and desktops with a powerful AArch64 processor and custom-designed graphics processor, there has been much speculation about whether the Apple M1 (and now M2) graphics are a clean-sheet Apple design or derived from Imagination PowerVR graphics that Apple had been using with earlier SoCs. There has been some similarities brought up before with the Asahi Linux team working on enabling the Apple M1/M2 under Linux while the latest Mesa driver activity points to more common bits between PowerVR graphics hardware and the Apple AGX graphics...

https://www.phoronix.com/news/Mesa-AGX-More-PVR-Reference

[carewolf]

Hasn't that been known for a while, that Apple basically copied the API of PowerVR, so they didn't have to change their drivers that much?

[CommunityMember]

Hasn't that been known for a while, that Apple basically copied the API of PowerVR, so they didn't have to change their drivers that much?

While the details and contracts have typically been under various forms of NDA, the fact that Apple (eventually, after being sued) agreed to pay Imagination licensee fees for (at least some) Imagination IP strongly suggested that Apple's new GPU was not a true clean-sheet design. Designing a new clean-sheet GPU that is competitive is hard work that takes time (look to Intel's ARC), and while Apple either has, or can hire, engineers to do so, they have not yet released such a part (the time to do so would likely be a new Mac Pro with such a new GPU architecture).

[Developer12]

Hasn't that been known for a while, that Apple basically copied the API of PowerVR, so they didn't have to change their drivers that much?

The real question has been how much of PowerVR survived in the design, which is still somewhat unresolved. Even between the M1 and iphone GPUs there have been sizable architectural changes.

[Developer12]

I expect the biggest limiting factor on upstreaming will be how quickly rust gets into the mainline kernel. Not only does the basic rust support, currently being pushed, need to get in but also all of the bindings that the GPU driver is dependent on. One example is the entirely new binding that Asahi Lina needed to invent for GEM. The rust patchset has been broken up so each binding can be reviewed in-turn, but that means it will probably take time for each of them to make their way in.

[stormcrow]

Hasn't that been known for a while, that Apple basically copied the API of PowerVR, so they didn't have to change their drivers that much?

Presenting a familiar API to developers is one thing, but an API can hide the hardware interfaces behind it which may not be anything like the PowerVR chips themselves. That's not to say there isn't PowerVR IP in the M GPUs, only that they don't actually have to be anything beyond the software interfaces that resembles the originating technology. APIs are both patentable (traditionally) and copyrightable (thanks to Oracle v. Google) in the US. That's not to say there aren't interop fair use arguments that aren't applicable, but in this case I doubt they would apply.

How much the hardware behind the interfaces resembles PowerVR is an open question.

[SteamPunker]

APIs are both patentable (traditionally) and copyrightable (thanks to Oracle v. Google) in the US.

Sorry , but I have to correct you there. Justice Stephen Breyer (since retired) wrote in his majority opinion that even if APIs were copyrightable, then implementing or using third party APIs would still constitute fair use, so Google did not infringe Sun's (later Oracle's) IP by implementing Java APIs in android. In the Google v Oracle decision, SCOTUS declined to settle the question of whether APIs are copyrightable in the first place.

[coder]

Presenting a familiar API to developers is one thing, but an API can hide the hardware interfaces behind it which may not be anything like the PowerVR chips themselves.

Not likely. If the hardware works in a fundamentally different way, the API needs to reflect that for the sake of efficiency (not to mention reducing complexity). And Apple cares about nothing more than efficiency.

[Dukenukemx]

I made that prediction about Apple's GPU's. Turns out it took nearly a month to find out.

https://www.phoronix.com/forums/forum/linux-graphics-x-org-drivers/x-org-drm/1340093-asahi-linux-may-pursue-writing-apple-silicon-gpu-driver-in-rust?p=1340130#post1340130

Developer12 was wrong.

https://www.phoronix.com/forums/forum/linux-graphics-x-org-drivers/x-org-drm/1340093-asahi-linux-may-pursue-writing-apple-silicon-gpu-driver-in-rust/page2#post1340143

[Boland]

I made that prediction about Apple's GPU's. Turns out it took nearly a month to find out.

https://www.phoronix.com/forums/forum/linux-graphics-x-org-drivers/x-org-drm/1340093-asahi-linux-may-pursue-writing-apple-silicon-gpu-driver-in-rust?p=1340130#post1340130

Developer12 was wrong.
https://www.phoronix.com/forums/foru...e2#post1340143

Apple never stopped paying imagination for their IP. The AGX is a TBDR GPU after all. And imagination hold bucket loads of patents related to that tech.

But from what Andrei hinted at before he left Anandtech, the actual cores in Apples GPUs diverged a lot from the powerVR designs.