Dissecting the Apple M1 GPU, the end

Too bad it was not Apple who hired her for M4, but in business leaders are always the most closed ones.

Apple is too much about beeing closed and creating barriers not sure that would have been a good fit. Plus that's a good way to flee a country quickly degrading.

Thanks Jesus it's Intel and not Apple, Intel has been extremely good at working upstream and has immense contributions in the Linux kernel, mesa, and elsewhere. Wasting such talent on Apple would make the world worse for us all.

Honestly if Apple had embraced Linux, the Apple Silicon CPUs would have been amazing for all sorts of server, scientific, and AI/LLM work. Too bad they are clamping down on the walled garden to focus on consumer toys instead.

> Honestly if Apple had embraced Linux, the Apple Silicon CPUs would have been amazing for all sorts of server, scientific, and AI/LLM work.

You can already do this work on M1/M2 using Asahi. A compute server doesn't need fully working peripherals and external displays.

I do have an M2 Macbook running Asahi, which works amazingly well for my casual use, but I think that there is no way that anyone will use last last gen hardware on a volunteer-developed OS for any actual work, server use cases, and so on.

How would GPGPU work in such a scenario?

I don't see a future for intel, frankly, but I'm very happy she found a good paying job.

Really a matter of perspective, tbh. Linux is also quite toy-like in its mess.

The M1/M2 GPU is supported via Vulkan Compute. (Or OpenCL/SYCL, going through rusticl.)

Thank you!

The real shame is the longevity, M1 Pro and M1 Max got discontinued two and a half years ago so they're on their way to the vintage list and could be entirely obsoleted by the end of this decade! Linux support is the only thing that will keep these machines usable after that.

There was a time when people said that about AMD.

Don't get me wrong, Intel's outlook is IMO currently indeed rather bleak, but I would not completely write it off just yet.

Why? Apple makes good products. It seems, unlike Intel.

> There was a time when people said that about AMD.

And Apple, to complete the circle.

From what I recall, Apple forbids its employees from participating in open source work it doesn't approve of. And given Apple's culture of secrecy, its agenda of maintaining a walled garden with their products, and her work basically contradicting the two, I doubt her being hired by Apple would benefit anyone other than Apple.

Apple doesn’t contribute back to the community.

Apple doesn't have any contributions to the Linux kernel or other parts of the Linux graphics stack. They are unlikely to hire someone who wants to work on open source.

Apple hiring her would essentially prevent her from doing it again on other models keeping the moat intact

Intel seem to be pretty good in the .Net framework too, pushing a bunch of performance upgrades.

To be fair they don't have anything to do with Linux so there is nothing to contribute back towards. They use BSD licensed software for a reason.

Apple does have open source projects. https://opensource.apple.com But the scope is rather limited. For someone of Alyssa's skillset there really isn't anything there.

AMD is equally fucked. Building off of IP-locked architectures is just a graveyard. Even apple will hit a wall one day.

Sure but linux isn't necessary. Legacy software can be virtualized

There are a myriad of companies that have thrived in "IP locked" environments, a host that have failed too. Equally there are heaps that have thrived and failed in "IP open" environments.

I think at best you could say it's more challenging or perhaps risky being a bit restricted with IP, but I'd call it miles away from a "graveyard".

You can hardly call Intel/amd/qualcomm etc all struggling due to the architectures being locked down.

Look at powerpc/Isa. It's (entirely?) open and hasn't really done any better than x86.

Fundamentally you're going to be tied to backwards compatibility to some extent. You're limited to evolution, not revolution. And I don't think x86 had failed to evolve? (eg avx10 is very new)