More Progress Is Made Understanding Apple's M1 GPU, Working Towards An Open Driver

[GruenSein]

I find it baffling that there are always some people who get really upset because they think that Apple intentionally breaks OSS support or cripples OpenGL/Vulkan performance when the easiest explanation is that they simply do not care about OSS or whether something outside of their ecosystem breaks.
Regarding the lacking hardware features: In the end, the goal of all GPUs is the same so they certainly provide some way to do it in an efficient fashion. The question is not whether or not it can run stuff people do with Vulkan. Vulkan is just a way of telling the GPU what to do. Same goes for Metal. The question is how easy it is to translate the intention phrased in Vulkan to something the M1 understands.

[Imroy]

May an Ampere Altra 8-core mini-sized computer exist.

I'm more hopeful for RISC-V myself.

[stingray454]

Still great news. While I don't particularly care for Apple, I must admit that their M1 chips are impressive. With rumors about M1X / M2 and so on, and computers with more ports and such, it's an interesting option. Sure I'd love to see other ARM-based laptops with similar performance, especially for emulating x86 software which we probably will need to for a while longer.. But as it stands, it seems to be 1-2 years before we even have something competitive on the market. Would consider an M1 if/when the hardware is supported, so great to see some progress.

[tildearrow]

May the M1 die in fire. May the whole Apple die in fire. The sooner the better.

May an Ampere Altra 8-core mini-sized computer exist.

[kpedersen]

May the M1 die in fire. May the whole Apple die in fire. The sooner the better.

Kudos to anyone working to free it: I myself would not spend a single second of mine to reverse engineer anything these days.

Agreed. Fair enough if she is doing it for fun but we will *always* be playing catchup. Apple will *always* break the work in the next release.

For average consumers, we should ditch the trash and buy stuff that is intended to work for us.

[lucrus]

May the M1 die in fire. May the whole Apple die in fire. The sooner the better.

Kudos to anyone working to free it: I myself would not spend a single second of mine to reverse engineer anything these days. It's not the early 90's anymore, when Linux desperately needed to work on something and reverse engineering was more or less the only way out.

Today we have 3 out of the 4 most relevant graphics hardware makers that support Linux one way or the other (yes, I'm including nvidia too, though I'd never buy it again), and one that stands in the way on purpose instead. Let it die in fire, we do not need bitten and quickly rotting fruits anymore (had we ever needed those anyway?).

[Ironmask]

How adorable, they specifically designed M1 to not support Vulkan so when benchmarked, Metal will out-perform it. It's just like how they intentionally slowed down iTunes on Windows. This company is nothing but smoke and mirrors.

[smitty3268]

I have an idea. Why not support exclusively Vulkan, and then do the rest using Zink (that keeps getting faster and faster)?
This way you could finish the driver in one year or two.

I think Alyssa probably has more of a background in GLES, which I imagine is why she's starting there. But it's certainly an interesting idea to rely on Zink for this hardware. Probably still quite a ways to go before it's really relevant - lots of stuff like getting the shader compiler to work with control flow, texturing support, etc. is going to be shared no matter what API you support and is a lot of work to get going.

[mangeek]

"Why not support exclusively Vulkan"

I'm not expert, but it sounds like the hardware doesn't necessarily have all the features it would need to implement Vulkan completely. Apple built something that was made for Metal. This might end up being a weird driver that does some things in CPU/software to get Vulkan working well enough to run Zink on.

[emblemparade]

Such an interesting read. Apple's Metal is indeed even lower-level than Vulkan or DX12. There are some obvious advantages at making the GPU as simple as possible while letting the CPU handle the overhead required for managing APIs. I have to wonder if Microsoft's success in pushing DirectX as a de facto standard in gaming has led to a lot of cruft (at an expense) in GPUs in order to compete in accelerating that API series.