3 ms·

> Regrettably, the M1 doesn’t map well to any graphics standard newer than OpenGL ES 3.1. While Vulkan makes some of these features optional, the missing featur

> Regrettably, the M1 doesn’t map well to any graphics standard newer than OpenGL ES 3.1. While Vulkan makes some of these features optional, the missing features are required to layer DirectX and OpenGL on top. No existing solution on M1 gets past the OpenGL 4.1 feature set. I'm very curious to know the performance impact of this, particularly compared to using Metal on macOS. (I'm sure the answer is "it depends", but still.) It's possible the article answers this question, but I didn't understand most of it. :(

gilgoomesh 3y ago
There isn't necessarily much difference between implementing features in driver compute code versus GPU hardware support. Even the "hardware support" is usually implemented in GPU microcode. It often goes through the same silicon. Any feature could hit a performance bottleneck and it's hard to know which feature will bottleneck until you try.
ribit 3y ago
Alyssa chooses some very odd language here, it seems to me. Yes, Apple GPUs do not support geometry shaders natively because geometry shaders are a bad design and do not map well to GPU hardware (geometry shaders are known to be slow even on hardware that allegedly supports it — there is a reason why Nvidia went ahead to design mesh shading). Transform feedback (ability to write transformed vertex data back to memory) is another feature that is often brought up in these discussions, but Apple GPUs can write to arbitrary memory locations from any shader stage, which makes transform feedback entirely superfluous. The core of the issue is that Apple chose to implement a streamlined compute architecture, and they have cut a lot of legacy cruft and things that were known not to work well in the process. I don't think that the rhetorics of "M1 getting stuck at OpenGL 4.1" is appropriate. I stopped following OpenGL many years ago, so I don't know specifically which features past 4.1 she might refer to. What I can say is that I'd be very surprised if there is something that OpenGL offers that cannot be done in Metal, but there are plenty of things possible in Metal that cannot be done at all in OpenGL (starting with a fact that Metal shading language has fully featured pointers).

gilgoomesh 3y ago

There isn't necessarily much difference between implementing features in driver compute code versus GPU hardware support. Even the "hardware support" is usually implemented in GPU microcode. It often goes through the same silicon. Any feature could hit a performance bottleneck and it's hard to know which feature will bottleneck until you try.

ribit 3y ago

Alyssa chooses some very odd language here, it seems to me. Yes, Apple GPUs do not support geometry shaders natively because geometry shaders are a bad design and do not map well to GPU hardware (geometry shaders are known to be slow even on hardware that allegedly supports it — there is a reason why Nvidia went ahead to design mesh shading). Transform feedback (ability to write transformed vertex data back to memory) is another feature that is often brought up in these discussions, but Apple GPUs can write to arbitrary memory locations from any shader stage, which makes transform feedback entirely superfluous. The core of the issue is that Apple chose to implement a streamlined compute architecture, and they have cut a lot of legacy cruft and things that were known not to work well in the process. I don't think that the rhetorics of "M1 getting stuck at OpenGL 4.1" is appropriate. I stopped following OpenGL many years ago, so I don't know specifically which features past 4.1 she might refer to. What I can say is that I'd be very surprised if there is something that OpenGL offers that cannot be done in Metal, but there are plenty of things possible in Metal that cannot be done at all in OpenGL (starting with a fact that Metal shading language has fully featured pointers).