3 ms·

Recently I bought a Mac Studio and wanted to experiment with Apple's GPU-accelerated ML API under the Metal Performance Shaders framework. I downloaded a sampl

by pavlov 4y ago

Recently I bought a Mac Studio and wanted to experiment with Apple's GPU-accelerated ML API under the Metal Performance Shaders framework. I downloaded a sample code project from WWDC 2019: https://developer.apple.com/documentation/metalperformanceshaders/training_a_neural_network_with_metal_performance_shaders https://developer.apple.com/documentation/metalperformancesh... It didn't build on latest Xcode on the Mac Studio. I'm experienced with Cocoa and Apple's APIs, but couldn't fix the problem in 30 minutes of poking around. Then I found another sample code project from WWDC 2020 which is apparently using a similarly named but different API for the same purpose: https://developer.apple.com/documentation/metalperformanceshadersgraph/training_a_neural_network_using_mps_graph https://developer.apple.com/documentation/metalperformancesh... This one looked promising, but failed with a runtime assertion and I was unable to figure it out. At this point I wish Apple spent more effort on making their existing frameworks usable. If the only available sample code doesn't even work on a brand new Mac, the API isn't going to be used by third parties.

Invictus0 4y ago
I agree. It's a miracle anything gets written for iOS at all. Poking around in Apple's documentation is just miserable.
hedgehog 4y ago
Depending what you want to do you do you might find it faster to prototype with the Python bindings: https://coremltools.readme.io/docs/model-prediction https://coremltools.readme.io/docs/model-prediction
- pavlov 4y ago
  I don't want to use CoreML or Python, I wanted to learn about the lower-level implementation using Metal Performance Shaders. And the only available sample code doesn't even run. This is sadly not a unique situation for Apple's more specialized public frameworks — they're more like semi-private because nobody outside Apple can make them work.
  - hedgehog 4y ago
    Ah, yeah, and that's an improvement over the historical state of their ML support. For inference the M1 is a nice platform (fastish and power efficient) but for research and training not fast or well supported enough to justify not using NVIDIA.

by pavlov 4y ago

Invictus0 4y ago

I agree. It's a miracle anything gets written for iOS at all. Poking around in Apple's documentation is just miserable.

hedgehog 4y ago

Depending what you want to do you do you might find it faster to prototype with the Python bindings: https://coremltools.readme.io/docs/model-prediction https://coremltools.readme.io/docs/model-prediction

pavlov 4y ago
I don't want to use CoreML or Python, I wanted to learn about the lower-level implementation using Metal Performance Shaders. And the only available sample code doesn't even run. This is sadly not a unique situation for Apple's more specialized public frameworks — they're more like semi-private because nobody outside Apple can make them work.
- hedgehog 4y ago
  Ah, yeah, and that's an improvement over the historical state of their ML support. For inference the M1 is a nice platform (fastish and power efficient) but for research and training not fast or well supported enough to justify not using NVIDIA.