I've added MTLCompilerService crash-log, provided by user, to feedback-assistance ticket.

I've added permalink to shader, here in first post and in feedback assistant: [quote='772346021, Try9998, /thread/772346, /profile/Try9998'] https://shader-playground.timjones.io/27565de40391f62f078c891077ba758c [/quote]

Done: https://feedbackassistant.apple.com/feedback/16297235. Thanks for pointing out to this.

OK, thanks. useResources(every_blas) helped. Yet, need to note that this behavior is not documented on developer webside - so please update docs. Also making call to useResource* and tracking blas'ses introduces large overhead for engine. Is it possible, can you change api then? Metal knows list of blas'es anyway, from when tlas was created.

There can't be missing useResource, as they do not required: no argument buffers are in use(or similar). Tested locally useResource(tlas) - no changes

I have culling in shader. Without it it's 3X slower. GPU wise M1 is only hardware that I have with Mac.

such as, dynamic geometry expansion / culling Hmm, let me clarify: this is the case. There are 2 types of culling implemented in the game: Frustrum (in shadow-pass) 1.1 For whole object bounding sphere, in case if props/animated objects 1.2 Per cluster, for landscape 2. HiZ (in main-pass) Frustrum culling is done as part of mesh-stage, not a task-shader. Task/Object shaders are still TODO for me. so you should try both paths on different GPUs Unfortunately I don't have any other GPU :(