I love Erlang and the concept of multinode programming, but sometimes I wish it could lean back into the multicore/SIMD world. A lot of the time, actors are the right choice; but for things like nodes in a physics engine, you obviously want to take a big pile of them and stream them through a matrix multiplication or something of the sort, if you can.
If there was a language that let me define something in terms of an individual actor-modelled components, but then allowed you to optimize some of them into batched GPGPU passes (and apply strict simulation ticking and so forth in the process), I feel like it would "win" all sorts of domains. Erlang could be that language with a little work.
If there was a language that let me define something in terms of an individual actor-modelled components, but then allowed you to optimize some of them into batched GPGPU passes (and apply strict simulation ticking and so forth in the process), I feel like it would "win" all sorts of domains. Erlang could be that language with a little work.