Same speed. Python uses GIL by default, so all threads are actually sequential (interleaved) at the interpreter level, unless you mix in some native libraries and disable GIL while they run.

You'd need to run multiple processes instead (multiprocessing module) but at that point you'll need some way to have a shared concurrency-safe hashmap.

Moral? it's just a high-level kang, not a high-performance kang.


Well, that's it, we're screwed. My 3-kangaroo C program also runs on Raspberry Pi. I also have it an Android app with native JNI bindings to it. So every time I wake it up it runs a few batches to mine DPs. Next step is to implement kang on mobile GPUs using Vulcan.