Doing the "wrong things" in parallel for a net benefit is already a very common theme in all parallel computing, especially in GPUs. All interesting parallel al

Doing the "wrong things" in parallel for a net benefit is already a very common theme in all parallel computing, especially in GPUs. All interesting parallel algorithms (more than just a parallel-for over tasks) do more total work than their serial versions, but have shorter critical sections. I found Feynman's lecture and it's a good analogy for the differences between GPUs and CPUs. The GPU has thousands of dumb clerks who are fast at using their scratch paper and doing a few different arithmetic operations, but there's only one boss giving the whole group their instructions. https://youtu.be/EKWGGDXe5MA?t=6m https://youtu.be/EKWGGDXe5MA?t=6m