We somewhat have this with object picking. You can render objects with unique IDs to a framebuffer, so it's all in parallel. Then you just query the buffer at y

We somewhat have this with object picking. You can render objects with unique IDs to a framebuffer, so it's all in parallel. Then you just query the buffer at your mouse coordinates. Of course you lose some properties of the object and it's on a one frame delay. There's the added complexity of collision response, working in 3-dimensions, etc. But, to the point of just "eyeballing" collision, we can actually parallelize it!