Q1 of two

I’m posting two separate threads because, whilst linked, the questions are very much different.

Today (on linux) I’m suddenly getting CUDA compilation failed errors (separate thread), and to ensure that it isn’t my scene, I switched testingly to a low sample CPU render. GPU is a 670, 2GB, and CPU is a Phenom 965 X4. On the GPU (when it was working) I needed to take it to 2000 samples for the render, and it was still 5 - 6 hours. On my CPU test, testing with just 50 samples (for speed) the render is as good as 200 - 300 on GPU.

Is there a fundamental difference between the two that means CPU will produce better images at fewer samples? If so, where’s the sweetspot?