Because it's more optimised towards AI, you start off with an exynos chip which is ok for gaming and ok-ish as far as neural networks are concerned. The more you optimise it for AI tasks, the less easy it is to make it as available as it was before optimisation for rendering. So a G3 isn't meant to be architecturaly better at gaming than a G2, the goal is to be way better at AI, and unless there is a leap in exynos architecture, that should increase with tha G4... Until google get it's independant chip designs around ARM achitecture going. Until then they have to work within the limits of exynos limitations.