It feels weird to not include some sort of ML inference tasks when benchmarking modern compute hardware. The GPU capabilities are a major factor of the M3 upgrade it seems?