That doesn't make much sense to me. The GPU's task energy is so much lower than even the e cores, and AFIAK the GPU's compute isn't even fully utilized for loca

That doesn't make much sense to me. The GPU's task energy is so much lower than even the e cores, and AFIAK the GPU's compute isn't even fully utilized for local inference.