GeeCoolGuest
GeeCoolFor the AI pipeline itself the iGPU is insurance, not performance. Prefill, embedding and quantization follow the same CPU-plus-GPU split either way; the integrated graphics matter when the discrete card throws a fit or a driver update bricks the display — you still have a screen to fix things from.
Take the premium if a spare display path has real value in your routine; skip it if every dollar should sit inside inference hardware. Cores, threads and the 5.6GHz boost are identical across both parts, and the iGPU's contribution ends there — do not expect it inside model math of any kind.