newer process node are smaller but very expensive compared to mature ones, each wafer from TSMC latest process is costly and with lower yield due to GPU large die size (+700mm2 compared to around 60mm2 per DRAM die)