haxxiy on 04 February 2023
A worrying trend: flagship GPUs have increased power consumption by ~15% per generation (in real-world scenarios, although the result of estimating it using TDP is similar). This is happening against a backdrop of decreasing gains in efficiency with rare exceptions due to architectural improvements.
So far it has been used to average ~50% gains per generation for the last decade, but ahead of us lies a bunch of nodes with worse power characteristics all the way to the end of CMOS die shrinking.
Will we see a 5090 with a 560W TDP to get to that 40% improvement? A 6090 keeping said TDP but consuming closer to it? A 7090 going for 650W? Or will consumer pushback and/or a novel innovative architecture save us from going that path?
