AI performance per watt

https://extra-wiki.win/index.php/Why_an_Open_AI_Platform_Matters_for_Scalable_Machine_Learning

AI performance per watt is the real metric that matters when you're running machine learning models at scale—it's not just about how fast a chip can crunch data, but how efficiently it does so without jacking up your electricity bill. Switching from older GPUs to newer ones designed specifically for AI workloads, like those with tensor cores or sparse compute, can double your throughput while cutting power draw nearly in half, which is huge for data centers trying to stay green and profitable.