Comment by fooker
Is there publicly available evidence that TPU perf/$ is better than Blackwell ?
I know it seems intuitively true but was surprised to not really find evidence for it.
Is there publicly available evidence that TPU perf/$ is better than Blackwell ?
I know it seems intuitively true but was surprised to not really find evidence for it.
yeah ... poly market and other makers seem to be betting that Google by year's end or sometime next year or so will have teh best gen ai models on the market ... but I've been using Claude sonnet 4.5 with GitHub Copilot and swear by it.
anyways would be nice to really see some apples-to-apples benchmarks of the TPU vs Nvidia hardware but how would that work given CUDA is not hardware agnostic?