every GPU cluster we've found is fairly generic with the same setup i think this is because they just want to support a wide range of workloads without overspecializing but this is why no one can match deepseek's price on inference. we're going to have to build our own cluster