@prologic@twtxt.net yeah. I’d add “Big Data” to that hype list, and I’m sure there are a bunch more that I’m forgetting.
On the topic of a GPU cluster, the optimal design is going to depend a lot on what workloads you intend to run on it. The weakest link in these things is the data transfer rate, but that won’t matter too much for compute-heavy workloads. If your workloads are going to involve a lot of data, though, you’d be better off with a smaller number of high-VRAM cards than with a larger number of interconnected cards. I guess that’s hardware engineering 101 stuff, but still…