There are many different recommendations floating around, typically the limit depends on how well you know Cassandra and your workload.
If your workload is CPU bound, you should go for more, less dense nodes. If not, you can sustain higher data density per node. Typically I'd say the usable range is anywhere from 800GB - 4TB, with a lower amount for read heavy workloads. Aiming to keep nodes between 50-70% capacity. Keep in mind that there are a lot of caveats to very dense nodes (>2GB). The most important ones are bootstrapping/replacement times and repairs. These will be greatly affected by how dense your nodes are. Compaction is another notable one as that will only cause problems long after you've started. You really need a good replacement strategy and an operations team that really know their stuff before embarking on the very dense nodes adventure. Start with smaller nodes if this is your first experience with Cassandra, even if it costs you more.