So, when do you actually need to pull in a separate caching service?

we're currently running redis as an in-memory cache alongside our main application instances. its been fantastic for speeding up reads and reducing database load. but, as our traffic grows, we're starting to see some contention on the redis instance during peak loads, impacting both cache performance and application responsiveness slightly. we're wondering at what point do you all consider it worthwhile to extract redis into its own dedicated, potentially clustered, service, rather than keeping it as a co-located resource? what are the key metrics or indicators you look for to make that decision?