Performance implications of using lowp precision qualifiers on varyings

I was considering the usage of lowp varyings for cases where we could potentially afford the lack of precision. Am I right in assuming that there will be a 4x (presumable) savings in bandwidth (write + read) when compared to mediump? Or could this be strongly HW (GPU model/family) dependent?

According to Arm Mali GPU Best Practices Developer Guide there's some encouragement to drop some things to mediump but not lowp. So I was wondering what's Arm's guidance on this (even if it depends on certai GPU families)?

More questions in this forum