Does >8 GB RAM reduce latency? + device recommendation

[Alexei-WLU]

I’m evaluating Rtranslator for a hands-free, walkie-talkie–style interpreter between Russian and English (one phone on the table that speaks aloud) for personal use by a patient in settings like nursing homes, hospitals, or doctor visits. For this use case, translation latency is critical.

Assuming four Android devices with the same CPU/chipset and no other user-launched apps:
8 GB RAM

10 GB RAM

12 GB RAM

16 GB RAM

All else equal, would 10/12/16 GB deliver measurably lower end-to-end translation latency than 8 GB? Or is additional RAM beyond 8 GB unlikely to help if the processor is identical?

Separately, which specific Android device(s) would you recommend today for the lowest possible RU↔EN latency in Rtranslator?

Thanks!

[niedev]

Hi, as stated here, less than 8gb will deliver worse performance, but with 8gb or more the performance should be identical (but to be sure maybe 10gb or more is preferable). In the future however I will implement the option to use bigger models, and for those 12gb will probably be the minimum.
For the best latency I don't have a specific model to suggest, but I suggest one with the fastest possible CPU, and also efficient, because otherwise RTranslator will consume a lot of battery with long usages. On top of that I suggest to use a phone with a Qualcomm chip, because onnxruntime supports QNN and maybe in the future I will use this library to take advantage of the NPU or GPU of the chip. So the best fit would be a phone with a big enough battery, 12gb of RAM or more and a snapdragon 8 elite gen 4 or 5.

[Alexei-WLU]

Is Xiaomi Pad 8 Pro fully compatible?
Is Xiaomi brand in general compatible? I know it has some deviations from standard Android/Google

[niedev]

Yes, it is fully compatible