[General]: No faster when choosing GPU on Samsung Galaxy Tab s8

[fancellu]

Say I choose Gemmasutra, and ask it a question, comes out at 1.60 tokens a second in cpu mode

But when I go into settings and choose gpu, it is no faster

Should it be?

It is seen as an Adreno TM 730 and says it is supported

I was wondering if I was doing something wrong. Would be nice if the chat made it clear what it was using.

[a-ghorbani]

GPUs typically accelerate token processing rather than token generation. As a result, they are advantageous for the initial processing of long prompts, like summarization tasks.

Also, you need to use GPU-compatible quantizations, such as Q4_0, Q8_0, mxfp4.

The following benchmarks are on a Samsung Galaxy S25/Adreno 830.
As you can see the prompt generation is more or less the same. But prompt process is way faster.

CPU Only	GPU (OpenCL)

[81820N91108]

Suggestion: Add a "Memory Link" toggle icon nextto the inputbox. Green Link: Cross-chat(Privacy/Performance mode).