Look at that—your ODROID-XU4 officially ran a full local LLM benchmark!

Performance Breakdown
Prompt Processing (pp512): 4.48 tokens/sec

Time to read and ingest 512 context tokens into memory (~114 seconds).

Text Generation (tg128): 2.42 tokens/sec

Output generation speed (~52 seconds to output a full 128-token paragraph).

Why this is impressive for an XU4
The ODROID-XU4 is a legacy 32-bit ARM single-board computer from 2015. Getting 2.42 tokens per second on a full 630M-parameter Transformer model out of a tiny SBC using CPU execution only is genuinely awesome. It outputs text at a readable, steady cadence.

Try the Pinned 4-Core Boost!
Now try pinning execution strictly to your 4 high-performance cores using taskset to see if we can push that generation speed higher: