Arena
The arena is empty. No runs have been published yet.
This page describes the shape of what will land here: what a run is, what it will measure, and the first device it will measure it on. None of it is a result — it is the method, published before any number exists.
One device, one setup, measured whole
A run pairs one specific Jetson device with one setup — a single model, a mixture of models, or an ordered pipeline, under a pinned config — and measures that pairing as a single system, not as isolated parts. Every run will ship with the Docker recipe that reproduces it: the exact containers, models, and flags. Own the device, rerun the arena.
What every run will carry
Latency
End to end, and per stage
The wall-clock story of one pass through the pipeline, and where inside it the time goes. Every run will carry the full breakdown, not a single headline figure.
Memory signature
On unified memory
CPU and GPU share one pool on a Jetson, so the setup's whole footprint is what matters. Runs will record what each stage holds resident, and what remains for everything else on the board.
Quality
Honest, per stage
A pipeline that answers fast but gets a stage wrong has not won anything. Each stage will be scored against its own task, and every trade-off made to fit the device will be stated with the run.
Jetson Thor
A realtime voice loop, running whole on one board: speak, and hear the reply. It is the first setup the arena will publish a run for.
- VADvoice activity detection
- STTspeech to text
- LLMlanguage model
- TTStext to speech
Four stages contending for one pool of unified memory, in a loop a person can feel. None of it is measured yet — this is only where the result will go once it is. Next in line: Jetson AGX Orin on JetPack 7.2, then Jetson Orin Nano Super 8GB.