Power & Backend
Jetson AGX Orin · RTX 5090 rig · ai.priontific.co.uk
Backend
Serving now:
—
turbo
Switching pulls whoever is mid-session onto the other
model, so a switch away from a backend used in the last 2 minutes asks
first.
Orin
…
Power draw—
Shim—
Last request—
Updated—
Boot to serving takes about 2–3 minutes.
Rig (RTX 5090)
…
Power draw—
vLLM—
Last request—
Boot to serving takes about 3 minutes. The rig is shared
with Maddie’s work and puts itself to sleep when idle.