Benchmarks
One honest number: our first Kria demo.
The RVPU is pre-demo. Rather than publish a comparison we have not run, we are publishing exactly one target — the pipeline we are bringing up first, on the hardware it will run on. A full comparison brief follows once the demo is measured on our own Kria hardware.
First published benchmark · the M5 demo
Depth Anything V2 monocular depth + stereo matching
AMD Kria KV260
Our first published benchmark is the M5 demo: a Depth Anything V2 monocular-depth plus stereo-matching pipeline running on an AMD Kria KV260 at real-time throughput (30+ Hz, ~33 ms per frame) within the KV260 power envelope. These are preliminary target figures pending the demo — we do not publish precise wattage, GOPS, or utilization numbers we have not measured.
- Throughput
- 30+ Hz
- Latency
- ~33 ms / frame
- Power
- Within KV260 envelope
Real-time · target
Per-frame · target
No measured wattage published yet
What this is, and isn't
Under-claiming, on purpose.
- 01This page is deliberately short and honest. The single figure above is a preliminary target for the pipeline we are bringing up first — not a measured, third-party-verified result. We will replace it with measured numbers once the demo lands.
- 02A full comparison brief — RVPU pipelines against alternative perception stacks — will follow after the demo is measured on our own hardware. We do not publish another vendor’s silicon performance, or multipliers against it, as our own.
- 03Design partners get earlier visibility: preliminary demo data and, as they complete, qualification reports are shared under NDA.
Want earlier visibility into the demo data?
Design partners get preliminary demo data and, as they complete, qualification reports under NDA. Tell us your platform and the pipelines you need.