6 TFLOPS at roughly 3 W. The dataflow core everything is built on.
At-memory compute, no external DRAM, deterministic latency.
12 to 100 TFLOPS across the module family.
Drop-in edge AI for Pi 5 builds.
Add inference to any existing host.
Ready-to-deploy boxes for production inference.
Rack-scale inference at data-center density.
Buy outright through our distributors.
Deploy on subscription, scale as you go.
Pay per inference, no hardware to manage.
Product of the Year Award — independently benchmarked as the most power-efficient inference accelerator in its class.
Manufacturing, automation, edge computing.
AI-powered surveillance and smart vision.
Automotive and mobility applications.
On-device intelligence for edge products.
Single-chip inference at the sensor.
Scale the same architecture to the rack.
SDK, model zoo and compiler toolchain.
Integrators and ecosystem partners.
The same dataflow core scales from a 3 W chip to a full rack — one toolchain, one model pipeline, no rewrites.
SDK, model zoo, compiler toolchain and the MemryX Developer Hub.