Our team of engineers and scientists create the most efficient neural network inference compute engine and package it with single command line ease of use.

They crafted Zebra to deliver performance and accuracy out of the box, with ease of use for all image based Machine Learning applications.

What you can expect when using Zebra to deploy your neural networks in the real world:

• Use the exact same neural network that is already trained by your scientist

• No new tools or knowledge, no changes whatsoever

• Select your FPGA-based card, plug it and forget the hardware, execute your inference on Zebra with a single command

• Get high performance or low latency without long profiling or having to understand the hardware

• Keep the accuracy while Zebra transparently quantizes the neural network for you – don’t need scientists to retrain

• Keep up with the latest Machine Learning progress as Zebra gets refreshed regularly to accelerate what specialized ASICs could not plan for a few years back

• Prefer to use Zebra for embedded application? Zebra comes also as an IP and the same easy-to-use tools

• Run for years thanks to the long lifespan of FPGAs, with high stability and lower Total cost of ownership

• Always higher performance with new generations of AI-oriented FPGAs every year and zero adoption effort thanks to Zebra