For questions, a demo or a trial license, contact us.
Including the ones that do not flatter us. What each set of results does and does not show is in each model’s benchmark report.
Models we haven't ported yet. Our team ports every model into NukeTorch today, so tell us which one you need.
No. NukeTorch's Tensor Engine is new, built from the ground up on our own mathematical principles. We pursue one question: what are numbers? We designed its architecture ourselves, without studying other engines.
NukeTorch is its own Tensor Engine, built on a different formulation of the tensor. Its gains come from that design and don't fit inside another engine.
Every model has a benchmark and a reproducible path. We run private demos, and downloads are available by permission after an NDA is signed.
We license NukeTorch commercially to selected partners. Tell us what you are building, and we’ll find the right fit.
Our published benchmarks were done on Apple silicon. We will release additional hardware support.
The model stays the same. For optimization and every last bit of performance, we can work with you.
We publish the mathematics: see Publication for our papers on the formulation. The engine code itself is not published.