NVIDIA Deep Learning Accelerator IP now released at NVDLA.org

The NVIDIA Deep Learning Accelerator (NVDLA) is a free and open architecture that promotes a standard way to design deep learning inference accelerators. Today, the project is live at NVDLA.org with the RTL sources hosted on GitHub.

Download the packages today from NVDLA.org and get started designing your own smart IoT or SoC devices.

Making a cool device or platform integrated with NVDLA? Share with others and let us know about it!

Is there a forum for NVDLA that we can ask questions?

Hi ellie_s, please post your questions/issues to the NVDLA GitHub repo’s Issue tracker tab, thank you.

Thank you, dusty.
My question is more on how a module works, not the code itself.
For example, I want some clarification on RUBIK and ACC. Specifically, I would like to know if RUBIK masks some of rams if number of channels are smaller than 32, and combines the results of one clock cycle to result of next.

Is this type of question appropriate for github repo?

Sure thing, any questions about NVDLA you have, feel free to post them there. At this time there hasn’t been a general forum board created for it here.