Submitter: | Steven van der Vlugt |
Description: | October 10th AMD held it's Advancing AI 2024 Event in San Francisco. We were invite to join the event because we participated in the AMD Pervasive AI Hackster competition. During the event we heard that we landed second place with our project TINA in the PC AI category. TINA is an acronym for "This Is Not AI". With our TINA framework we enable the implementation of non Neural Network (NN) signal processing algorithms on NN accelerators such as GPUs, TPUs or FPGAs. In this project we demonstrated that we can easily port an already existing GPU implementation of a Polyphase Filter Bank to the Ryzen AI engines (NPU/XDNA). TINA is a collaboration between TU Delft (Christiaan Boerkamp and Zaid Al-Ars) and ASTRON (Steven van der Vlugt). The Ryzen AI PC that we could use in this competition allowed us to test our implementation on the Ryzen AI engines. This prize allows us to further explore the potential of the Ryzen AI engines. The technology developed in project TINA will allow us to leverage special purpose AI/ML/LLM hardware for acceleration of radio astronomical software. The picture shows Christiaan receiving the prize at the AMD AI event and meeting Dr. Lisa Su. Thank you AMD for the invitation and this amazing competition where we could share about our project! The Hackster project: https://www.hackster.io/tina/tina-running-non-nn-algorithmns-on-an-amd-ryzen-npu-0cc58c TINA Publication with MLSP2024: https://arxiv.org/abs/2408.16551 TINA with AMD FINN (FPL2024 tutorial): https://github.com/Xilinx/finn/blob/github-pages/docs/finn-fpl24/FPL24_poster_TINA.pdf |
Copyright: | Christiaan Boerkamp |
Tweet | ![]() |