Speakers
Description
High Energy Physics experiments at the energy and intensity frontier face O(Tbps) data rates that must be processed with microsecond latency. To handle the large amount of data, a two-stage selection strategy is typically deployed to select and record the most interesting events in real time, where the first stage is based on custom electronics and the second one on a heterogeneous CPU/GPU computing farm. Recent advances in machine learning and high-speed networking enable a paradigm shift in data selection strategies. A-GHOST is an R&D project contributing towards this direction. It uses custom FPGA-based boards for data aggregation and ML-based feature extraction/compression to efficiently stream in a powerful GPU backend. The FPGA-to-GPU streaming happens via the NVIDIA DAQIRI software stack, which is optimised for high-throughput and low-latency streaming directly to the GPU memory. The project is currently in its prototype phase with the use of the NVIDIA IGX Thor platform, leveraging a 100 Gb/s UDP input stream generated in custom FPGA boards. This presentation will show the results obtained by deploying DAQIRI in a realistic throughput scenario and achieving stable inference throughput at 80 Gb/s.