Category :
Others
Date :
2026-09-01 (Tuesday)
Nature :
External
The IEEE AICAS 2026 Grand Challenge (8th IEEE International Conference on Artificial Intelligence Circuits and Systems Grand Challenge) is a premier global competition focused on pushing the boundaries of AI chip architecture, hardware-software co-design, and embedded systems. The challenge requires teams to deploy complex AI workloads efficiently within strict hardware constraints.
Competing against top talents from around the world, Yeung Yeung from Class 5C demonstrated outstanding technical execution and high-performance system design. By developing a systematic optimization framework—specifically targeting Vision-Language Models (VLMs) like Qwen3-VL-2B-Instruct—the team overcame the kernel launch bottleneck using operator fusion and CUDA Graphs. Their innovative approach achieved an incredible 11× speedup over the competition's baseline, enabling real-time processing capabilities of about 300 words per second without batch processing.
For these remarkable accomplishments, Yeung Yeung successfully took home the global First Runner-Up award, with their research work also accepted for publication as a conference paper titled "Overcoming the Kernel Launch Bottleneck: 11× Faster VLM Inference with Operator Fusion and CUDA Graphs."