🤖 AI Summary
Beamforming in multi-sensor signal processing suffers from high computational density, poor hardware adaptability, and insufficient flexibility in precision. Method: This paper introduces the first tensor-core-oriented general-purpose beamforming acceleration library. It pioneers the generalization of GPU tensor cores for beamforming computation, integrates mixed-precision (FP16/1-bit) design, and achieves cross-platform high-efficiency deployment on both NVIDIA and AMD GPUs via dual-stack heterogeneous optimization using CUDA and HIP. Contributions/Results: The library achieves over 600 TeraOps/s (FP16) on AMD MI300X with near 1 TeraOp/J energy efficiency; on NVIDIA A100, it delivers 3 PetaOps/s in 1-bit mode with >10 TeraOps/J efficiency. It enables, for the first time, ultra-low-precision real-time beamforming and has been successfully deployed in clinical ultrasound and radio astronomy systems.
📝 Abstract
Beamforming is a well-known technique to combine signals from multiple sensors. It has a wide range of application domains. This paper introduces the Tensor-Core Beamformer: a generic, optimized beamformer library that harnesses the computational power of GPU tensor cores to accelerate beamforming computations. The library hides the complexity of tensor cores from the user, and supports 16-bit and 1-bit precision. An extensive performance evaluation on NVIDIA and AMD GPUs shows that the library outperforms traditional beamforming on regular GPU cores by a wide margin, at much higher energy efficiency. In the 16-bit mode, it achieves over 600 TeraOps/s on an AMD MI300X GPU, while approaching 1 TeraOp/J. In the 1-bit mode, it breaks the 3 PetaOps/s barrier and achieves over 10 TeraOps/J on an NVIDIA A100 GPU. The beamforming library can be easily integrated into existing pipelines. We demonstrate its use for medical ultrasound and radio-astronomical instruments.