High-Throughput Parallel Viterbi Decoder on GPU Tensor Cores

TL;DR

This paper introduces a novel parallel Viterbi decoding algorithm optimized for GPU Tensor Cores, achieving significant throughput improvements over previous GPU-based implementations.

Contribution

It presents a new parallel implementation of Viterbi decoding leveraging Tensor Cores, enhancing performance on modern GPUs.

Findings

Significant throughput improvements over prior GPU implementations

Efficient utilization of Tensor Cores for decoding algorithms

Demonstrated performance gains through experimental evaluation

Abstract

Many research works have been performed on implementation of Vitrerbi decoding algorithm on GPU instead of FPGA because this platform provides considerable flexibility in addition to great performance. Recently, the recently-introduced Tensor cores in modern GPU architectures provide incredible computing capability. This paper proposes a novel parallel implementation of Viterbi decoding algorithm based on Tensor cores in modern GPU architectures. The proposed parallel algorithm is optimized to efficiently utilize the computing power of Tensor cores. Experiments show considerable throughput improvements in comparison with previous works.