Scholarlink

<p>Efficient use of resources of FPGA-based system-on-modules (SoMs) is critical for deploying deep neural networks at the edge. This work quantifies the impact of software multithreading on the AMD Kria KV260, built around a Zynq UltraScale+ MPSoC with a Quad-Core Cortex-A53 and a DPU accelerator, on an image classification task. Three image classification models (MobileNetV2, ResNet-50, and SqueezeNet) were benchmarked under identical conditions, while varying the number of threads for each test. Each thread drives an independent Vitis-AI runner instance. The accuracies of the floating point and quantized models were recorded on a host PC, and the KV260 inference throughput was evaluated on a subset of 500 images from the ImageNet dataset. Thread concurrency delivered a throughput gain of approximately 3.1 <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> to 3.67 <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> across the three models, up to an optimal threshold of four threads without degrading the models’ Top-1 accuracy. Results provide board-specific evidence that lightweight software multithreading can unlock a significant portion of the KV260 performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Throughput impact of software multithreading for deep-learning inference on the AMD Kria KV260

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月19日

查看全文
  • Claudino Costa,
  • José Henrique Brito

摘要

Efficient use of resources of FPGA-based system-on-modules (SoMs) is critical for deploying deep neural networks at the edge. This work quantifies the impact of software multithreading on the AMD Kria KV260, built around a Zynq UltraScale+ MPSoC with a Quad-Core Cortex-A53 and a DPU accelerator, on an image classification task. Three image classification models (MobileNetV2, ResNet-50, and SqueezeNet) were benchmarked under identical conditions, while varying the number of threads for each test. Each thread drives an independent Vitis-AI runner instance. The accuracies of the floating point and quantized models were recorded on a host PC, and the KV260 inference throughput was evaluated on a subset of 500 images from the ImageNet dataset. Thread concurrency delivered a throughput gain of approximately 3.1 \(\times \) × to 3.67 \(\times \) × across the three models, up to an optimal threshold of four threads without degrading the models’ Top-1 accuracy. Results provide board-specific evidence that lightweight software multithreading can unlock a significant portion of the KV260 performance.