面向视频解析任务的异构GPU 性能建模与中立评估框架研究

© 2026 by the Author(s). Licensee Art and Technology, USA. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution -Noncommercial 4.0 International License (CC BY-NC 4.0) ( https://creativecommons.org/licenses/by-nc/4.0/ )

Download PDF

Cite

XML

HTML

Abstract

视频结构化解析是智能安防系统的核心技术，但面临多源异构硬件选型难题。针对当前评估中" 硬件参数导向" 的局限性，本文提出一种面向视频解析任务的异构GPU 中立评估框架。通过分析GDDR 与HBM 显存架构的技术差异，构建包含有效解码并发度、推理吞吐量、带宽- 计算平衡性、跨框架适配成本、能耗效率的五维评估模型。基于该框架对NVIDIA Ampere、AMD CDNA2、华为昇腾等架构进行建模与验证，发现传统显存位宽指标存在架构依赖性偏差。提出的任务- 架构匹配度指标（TAM）为多品牌GPU 选型提供了方法论参考。

Keywords

异构计算

视频结构化解析

GPU 性能建模

中立评估框架

References

[1] 王坤峰, 苟超, 王飞跃. 平行视觉: 基于ACP 的智能视觉计算方法[J]. 自动化学报, 2016, 42(10): 1490-1500.
[2]Abdelkhalik H, Arafa Y, Santhi N, et al. Demystifying the Nvidia Ampere Architecture through Microbenchmarking and Instruction-level Analysis[C]. IEEE HPEC, 2022. arXiv:2208.11174.
[3]NVIDIA Corporation. NVIDIA A100 Tensor Core GPU Datasheet[EB/OL]. 2021.
[4]NVIDIA Corporation. NVIDIA A40 GPU Datasheet[EB/OL]. 2021.
[5]AMD Corporation. AMD Instinct MI210 Accelerator Specifications[EB/OL]. 2021.
[6]Williams S, Waterman A, Patterson D. Roofline: An insightful visual performance model for multicore architectures[J]. Communications of the ACM, 2009, 52(4): 65-76.
[7] 华为技术有限公司. 昇腾910 AI 处理器技术白皮书[R]. 2021.
[8]MLCommons. MLPerf Inference v2.1 Results[EB/OL]. https://mlcommons.org/benchmarks/inference-datacenter/, 2022.
[9]NVIDIA Corporation. NVIDIA A100 Tensor Core GPU Architecture[EB/OL]. 2020.

Previous article in this issue

Next article in this issue