UVMBench
收藏资源简介:
UVMBench是由俄勒冈州立大学电气工程与计算机科学学院开发的综合性UVM基准套件,包含32个来自不同应用领域的代表性基准。该数据集旨在通过统一的编程实现和多样化的内存访问模式,帮助研究人员全面评估和比较当前最先进的UVM技术。UVMBench特别增加了与机器学习相关的负载,以探索UVM在数据密集型机器学习应用中的应用潜力。数据集的创建过程涉及重新实现现有基准,并开发新的机器学习负载。UVMBench的应用领域主要集中在GPU的UVM研究,旨在解决UVM引入的性能开销问题,并探索在内存超额订阅情况下的性能表现。
UVMBench is a comprehensive UVM benchmark suite developed by the College of Engineering and Computer Science, Oregon State University, which consists of 32 representative benchmarks from diverse application domains. This dataset aims to help researchers comprehensively evaluate and compare state-of-the-art UVM technologies via unified programming implementations and diversified memory access patterns. UVMBench specifically incorporates machine learning-related workloads to explore the application potential of UVM in data-intensive machine learning applications. The creation of this dataset involves re-implementing existing benchmarks and developing new machine learning workloads. UVMBench is primarily targeted at GPU-oriented UVM research, with the goals of addressing the performance overhead introduced by UVM and investigating performance characteristics under memory oversubscription scenarios.




