CUTLASS GEMM Profiling Data
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/pavlyhalim/GPPerf
下载链接
链接失效反馈官方服务：
资源简介：
该数据集汇总了从不同配置的CUTLASS SGEMM内核分析中收集的性能指标数据。这些指标包括运行时间、吞吐量、功耗、能耗、内存利用率和GPU利用率，这些数据是通过CUTLASS分析器和NCU工具收集的。数据涵盖了各种矩阵维度、布局、块大小以及alpha-beta标量的配置。该数据集的任务是对GEMM操作的性能进行基准测试和分析。
This dataset aggregates performance metric data collected from the analysis of CUTLASS SGEMM kernels with different configurations. The metrics include runtime, throughput, power consumption, energy consumption, memory utilization, and GPU utilization, which were collected using the CUTLASS analyzer and the NCU tool. The dataset covers various configurations of matrix dimensions, layouts, block sizes, and alpha-beta scalars. The task of this dataset is to benchmark and analyze the performance of GEMM operations.
提供机构：
NVIDIA
