TGG BenchSEE v1.2.3 Comments - Part II

12 November, 2021 | Report

This paper will continue BenchSEE architectural assessment focusing on power analyzers averaging interval (TGG27), and BenchSEE scoring assessment to address DRAM frequency scaling (part of TGG6), OS, core count, socket count and node count scaling. Rank analysis was conducted on a limited number of server systems run on BenchSEE and SERT tools. Servers with one installed processor shows ranking inversion (Figure 6) between BenchSEE and SERT scores. This could be attributed to the known 1P issues (TGG7). Similar paired analysis was conducted on 2P server systems which shows good correlation on ranking analysis except for two paired configurations. While this could be tied to TGG1 and TGG2 issues, this needs to be further understood (TGG28). Similar anomaly was observed on ranking analysis within a product family consisting of low end and high end configurations. BenchSEE efficiency score was higher on low end configuration than high end configuration, an inversion that needs further investigation (TGG29).

Topics: BenchSEE, SERT, SPEC