search

karlrupp / microprocessor-trend-data

Data repository for my blog series on microprocessor trend data.
https://www.karlrupp.net/2018/02/42-years-of-microprocessor-trend-data/
Other
499 stars 79 forks source link

Recent single-core SPECint data misleading #4

Open codebent opened 4 years ago

codebent commented 4 years ago

I know that it's just the way things have progressed in the benchmarking game, but most of the "single-core" speed results use auto-parallelization and do not truly represent single-core performance. The last non autopar'd x86 results were submitted in 2009 and it topped out at 23.1. I would guess that without autopar current systems would be in the ~50 range. So not a 4x increase in performance over 10 years, but closer to 2.

karlrupp commented 4 years ago

@codebent thanks for pointing this out. Yes, SPECint is only an approximation to single-core performance and somewhat biased. However, it is a standardized metric that can be compared across a broad range of processors and many people are aware of its shortcomings.

In the end, it's a log-scale plot, so if we overestimate single-core performance by a factor of two, it doesn't change the overall picture :-)