In shinra-dev/okcpuid: Interface to the cpuid library

library(Rfiglet)
options(scipen = 3)
knitr::opts_chunk$set(
  comment = NA,
  error   = FALSE,
  tidy    = FALSE)

library(Rcpuid)

font <- "standard"
header <- c(figlet("Introducing", font=font), figlet("Rcpuid", font=font))
class(header) <- "figlet"

header

Drew Schmidt

Rcpuid

libcpuid is a cross-platform C library for x86 CPU detection. The Rcpuid package is a set of R bindings (and a few minor improvements) for cpuid. Along with some helper utilities, the package offers three main functions: cpu_clock(), cpu_id(), and cpu_ins().

CPU Clock

The return is a list containing the total number of cores detected on the system (note: only cores available on the calling node are detected), the CPU clock in MHz as reported by the operating system (more on this later), the CPU clock in MHz as detected by a simple test, and the estimated peak FLOPS.

library(Rcpuid)

cpu_clock()

Why the two reports of CPU clock, you may be wondering. The reason is, all modern operating systems downscale CPU clocks when they're not very busy, and upscale when there's work to do. Ostensibly this is a power/energy-saving feature; in reality, these features tend to not work very well. Generally you can expect your system to underperform for almost no energy savings...but it makes people feel good, so here we are.

So you will probably see a huge discrepancy between the value reported by the OS and the value reported by the test. For example, when using the "ondemand" CPU governor (Linux), my OS reports that my clock is 830 MHz. But when I set the governor to performance, the OS reports 2159 MHz, which isn't too far from the tested value of 2166 MHz.

It's worth noting that given how the modern CPU works, this value is mostly (but not entirely) worthless; the above is just one demonstration.