turbostat - Report processor frequency and idle statistics turbostat [ Options ] command

turbostat [ Options ] [ "-i interval_sec" ]

turbostat reports processor topology, frequency and idle power state statistics on modern X86 processors. Either command is forked and statistics are printed upon its completion, or statistics are printed periodically. turbostat requires that the processor supports an "invariant" TSC, plus the APERF and MPERF MSRs. turbostat will report idle cpu power state residency on processors that additionally support C-state residency counters. The -p option limits output to the 1st thread in 1st core of each package.

The -P option limits output to the 1st thread in each Package.

The -S option limits output to a 1-line System Summary for each interval.

The -v option increases verbosity.

The -s option prints the SMI counter, equivalent to "-c 0x34"

The -c MSR# option includes the delta of the specified 32-bit MSR counter.

The -C MSR# option includes the delta of the specified 64-bit MSR counter.

The -m MSR# option includes the the specified 32-bit MSR value.

The -M MSR# option includes the the specified 64-bit MSR value.

The -i interval_sec option prints statistics every \fiinterval_sec seconds. The default is 5 seconds.

The command parameter forks command and upon its exit, displays the statistics gathered since it was forked.

pk processor package number.
cor processor core number.
CPU Linux CPU (logical processor) number.
Note that multiple CPUs per core indicate support for Intel(R) Hyper-Threading Technology.
%c0 percent of the interval that the CPU retired instructions.
GHz average clock rate while the CPU was in c0 state.
TSC average GHz that the TSC ran during the entire interval.
%c1, %c3, %c6, %c7 show the percentage residency in hardware core idle states.
%pc2, %pc3, %pc6, %pc7 percentage residency in hardware package idle states.

Without any parameters, turbostat prints out counters ever 5 seconds. (override interval with "-i sec" option, or specify a command for turbostat to fork). The first row of statistics is a summary for the entire system. Note that the summary is a weighted average. Subsequent rows show per-CPU statistics.

[root@x980]# ./turbostat
cor CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
 0.09 1.62 3.38 1.83 0.32 97.76 1.26 83.61
 0 0 0.15 1.62 3.38 10.23 0.05 89.56 1.26 83.61
 0 6 0.05 1.62 3.38 10.34
 1 2 0.03 1.62 3.38 0.07 0.05 99.86
 1 8 0.03 1.62 3.38 0.06
 2 4 0.21 1.62 3.38 0.10 1.49 98.21
 2 10 0.02 1.62 3.38 0.29
 8 1 0.04 1.62 3.38 0.04 0.08 99.84
 8 7 0.01 1.62 3.38 0.06
 9 3 0.53 1.62 3.38 0.10 0.20 99.17
 9 9 0.02 1.62 3.38 0.60
 10 5 0.01 1.62 3.38 0.02 0.04 99.92
 10 11 0.02 1.62 3.38 0.02

The "-s" option prints the column headers just once, and then the one line system summary for each sample interval.

[root@x980]# ./turbostat -s
 %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
 0.23 1.67 3.38 2.00 0.30 97.47 1.07 82.12
 0.10 1.62 3.38 1.87 2.25 95.77 12.02 72.60
 0.20 1.64 3.38 1.98 0.11 97.72 0.30 83.36
 0.11 1.70 3.38 1.86 1.81 96.22 9.71 74.90

The "-v" option adds verbosity to the output:

GenuineIntel 11 CPUID levels; family:model:stepping 0x6:2c:2 (6:44:2)
12 * 133 = 1600 MHz max efficiency
25 * 133 = 3333 MHz TSC frequency
26 * 133 = 3467 MHz max turbo 4 active cores
26 * 133 = 3467 MHz max turbo 3 active cores
27 * 133 = 3600 MHz max turbo 2 active cores
27 * 133 = 3600 MHz max turbo 1 active cores

The max efficiency frequency, a.k.a. Low Frequency Mode, is the frequency available at the minimum package voltage. The TSC frequency is the nominal maximum frequency of the processor if turbo-mode were not available. This frequency should be sustainable on all CPUs indefinitely, given nominal power and cooling. The remaining rows show what maximum turbo frequency is possible depending on the number of idle cores. Note that this information is not available on all processors. If turbostat is invoked with a command, it will fork that command and output the statistics gathered when the command exits. eg. Here a cycle soaker is run on 1 CPU (see %c0) for a few seconds until ^C while the other CPUs are mostly idle:

[root@x980 lenb]# ./turbostat cat /dev/zero > /dev/null
^C
cor CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
 8.86 3.61 3.38 15.06 31.19 44.89 0.00 0.00
 0 0 1.46 3.22 3.38 16.84 29.48 52.22 0.00 0.00
 0 6 0.21 3.06 3.38 18.09
 1 2 0.53 3.33 3.38 2.80 46.40 50.27
 1 8 0.89 3.47 3.38 2.44
 2 4 1.36 3.43 3.38 9.04 23.71 65.89
 2 10 0.18 2.86 3.38 10.22
 8 1 0.04 2.87 3.38 99.96 0.01 0.00
 8 7 99.72 3.63 3.38 0.27
 9 3 0.31 3.21 3.38 7.64 56.55 35.50
 9 9 0.08 2.95 3.38 7.88
 10 5 1.42 3.43 3.38 2.14 30.99 65.44
 10 11 0.16 2.88 3.38 3.40

Above the cycle soaker drives cpu7 up its 3.6 Ghz turbo limit while the other processors are generally in various states of idle. Note that cpu1 and cpu7 are HT siblings within core8. As cpu7 is very busy, it prevents its sibling, cpu1, from entering a c-state deeper than c1. Note that turbostat reports average GHz of 3.63, while the arithmetic average of the GHz column above is lower. This is a weighted average, where the weight is %c0. ie. it is the total number of un-halted cycles elapsed per time divided by the number of CPUs. On Intel Nehalem and newer processors, MSR 0x34 is a System Management Mode Interrupt (SMI) counter. Using the -m option, you can display how many SMIs have fired since reset, or if there are SMIs during the measurement interval, you can display the delta using the -d option.

[root@x980 ~]# turbostat -m 0x34
cor CPU %c0 GHz TSC MSR 0x034 %c1 %c3 %c6 %pc3 %pc6
 1.41 1.82 3.38 0x00000000 8.92 37.82 51.85 17.37 0.55
 0 0 3.73 2.03 3.38 0x00000055 1.72 48.25 46.31 17.38 0.55
 0 6 0.14 1.63 3.38 0x00000056 5.30
 1 2 2.51 1.80 3.38 0x00000056 15.65 29.33 52.52
 1 8 0.10 1.65 3.38 0x00000056 18.05
 2 4 1.16 1.68 3.38 0x00000056 5.87 24.47 68.50
 2 10 0.10 1.63 3.38 0x00000056 6.93
 8 1 3.84 1.91 3.38 0x00000056 1.36 50.65 44.16
 8 7 0.08 1.64 3.38 0x00000056 5.12
 9 3 1.82 1.73 3.38 0x00000056 7.59 24.21 66.38
 9 9 0.09 1.68 3.38 0x00000056 9.32
 10 5 1.66 1.65 3.38 0x00000056 15.10 50.00 33.23
 10 11 1.72 1.65 3.38 0x00000056 15.05
^C
[root@x980 ~]# 

"turbostat " must be run as root. "turbostat " reads hardware counters, but doesn't write them. So it will not interfere with the OS or other programs, including multiple invocations of itself. turbostat may work poorly on Linux-2.6.20 through 2.6.29, as acpi-cpufreq periodically cleared the APERF and MPERF in those kernels. If the TSC column does not make sense, then the other numbers will also make no sense. Turbostat is lightweight, and its data collection is not atomic. These issues are usually caused by an extremely short measurement interval (much less than 1 second), or system activity that prevents turbostat from being able to run on all CPUS to quickly collect data. The APERF, MPERF MSRs are defined to count non-halted cycles. Although it is not guaranteed by the architecture, turbostat assumes that they count at TSC rate, which is true on all processors tested to date. "Intel® Turbo Boost Technology in Intel® Core™ Microarchitecture (Nehalem) Based Processors" http://download.intel.com/design/processor/applnots/320354.pdf "Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide" http://www.intel.com/products/processor/manuals/

/dev/cpu/*/msr

msr(4), vmstat(8)

Written by Len Brown <len.brown@intel.com>