3 <a name="disabling-cpu-frequency-scaling" />
5 ## Disabling CPU Frequency Scaling
10 ***WARNING*** CPU scaling is enabled, the benchmark real time measurements may be noisy and will incur extra overhead.
13 you might want to disable the CPU frequency scaling while running the
14 benchmark, as well as consider other ways to stabilize the performance of
15 your system while benchmarking.
17 Exactly how to do this depends on the Linux distribution,
18 desktop environment, and installed programs. Specific details are a moving
19 target, so we will not attempt to exhaustively document them here.
21 One simple option is to use the `cpupower` program to change the
22 performance governor to "performance". This tool is maintained along with
23 the Linux kernel and provided by your distribution.
25 It must be run as root, like this:
28 sudo cpupower frequency-set --governor performance
31 After this you can verify that all CPUs are using the performance governor
32 by running this command:
35 cpupower frequency-info -o proc
38 The benchmarks you subsequently run will have less variance.
40 <a name="reducing-variance" />
42 ## Reducing Variance in Benchmarks
44 The Linux CPU frequency governor [discussed
45 above](user_guide#disabling-cpu-frequency-scaling) is not the only source
46 of noise in benchmarks. Some, but not all, of the sources of variance
49 1. On multi-core machines not all CPUs/CPU cores/CPU threads run the same
50 speed, so running a benchmark one time and then again may give a
51 different result depending on which CPU it ran on.
52 2. CPU scaling features that run on the CPU, like Intel's Turbo Boost and
53 AMD Turbo Core and Precision Boost, can temporarily change the CPU
54 frequency even when the using the "performance" governor on Linux.
55 3. Context switching between CPUs, or scheduling competition on the CPU the
56 benchmark is running on.
57 4. Intel Hyperthreading or AMD SMT causing the same issue as above.
58 5. Cache effects caused by code running on other CPUs.
59 6. Non-uniform memory architectures (NUMA).
61 These can cause variance in benchmarks results within a single run
62 (`--benchmark_repetitions=N`) or across multiple runs of the benchmark
65 Reducing sources of variance is OS and architecture dependent, which is one
66 reason some companies maintain machines dedicated to performance testing.
68 Some of the easier and effective ways of reducing variance on a typical
69 Linux workstation are:
71 1. Use the performance governor as [discussed
72 above](user_guide#disabling-cpu-frequency-scaling).
73 1. Disable processor boosting by:
75 echo 0 | sudo tee /sys/devices/system/cpu/cpufreq/boost
77 See the Linux kernel's
78 [boost.txt](https://www.kernel.org/doc/Documentation/cpu-freq/boost.txt)
80 2. Set the benchmark program's task affinity to a fixed cpu. For example:
82 taskset -c 0 ./mybenchmark
84 3. Disabling Hyperthreading/SMT. This can be done in the Bios or using the
85 `/sys` file system (see the LLVM project's [Benchmarking
86 tips](https://llvm.org/docs/Benchmarking.html)).
87 4. Close other programs that do non-trivial things based on timers, such as
88 your web browser, desktop environment, etc.
89 5. Reduce the working set of your benchmark to fit within the L1 cache, but
90 do be aware that this may lead you to optimize for an unrealistic
93 Further resources on this topic:
95 1. The LLVM project's [Benchmarking
96 tips](https://llvm.org/docs/Benchmarking.html).
97 1. The Arch Wiki [Cpu frequency
98 scaling](https://wiki.archlinux.org/title/CPU_frequency_scaling) page.