turbostat.8 (20296B)
1.TH TURBOSTAT 8 2.SH NAME 3turbostat \- Report processor frequency and idle statistics 4.SH SYNOPSIS 5.ft B 6.B turbostat 7.RB [ Options ] 8.RB command 9.br 10.B turbostat 11.RB [ Options ] 12.RB [ "\--interval seconds" ] 13.SH DESCRIPTION 14\fBturbostat \fP reports processor topology, frequency, 15idle power-state statistics, temperature and power on X86 processors. 16There are two ways to invoke turbostat. 17The first method is to supply a 18\fBcommand\fP, which is forked and statistics are printed 19in one-shot upon its completion. 20The second method is to omit the command, 21and turbostat displays statistics every 5 seconds interval. 22The 5-second interval can be changed using the --interval option. 23.PP 24Some information is not available on older processors. 25.SS Options 26Options can be specified with a single or double '-', and only as much of the option 27name as necessary to disambiguate it from others is necessary. Note that options are case-sensitive. 28.PP 29\fB--add attributes\fP add column with counter having specified 'attributes'. The 'location' attribute is required, all others are optional. 30.nf 31 location: {\fBmsrDDD\fP | \fBmsr0xXXX\fP | \fB/sys/path...\fP} 32 msrDDD is a decimal offset, eg. msr16 33 msr0xXXX is a hex offset, eg. msr0x10 34 /sys/path... is an absolute path to a sysfs attribute 35 36 scope: {\fBcpu\fP | \fBcore\fP | \fBpackage\fP} 37 sample and print the counter for every cpu, core, or package. 38 default: cpu 39 40 size: {\fBu32\fP | \fBu64\fP } 41 MSRs are read as 64-bits, u32 truncates the displayed value to 32-bits. 42 default: u64 43 44 format: {\fBraw\fP | \fBdelta\fP | \fBpercent\fP} 45 'raw' shows the MSR contents in hex. 46 'delta' shows the difference in values during the measurement interval. 47 'percent' shows the delta as a percentage of the cycles elapsed. 48 default: delta 49 50 name: "name_string" 51 Any string that does not match a key-word above is used 52 as the column header. 53.fi 54.PP 55\fB--cpu cpu-set\fP limit output to system summary plus the specified cpu-set. If cpu-set is the string "core", then the system summary plus the first CPU in each core are printed -- eg. subsequent HT siblings are not printed. Or if cpu-set is the string "package", then the system summary plus the first CPU in each package is printed. Otherwise, the system summary plus the specified set of CPUs are printed. The cpu-set is ordered from low to high, comma delimited with ".." and "-" permitted to denote a range. eg. 1,2,8,14..17,21-44 56.PP 57\fB--hide column\fP do not show the specified built-in columns. May be invoked multiple times, or with a comma-separated list of column names. 58.PP 59\fB--enable column\fP show the specified built-in columns, which are otherwise disabled, by default. Currently the only built-in counters disabled by default are "usec", "Time_Of_Day_Seconds", "APIC" and "X2APIC". 60The column name "all" can be used to enable all disabled-by-default built-in counters. 61.PP 62\fB--show column\fP show only the specified built-in columns. May be invoked multiple times, or with a comma-separated list of column names. 63.PP 64\fB--show CATEGORY --hide CATEGORY\fP Show and hide also accept a single CATEGORY of columns: "all", "topology", "idle", "frequency", "power", "sysfs", "other". 65.PP 66\fB--Dump\fP displays the raw counter values. 67.PP 68\fB--quiet\fP Do not decode and print the system configuration header information. 69.PP 70\fB--interval seconds\fP overrides the default 5.0 second measurement interval. 71.PP 72\fB--num_iterations num\fP number of the measurement iterations. 73.PP 74\fB--out output_file\fP turbostat output is written to the specified output_file. 75The file is truncated if it already exists, and it is created if it does not exist. 76.PP 77\fB--help\fP displays usage for the most common parameters. 78.PP 79\fB--Joules\fP displays energy in Joules, rather than dividing Joules by time to print power in Watts. 80.PP 81\fB--list\fP display column header names available for use by --show and --hide, then exit. 82.PP 83\fB--Summary\fP limits output to a 1-line System Summary for each interval. 84.PP 85\fB--TCC temperature\fP sets the Thermal Control Circuit temperature for systems which do not export that value. This is used for making sense of the Digital Thermal Sensor outputs, as they return degrees Celsius below the TCC activation temperature. 86.PP 87\fB--version\fP displays the version. 88.PP 89The \fBcommand\fP parameter forks \fBcommand\fP, and upon its exit, 90displays the statistics gathered since it was forked. 91.PP 92.SH ROW DESCRIPTIONS 93The system configuration dump (if --quiet is not used) is followed by statistics. The first row of the statistics labels the content of each column (below). The second row of statistics is the system summary line. The system summary line has a '-' in the columns for the Package, Core, and CPU. The contents of the system summary line depends on the type of column. Columns that count items (eg. IRQ) show the sum across all CPUs in the system. Columns that show a percentage show the average across all CPUs in the system. Columns that dump raw MSR values simply show 0 in the summary. After the system summary row, each row describes a specific Package/Core/CPU. Note that if the --cpu parameter is used to limit which specific CPUs are displayed, turbostat will still collect statistics for all CPUs in the system and will still show the system summary for all CPUs in the system. 94.SH COLUMN DESCRIPTIONS 95.nf 96\fBusec\fP For each CPU, the number of microseconds elapsed during counter collection, including thread migration -- if any. This counter is disabled by default, and is enabled with "--enable usec", or --debug. On the summary row, usec refers to the total elapsed time to collect the counters on all cpus. 97\fBTime_Of_Day_Seconds\fP For each CPU, the gettimeofday(2) value (seconds.subsec since Epoch) when the counters ending the measurement interval were collected. This column is disabled by default, and can be enabled with "--enable Time_Of_Day_Seconds" or "--debug". On the summary row, Time_Of_Day_Seconds refers to the timestamp following collection of counters on the last CPU. 98\fBCore\fP processor core number. Note that multiple CPUs per core indicate support for Intel(R) Hyper-Threading Technology (HT). 99\fBCPU\fP Linux CPU (logical processor) number. Yes, it is okay that on many systems the CPUs are not listed in numerical order -- for efficiency reasons, turbostat runs in topology order, so HT siblings appear together. 100\fBPackage\fP processor package number -- not present on systems with a single processor package. 101\fBAvg_MHz\fP number of cycles executed divided by time elapsed. Note that this includes idle-time when 0 instructions are executed. 102\fBBusy%\fP percent of the measurement interval that the CPU executes instructions, aka. % of time in "C0" state. 103\fBBzy_MHz\fP average clock rate while the CPU was not idle (ie. in "c0" state). 104\fBTSC_MHz\fP average MHz that the TSC ran during the entire interval. 105\fBIRQ\fP The number of interrupts serviced by that CPU during the measurement interval. The system total line is the sum of interrupts serviced across all CPUs. turbostat parses /proc/interrupts to generate this summary. 106\fBSMI\fP The number of System Management Interrupts serviced CPU during the measurement interval. While this counter is actually per-CPU, SMI are triggered on all processors, so the number should be the same for all CPUs. 107\fBC1, C2, C3...\fP The number times Linux requested the C1, C2, C3 idle state during the measurement interval. The system summary line shows the sum for all CPUs. These are C-state names as exported in /sys/devices/system/cpu/cpu*/cpuidle/state*/name. While their names are generic, their attributes are processor specific. They the system description section of output shows what MWAIT sub-states they are mapped to on each system. 108\fBC1%, C2%, C3%\fP The residency percentage that Linux requested C1, C2, C3.... The system summary is the average of all CPUs in the system. Note that these are software, reflecting what was requested. The hardware counters reflect what was actually achieved. 109\fBCPU%c1, CPU%c3, CPU%c6, CPU%c7\fP show the percentage residency in hardware core idle states. These numbers are from hardware residency counters. 110\fBCoreTmp\fP Degrees Celsius reported by the per-core Digital Thermal Sensor. 111\fBPkgTmp\fP Degrees Celsius reported by the per-package Package Thermal Monitor. 112\fBGFX%rc6\fP The percentage of time the GPU is in the "render C6" state, rc6, during the measurement interval. From /sys/class/drm/card0/power/rc6_residency_ms. 113\fBGFXMHz\fP Instantaneous snapshot of what sysfs presents at the end of the measurement interval. From /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz. 114\fBPkg%pc2, Pkg%pc3, Pkg%pc6, Pkg%pc7\fP percentage residency in hardware package idle states. These numbers are from hardware residency counters. 115\fBPkgWatt\fP Watts consumed by the whole package. 116\fBCorWatt\fP Watts consumed by the core part of the package. 117\fBGFXWatt\fP Watts consumed by the Graphics part of the package -- available only on client processors. 118\fBRAMWatt\fP Watts consumed by the DRAM DIMMS -- available only on server processors. 119\fBPKG_%\fP percent of the interval that RAPL throttling was active on the Package. Note that the system summary is the sum of the package throttling time, and thus may be higher than 100% on a multi-package system. Note that the meaning of this field is model specific. For example, some hardware increments this counter when RAPL responds to thermal limits, but does not increment this counter when RAPL responds to power limits. Comparing PkgWatt and PkgTmp to system limits is necessary. 120\fBRAM_%\fP percent of the interval that RAPL throttling was active on DRAM. 121.fi 122.SH TOO MUCH INFORMATION EXAMPLE 123By default, turbostat dumps all possible information -- a system configuration header, followed by columns for all counters. 124This is ideal for remote debugging, use the "--out" option to save everything to a text file, and get that file to the expert helping you debug. 125.PP 126When you are not interested in all that information, and there are several ways to see only what you want. First the "--quiet" option will skip the configuration information, and turbostat will show only the counter columns. Second, you can reduce the columns with the "--hide" and "--show" options. If you use the "--show" option, then turbostat will show only the columns you list. If you use the "--hide" option, turbostat will show all columns, except the ones you list. 127.PP 128To find out what columns are available for --show and --hide, the "--list" option is available. For convenience, the special strings "sysfs" can be used to refer to all of the sysfs C-state counters at once: 129.nf 130sudo ./turbostat --show sysfs --quiet sleep 10 13110.003837 sec 132 C1 C1E C3 C6 C7s C1% C1E% C3% C6% C7s% 133 4 21 2 2 459 0.14 0.82 0.00 0.00 98.93 134 1 17 2 2 130 0.00 0.02 0.00 0.00 99.80 135 0 0 0 0 31 0.00 0.00 0.00 0.00 99.95 136 2 1 0 0 52 1.14 6.49 0.00 0.00 92.21 137 1 2 0 0 52 0.00 0.08 0.00 0.00 99.86 138 0 0 0 0 71 0.00 0.00 0.00 0.00 99.89 139 0 0 0 0 25 0.00 0.00 0.00 0.00 99.96 140 0 0 0 0 74 0.00 0.00 0.00 0.00 99.94 141 0 1 0 0 24 0.00 0.00 0.00 0.00 99.84 142.fi 143.PP 144.SH ONE SHOT COMMAND EXAMPLE 145If turbostat is invoked with a command, it will fork that command 146and output the statistics gathered after the command exits. 147In this case, turbostat output goes to stderr, by default. 148Output can instead be saved to a file using the --out option. 149In this example, the "sleep 10" command is forked, and turbostat waits for it to complete before saving all statistics into "ts.out". Note that "sleep 10" is not part of turbostat, but is simply an example of a command that turbostat can fork. The "ts.out" file is what you want to edit in a very wide window, paste into a spreadsheet, or attach to a bugzilla entry. 150 151.nf 152[root@hsw]# ./turbostat -o ts.out sleep 10 153[root@hsw]# 154.fi 155 156.SH PERIODIC INTERVAL EXAMPLE 157Without a command to fork, turbostat displays statistics ever 5 seconds. 158Periodic output goes to stdout, by default, unless --out is used to specify an output file. 159The 5-second interval can be changed with the "-i sec" option. 160.nf 161sudo ./turbostat --quiet --hide sysfs,IRQ,SMI,CoreTmp,PkgTmp,GFX%rc6,GFXMHz,PkgWatt,CorWatt,GFXWatt 162 Core CPU Avg_MHz Busy% Bzy_MHz TSC_MHz CPU%c1 CPU%c3 CPU%c6 CPU%c7 163 - - 488 12.52 3900 3498 12.50 0.00 0.00 74.98 164 0 0 5 0.13 3900 3498 99.87 0.00 0.00 0.00 165 0 4 3897 99.99 3900 3498 0.01 166 1 1 0 0.00 3856 3498 0.01 0.00 0.00 99.98 167 1 5 0 0.00 3861 3498 0.01 168 2 2 1 0.02 3889 3498 0.03 0.00 0.00 99.95 169 2 6 0 0.00 3863 3498 0.05 170 3 3 0 0.01 3869 3498 0.02 0.00 0.00 99.97 171 3 7 0 0.00 3878 3498 0.03 172 Core CPU Avg_MHz Busy% Bzy_MHz TSC_MHz CPU%c1 CPU%c3 CPU%c6 CPU%c7 173 - - 491 12.59 3900 3498 12.42 0.00 0.00 74.99 174 0 0 27 0.69 3900 3498 99.31 0.00 0.00 0.00 175 0 4 3898 99.99 3900 3498 0.01 176 1 1 0 0.00 3883 3498 0.01 0.00 0.00 99.99 177 1 5 0 0.00 3898 3498 0.01 178 2 2 0 0.01 3889 3498 0.02 0.00 0.00 99.98 179 2 6 0 0.00 3889 3498 0.02 180 3 3 0 0.00 3856 3498 0.01 0.00 0.00 99.99 181 3 7 0 0.00 3897 3498 0.01 182.fi 183This example also shows the use of the --hide option to skip columns that are not wanted. 184Note that cpu4 in this example is 99.99% busy, while the other CPUs are all under 1% busy. 185Notice that cpu4's HT sibling is cpu0, which is under 1% busy, but can get into CPU%c1 only, 186because its cpu4's activity on shared hardware keeps it from entering a deeper C-state. 187 188.SH SYSTEM CONFIGURATION INFORMATION EXAMPLE 189 190By default, turbostat always dumps system configuration information 191before taking measurements. In the example above, "--quiet" is used 192to suppress that output. Here is an example of the configuration information: 193.nf 194turbostat version 2017.02.15 - Len Brown <lenb@kernel.org> 195CPUID(0): GenuineIntel 13 CPUID levels; family:model:stepping 0x6:3c:3 (6:60:3) 196CPUID(1): SSE3 MONITOR - EIST TM2 TSC MSR ACPI-TM TM 197CPUID(6): APERF, TURBO, DTS, PTM, No-HWP, No-HWPnotify, No-HWPwindow, No-HWPepp, No-HWPpkg, EPB 198cpu4: MSR_IA32_MISC_ENABLE: 0x00850089 (TCC EIST No-MWAIT PREFETCH TURBO) 199CPUID(7): No-SGX 200cpu4: MSR_MISC_PWR_MGMT: 0x00400000 (ENable-EIST_Coordination DISable-EPB DISable-OOB) 201RAPL: 3121 sec. Joule Counter Range, at 84 Watts 202cpu4: MSR_PLATFORM_INFO: 0x80838f3012300 2038 * 100.0 = 800.0 MHz max efficiency frequency 20435 * 100.0 = 3500.0 MHz base frequency 205cpu4: MSR_IA32_POWER_CTL: 0x0004005d (C1E auto-promotion: DISabled) 206cpu4: MSR_TURBO_RATIO_LIMIT: 0x25262727 20737 * 100.0 = 3700.0 MHz max turbo 4 active cores 20838 * 100.0 = 3800.0 MHz max turbo 3 active cores 20939 * 100.0 = 3900.0 MHz max turbo 2 active cores 21039 * 100.0 = 3900.0 MHz max turbo 1 active cores 211cpu4: MSR_CONFIG_TDP_NOMINAL: 0x00000023 (base_ratio=35) 212cpu4: MSR_CONFIG_TDP_LEVEL_1: 0x00000000 () 213cpu4: MSR_CONFIG_TDP_LEVEL_2: 0x00000000 () 214cpu4: MSR_CONFIG_TDP_CONTROL: 0x80000000 ( lock=1) 215cpu4: MSR_TURBO_ACTIVATION_RATIO: 0x00000000 (MAX_NON_TURBO_RATIO=0 lock=0) 216cpu4: MSR_PKG_CST_CONFIG_CONTROL: 0x1e000400 (UNdemote-C3, UNdemote-C1, demote-C3, demote-C1, UNlocked: pkg-cstate-limit=0: pc0) 217cpu4: POLL: CPUIDLE CORE POLL IDLE 218cpu4: C1: MWAIT 0x00 219cpu4: C1E: MWAIT 0x01 220cpu4: C3: MWAIT 0x10 221cpu4: C6: MWAIT 0x20 222cpu4: C7s: MWAIT 0x32 223cpu4: MSR_MISC_FEATURE_CONTROL: 0x00000000 (L2-Prefetch L2-Prefetch-pair L1-Prefetch L1-IP-Prefetch) 224cpu0: MSR_IA32_ENERGY_PERF_BIAS: 0x00000006 (balanced) 225cpu0: MSR_CORE_PERF_LIMIT_REASONS, 0x31200000 (Active: ) (Logged: Transitions, MultiCoreTurbo, Amps, Auto-HWP, ) 226cpu0: MSR_GFX_PERF_LIMIT_REASONS, 0x00000000 (Active: ) (Logged: ) 227cpu0: MSR_RING_PERF_LIMIT_REASONS, 0x0d000000 (Active: ) (Logged: Amps, PkgPwrL1, PkgPwrL2, ) 228cpu0: MSR_RAPL_POWER_UNIT: 0x000a0e03 (0.125000 Watts, 0.000061 Joules, 0.000977 sec.) 229cpu0: MSR_PKG_POWER_INFO: 0x000002a0 (84 W TDP, RAPL 0 - 0 W, 0.000000 sec.) 230cpu0: MSR_PKG_POWER_LIMIT: 0x428348001a82a0 (UNlocked) 231cpu0: PKG Limit #1: ENabled (84.000000 Watts, 8.000000 sec, clamp DISabled) 232cpu0: PKG Limit #2: ENabled (105.000000 Watts, 0.002441* sec, clamp DISabled) 233cpu0: MSR_PP0_POLICY: 0 234cpu0: MSR_PP0_POWER_LIMIT: 0x00000000 (UNlocked) 235cpu0: Cores Limit: DISabled (0.000000 Watts, 0.000977 sec, clamp DISabled) 236cpu0: MSR_PP1_POLICY: 0 237cpu0: MSR_PP1_POWER_LIMIT: 0x00000000 (UNlocked) 238cpu0: GFX Limit: DISabled (0.000000 Watts, 0.000977 sec, clamp DISabled) 239cpu0: MSR_IA32_TEMPERATURE_TARGET: 0x00641400 (100 C) 240cpu0: MSR_IA32_PACKAGE_THERM_STATUS: 0x884c0800 (24 C) 241cpu0: MSR_IA32_THERM_STATUS: 0x884c0000 (24 C +/- 1) 242cpu1: MSR_IA32_THERM_STATUS: 0x88510000 (19 C +/- 1) 243cpu2: MSR_IA32_THERM_STATUS: 0x884e0000 (22 C +/- 1) 244cpu3: MSR_IA32_THERM_STATUS: 0x88510000 (19 C +/- 1) 245cpu4: MSR_PKGC3_IRTL: 0x00008842 (valid, 67584 ns) 246cpu4: MSR_PKGC6_IRTL: 0x00008873 (valid, 117760 ns) 247cpu4: MSR_PKGC7_IRTL: 0x00008891 (valid, 148480 ns) 248.fi 249The \fBmax efficiency\fP frequency, a.k.a. Low Frequency Mode, is the frequency 250available at the minimum package voltage. The \fBTSC frequency\fP is the base 251frequency of the processor -- this should match the brand string 252in /proc/cpuinfo. This base frequency 253should be sustainable on all CPUs indefinitely, given nominal power and cooling. 254The remaining rows show what maximum turbo frequency is possible 255depending on the number of idle cores. Note that not all information is 256available on all processors. 257.SH ADD COUNTER EXAMPLE 258Here we limit turbostat to showing just the CPU number for cpu0 - cpu3. 259We add a counter showing the 32-bit raw value of MSR 0x199 (MSR_IA32_PERF_CTL), 260labeling it with the column header, "PRF_CTRL", and display it only once, 261afte the conclusion of a 0.1 second sleep. 262.nf 263sudo ./turbostat --quiet --cpu 0-3 --show CPU --add msr0x199,u32,raw,PRF_CTRL sleep .1 2640.101604 sec 265CPU PRF_CTRL 266- 0x00000000 2670 0x00000c00 2681 0x00000800 2692 0x00000a00 2703 0x00000800 271 272.fi 273 274.SH INPUT 275 276For interval-mode, turbostat will immediately end the current interval 277when it sees a newline on standard input. 278turbostat will then start the next interval. 279Control-C will be send a SIGINT to turbostat, 280which will immediately abort the program with no further processing. 281.SH SIGNALS 282 283SIGINT will interrupt interval-mode. 284The end-of-interval data will be collected and displayed before turbostat exits. 285 286SIGUSR1 will end current interval, 287end-of-interval data will be collected and displayed before turbostat 288starts a new interval. 289.SH NOTES 290 291.B "turbostat " 292must be run as root. 293Alternatively, non-root users can be enabled to run turbostat this way: 294 295# setcap cap_sys_admin,cap_sys_rawio,cap_sys_nice=+ep ./turbostat 296 297# chmod +r /dev/cpu/*/msr 298 299.B "turbostat " 300reads hardware counters, but doesn't write them. 301So it will not interfere with the OS or other programs, including 302multiple invocations of itself. 303 304\fBturbostat \fP 305may work poorly on Linux-2.6.20 through 2.6.29, 306as \fBacpi-cpufreq \fPperiodically cleared the APERF and MPERF MSRs 307in those kernels. 308 309AVG_MHz = APERF_delta/measurement_interval. This is the actual 310number of elapsed cycles divided by the entire sample interval -- 311including idle time. Note that this calculation is resilient 312to systems lacking a non-stop TSC. 313 314TSC_MHz = TSC_delta/measurement_interval. 315On a system with an invariant TSC, this value will be constant 316and will closely match the base frequency value shown 317in the brand string in /proc/cpuinfo. On a system where 318the TSC stops in idle, TSC_MHz will drop 319below the processor's base frequency. 320 321Busy% = MPERF_delta/TSC_delta 322 323Bzy_MHz = TSC_delta/APERF_delta/MPERF_delta/measurement_interval 324 325Note that these calculations depend on TSC_delta, so they 326are not reliable during intervals when TSC_MHz is not running at the base frequency. 327 328Turbostat data collection is not atomic. 329Extremely short measurement intervals (much less than 1 second), 330or system activity that prevents turbostat from being able 331to run on all CPUS to quickly collect data, will result in 332inconsistent results. 333 334The APERF, MPERF MSRs are defined to count non-halted cycles. 335Although it is not guaranteed by the architecture, turbostat assumes 336that they count at TSC rate, which is true on all processors tested to date. 337 338.SH REFERENCES 339Volume 3B: System Programming Guide" 340https://www.intel.com/products/processor/manuals/ 341 342.SH FILES 343.ta 344.nf 345/dev/cpu/*/msr 346.fi 347 348.SH "SEE ALSO" 349msr(4), vmstat(8) 350.PP 351.SH AUTHOR 352.nf 353Written by Len Brown <len.brown@intel.com>