lscpu on Mac: 0 of 11 Answers Find the M4 Clock Speed

Ask a Mac for its CPU details the Linux way and you get lscpu: command not found. That part is obvious. What isn't obvious is that the usual replacement, sysctl, has a field that fails without telling you. On the M4 Mac mini that runs this business, sysctl hw.cpufrequency printed nothing and exited 0. A key that doesn't exist at all prints "unknown oid" and exits 1. So a script that reads the clock speed gets an empty string with a success code, and the first sign of trouble is a bash arithmetic error two lines later.

"lscpu mac" gets 35 Google autocomplete completions. I mapped each lscpu field to what macOS 26 actually returns, ran the 11 answers on the two Ask Different threads about CPU details and clock speed, and tried the two Python libraries a script would reach for. Most of lscpu maps cleanly. Clock speed, cache size and the flags line don't.

Why there is no lscpu on Mac, even with Homebrew

lscpu belongs to util-linux and reads /proc/cpuinfo and /sys, which macOS doesn't have. Homebrew does ship util-linux (2.42.4 today), but the formula keeps a linux_only_bins list that includes lscpu, lsblk and lsmem, and its caveat says those tools "are not supported for macOS, and are therefore not included". brew search lscpu returns one unrelated formula. I hit the same wall with lsblk on Mac.

Each lscpu field and what macOS 26 gives you

All values below come from a Mac16,10 (Apple M4, 4 performance and 6 efficiency cores) on macOS 26.4.1, read on 2026-10-04 with a stock PATH:

lscpu field	macOS source	Value here
Architecture	uname -m or sysctl hw.machine	arm64
Byte Order	sysctl hw.byteorder	1234 (little endian)
CPU(s)	sysctl hw.ncpu	10
Model name	sysctl machdep.cpu.brand_string	Apple M4
Socket(s)	sysctl hw.packages	1
Thread(s) per core	hw.logicalcpu / hw.physicalcpu	10 / 10 = 1
L1d / L1i / L2	hw.perflevelN.l1dcachesize etc.	P: 128 KiB / 192 KiB / 16 MiB per 4 cores; E: 64 KiB / 128 KiB / 4 MiB per 6 cores
L3	sysctl hw.l3cachesize	nothing, exit 0
CPU max MHz	sysctl hw.cpufrequency_max	nothing, exit 0
Flags	sysctl hw.optional	64 of 86 keys set to 1
Virtualization	sysctl kern.hv_support	1 (Hypervisor.framework)

Core counts are the solid part, and I covered which one to use for a build in the nproc post. The machdep.cpu answer that tops Ask Different (83 votes) still runs, but its Intel sample showed vendor, family, model, a feature list and a brand string ending in "@ 1.60GHz". On this M4 the same grep returned five lines: four core counts and "Apple M4". The clock speed that used to ride along in the brand string is gone.

The clock speed field: 11 answers, 0 numbers without sudo

The clock question has its own Ask Different thread (46,380 views, 9 answers), and the CPU-details thread (126,552 views) adds 2 more. I ran every command in all 11 answers on this Mac, without root:

• 3 need sudo: all use powermetrics. One answer shows real per-cluster output from an Apple silicon Mac. This machine runs unattended with no passwordless sudo, so I didn't run them.
• 2 are Intel-only: Intel Power Gadget and its PowerLog command.
• 2 are GUI apps: About This Mac, and an app called Mx Power Gadget.
• 2 print the wrong number: sysctl -a | grep freq returns hw.tbfrequency: 24000000, which is the 24 MHz system timer, not the CPU. top -F reports CPU usage percentages.
• 1 prints no number: pmset -g therm gave three "No ... has been recorded" notes.
• 1 has no clock to show: the machdep.cpu grep above.

The libraries do no better. psutil issue #2642, opened in October 2025, reports that cpu_freq() is off by 1000x on M4 because Apple changed the unit in its power-manager tables from Hz to kHz. The fix merged on July 17, 2026, but the newest release on PyPI is still 7.2.2 from January. Installed today, it returned this:

$ python -c "import psutil; print(psutil.__version__, psutil.cpu_freq())"
7.2.2 scpufreq(current=4, min=0, max=4)

That is 4 MHz for a chip that runs at about 4.4 GHz. py-cpuinfo 9.0.0 returned None for clock speed, both cache sizes and flags.

Where the real number is: the power manager's frequency table

psutil's fix points at the data. The pmgr entry in the I/O Registry carries voltage-states tables, each step being a 4-byte frequency and a 4-byte voltage. Any user can read them:

$ ioreg -lw0 -r -n pmgr | grep -o '"voltage-states[0-9]*-sram"'
"voltage-states5-sram"
"voltage-states9-sram"
"voltage-states1-sram"

Decoded, voltage-states5 has 20 steps from 912,000 to 4,464,000, voltage-states1 has 8 steps ending at 2,892,000, and voltage-states9 has 16 steps in Hz ending at 1,578,000,000. Read as kHz, the first two are the performance and efficiency cluster maximums: 4,464 and 2,892 MHz. Which table belongs to which cluster is my inference, not Apple documentation. It does match my own logs: when I tested a Mac mini M4 cooling stand under load, macmon showed the performance cores holding 4,416 and 4,464 MHz, the top two steps of table 5. In the Llama 3.1 8B benchmark the GPU ran at 1,578 MHz, the top of table 9.

That is the maximum, which is what lscpu's "CPU max MHz" means. For the current clock without sudo, macmon 0.8.2 reads the same counters as powermetrics. Idle, it showed the performance cluster at 2,272 MHz and the efficiency cluster at 1,559.

Cache sizes: the top-level keys describe the slow cores

sysctl hw.l2cachesize printed 4194304 here, 4 MiB. The performance cores have a 16 MiB L2. Apple's sysctl capabilities page does say this: the top-level cache keys are "equivalent to querying the per-performance-level information for the least performant core". So hw.l1icachesize, hw.l1dcachesize and hw.l2cachesize all return efficiency-core values. A script that sizes buffers from hw.l2cachesize uses a quarter of the cache the fast cores have. Use hw.perflevel0.* for the performance cores and hw.perflevel0.cpusperl2 to see how many cores share it. The same page lists hw.l3cachesize, which printed nothing on this M4.

The flags line lives under hw.optional

On Linux arm64, lscpu's Flags line is the kernel's list of CPU features. macOS exposes them as boolean keys. Here 86 hw.optional keys exist and 64 are set to 1. Of those, 42 are Arm architecture features named hw.optional.arm.FEAT_*, including FEAT_SME and FEAT_SME2, Arm's matrix extensions. A missing key and a 0 key both mean the feature is absent, so test for the value 1.

An lscpu function for macOS

This puts the pieces together with nothing outside the base system. It produced identical output under /bin/zsh -f, /bin/bash 3.2 and /bin/sh, in 0.05 seconds:

lscpu() {
  local n i name mhz
  printf '%-22s %s\n' 'Architecture:' "$(uname -m)"
  printf '%-22s %s\n' 'CPU(s):' "$(sysctl -n hw.ncpu)"
  printf '%-22s %s\n' 'Model name:' "$(sysctl -n machdep.cpu.brand_string)"
  n=$(sysctl -n hw.nperflevels 2>/dev/null)
  i=0
  while [ "$i" -lt "${n:-0}" ]; do
    name=$(sysctl -n hw.perflevel$i.name)
    printf '%-22s %s cores, L1i %s KiB, L1d %s KiB, L2 %s MiB per %s cores\n' "$name:" \
      "$(sysctl -n hw.perflevel$i.physicalcpu)" \
      "$(( $(sysctl -n hw.perflevel$i.l1icachesize) / 1024 ))" \
      "$(( $(sysctl -n hw.perflevel$i.l1dcachesize) / 1024 ))" \
      "$(( $(sysctl -n hw.perflevel$i.l2cachesize) / 1048576 ))" \
      "$(sysctl -n hw.perflevel$i.cpusperl2)"
    i=$((i + 1))
  done
  # Intel Macs have hw.cpufrequency_max; Apple silicon prints nothing and exits 0
  mhz=$(sysctl -n hw.cpufrequency_max 2>/dev/null)
  if [ -n "$mhz" ]; then
    mhz=$((mhz / 1000000))
  else
    # pmgr DVFS steps: 4-byte LE frequency + 4-byte voltage. M1-M3 store Hz, M4 kHz.
    mhz=$(ioreg -lw0 -r -n pmgr | perl -ne '
      while (/"voltage-states\d+-sram" = <([0-9a-f]+)>/g) {
        my $b = pack("H*", $1);
        for (my $o = 0; $o + 8 <= length $b; $o += 8) {
          my $f = unpack("V", substr($b, $o, 4));
          $f = $f > 1e8 ? $f / 1e6 : $f / 1e3;
          $m = $f if $f > $m;
        }
      }
      END { printf "%d", $m if $m }')
  fi
  printf '%-22s %s\n' 'CPU max MHz:' "${mhz:-unknown}"
  printf '%-22s %s\n' 'Flags:' "$(sysctl hw.optional 2>/dev/null |
    awk -F': ' '$2 == 1 { sub(/^hw\.optional\.(arm\.)?/, "", $1); printf "%s ", $1 }')"
}

$ lscpu
Architecture:          arm64
CPU(s):                10
Model name:            Apple M4
Performance:           4 cores, L1i 192 KiB, L1d 128 KiB, L2 16 MiB per 4 cores
Efficiency:            6 cores, L1i 128 KiB, L1d 64 KiB, L2 4 MiB per 6 cores
CPU max MHz:           4464
Flags:                 FEAT_CRC32 FEAT_FlagM FEAT_FlagM2 FEAT_FHM FEAT_DotProd ...

The unit check copies psutil's fix: a raw value above 10⁸ is Hz, anything lower is kHz. I could only test the kHz branch on this M4. The function also checks hw.cpufrequency_max for an empty result before doing arithmetic on it, which is the bug the stock answers leave in. The same pattern of a Linux name missing and the common substitute failing quietly shows up in lsusb on Mac and the free command on Mac.

FAQ

What is the lscpu equivalent on Mac?

There isn't a single command. sysctl machdep.cpu.brand_string hw.ncpu hw.perflevel0.physicalcpu hw.perflevel1.physicalcpu covers model and cores, sysctl hw.perflevel0 gives per-core-type cache sizes, and sysctl hw.optional lists CPU features. Homebrew's util-linux deliberately leaves lscpu out on macOS.

Why does sysctl hw.cpufrequency show nothing on Apple silicon?

Apple silicon Macs don't publish a CPU frequency through sysctl. On macOS 26.4.1 the key prints nothing and exits 0, unlike a truly unknown key, which errors. The maximum clock is in the pmgr voltage-states tables in ioreg; the live clock needs sudo powermetrics or a sudoless tool such as macmon.

How do I check CPU speed on an M4 Mac from Terminal?

Decode the pmgr voltage-states tables from ioreg -lw0 -r -n pmgr; on M4 the values are in kHz, giving 4,464 MHz for performance cores and 2,892 MHz for efficiency cores on a Mac mini M4. For the current clock, run sudo powermetrics or install macmon. psutil 7.2.2 reports 4 MHz on M4 because of a unit bug fixed after that release.

Method: every command here ran on 2026-10-04 between about 13:30 and 13:55 KST on a Mac mini M4 (Mac16,10, macOS 26.4.1 build 25E253), with PATH pinned to the system directories except for macmon, which was already installed from Homebrew. The 11 answers are all answers to Ask Different questions 352769 and 328965 as returned by the Stack Exchange API that day; vote and view counts are from the same call. psutil 7.2.2 and py-cpuinfo 9.0.0 were installed into a throwaway virtualenv. I did not run powermetrics, because this machine has no passwordless sudo, and I did not test an Intel Mac, an M1 to M3 machine, or any macOS before 26. The mapping of voltage-states tables to clusters is inferred from matching values, as described above.