CPU watt solution

[birdie-github]

Under Linux it's absolutely possible to read CPU watt usage without (lm-) sensors.

It involves calling turbostat (which works for both Intel and AMD CPUs, I'm not sure about Qualcomm but they are still extremely rare), e.g.

turbostat --quiet --num_iterations 1 --interval 1 --show PkgWatt --Summary | tail -1 | tr -d '\n'

Or you could do whatever turbostat is doing yourself. There's an API for that.

1. In both cases, the msr kernel module must be loaded.
2. setcap requires this to work under a non-privileged user:

   setcap cap_sys_rawio,cap_sys_nice=+ep /usr/bin/turbostat

Prolly mangohud will require it too.

This solution is universal, supported for all AMD/Intel CPUs released in the past decade and will be supported going forward.

[flightlessmango]

you can't setcap a shared library, which mangohud is. At best you could wrap turbostat in the code and make the user manually apply setcap to it. Either way, what problem is being solved here? in general there shouldn't be any issue getting cpu wattage for intel or amd

[birdie-github]

you can't setcap a shared library, which mangohud is.

You're absolutely right.

you can't setcap a shared library, which mangohud is.

Probably that's the way.

Either way, what problem is being solved here? in general there shouldn't be any issue getting cpu wattage for intel or amd

But it's shown as 0W here for the Ryzen 7 5800X. AFAIK you rely on the unofficial zenpower3 driver which hasn't been maintained in years and not a single Linux distro comes with it by default.

[flightlessmango]

Using a kernel module like zenpower3 is a better solution than wrapping another application and depending on it

[birdie-github]

• The vast majority of users will never install it for various reasons
• zenpower3 hasn't been maintained for years and may break any minute

OK, you don't like the idea of calling turbostat.

What about RAPL?

https://web.eece.maine.edu/~vweaver/projects/rapl/ https://github.com/deater/uarch-configure/blob/master/rapl-read/rapl-read.c

Only requires read access to /dev/cpu/N/msr

[flightlessmango]

The vast majority of users will never install it for various reasons

How is it any better that the user would have to install turbostat instead that they also need to set special permissions for?

zenpower3 hasn't been maintained for years and may break any minute

Does it need maintenance? Do you know that the developer has abandoned it or something?

Another option is https://github.com/BoukeHaarsma23/zenergy

[birdie-github]

Given proper MSR access, no root or anything is required to read RAPL data.

That's as easy as:

/etc/udev/rules.d/99-local-msr.rules
SUBSYSTEM=="msr", MODE="0640", GROUP="wheel"

Could be adjusted for your use case.

Out of tree kernel modules are unavailable to 99.9% of Linux users.

[flightlessmango]

The problem with making msr accessible by the user is that it doesn't only let mangohud access it

[birdie-github]

The problem with making msr accessible by the user is that it doesn't only let mangohud access it

1. The vast majority of desktop Linux users are already local admins via sudo, su or root access anyways.
2. Read only MSR access is not something to really worry about. It's not like MSR is direct RAM access, no, it only allows to read certain known CPU registers (e.g. performance counters and power use information).

[flightlessmango]

Using sudo is limited exposure, intentionally allowing a specific application access. This is giving all applications access. msr information can be used for exploits, as such the question is if it's worth exposing users to it than making a proper solution.

We already do some of this by exposing intel rapl, but question is if we should double down on it

[birdie-github]

msr information can be used for exploits, as such the question is if it's worth exposing users to it than making a proper solution.

You're right.

Looks like Linux users are screwed and there's no solution. I'm closing this ticket because I have no idea how to solve this problem. I've exhausted all my options and ideas.

[Kraust]

Seems like the MangoHud team is having the same issues I am having now.

None of the tools out there right now that support Zen Temperature Readings except for lm_sensors properly support Zen 4 (and probably Zen 5). zenpower3 was called out here, but does not have Zen 4+ support. The same can be said for every other Zen-integration for system monitoring except for cpu-speed which uses lm_sensors under the hood.

Came here to see if there was any progress with MangoHud and integrating per-CPU Clocks or even per-CCD Temps / Power but it looks like everyone is limited by the same issue - nobody seems to know how to effectively get the data out of the Zen Power Tables. I made an attempt but without knowing how to map the Power Tables no progress can really be made.

[birdie-github]

@Kraust

turbostat works just fine even without root access and I've been using it for years. What's more it supports pretty much all the CPUs that can report power consumption. This method doesn't need lm-sensors support.

It's just unhandy to implement since you need to system() an external application.

I'm not a good C programmer, so I just walked away. I guess a patch to implement this feature will be fewer than 100 lines of code.

You create a new thread, run every second turbostat --quiet --num_iterations 1 --interval 1 --show PkgWatt --Summary | tail -1 | tr -d '\n' in it and assign the result of this function to a variable. That's it.

In a perfect world you will not use tail and tr and instead parse the output using C. It's gonna be faster and more efficient.

And of course you need to make sure that tr 1) doesn't hang up and 2) exits successfully. You don't want to end up running thousands of turbostat processes.

[Kraust]

@birdie-github Have you actually tried this on Zen 4?

[nix-shell:~]$ turbostat --quiet --num_iterations 1 --interval 1 --show PkgWatt --Summary

The fields you think that are included are not included. I think that's the thing you're missing in this chain and why you likely closed the issue. This is a software development issue and nobody is willing to maintain because it requires the end users to figure out mappings which are not immediately obvious (e.g. I struggled with it for a few days and made minimal progress.

[nix-shell:~]$ turbostat --quiet --num_iterations 1 --interval 1 --no-perf
Die     Core    CPU     TSC_MHz IRQ     POLL    C1      C2      C3      POLL%   C1%     C2%     C3%
-       -       -       4198    5858    25      108     1046    4221    0.00    0.09    1.92    96.90
0       0       0       4200    245     0       3       39      193     0.00    0.09    1.91    97.40
0       0       16      4200    14      0       0       4       10      0.00    0.00    0.31    99.67
0       1       1       4200    72      0       0       9       55      0.00    0.00    0.24    99.54
0       1       17      4200    12      0       0       0       11      0.00    0.00    0.00    99.96
0       2       2       4200    408     0       9       86      305     0.00    0.25    4.33    94.21
0       2       18      4200    5       0       0       1       4       0.00    0.00    0.01    99.98
0       3       3       4200    114     0       0       4       110     0.00    0.00    0.13    99.59
0       3       19      4200    4       0       0       1       3       0.00    0.00    0.02    99.97
0       4       4       4200    236     0       10      56      134     0.00    0.04    3.20    94.52
0       4       20      4200    53      0       13      7       25      0.00    0.79    0.40    98.73
0       5       5       4200    329     0       2       59      251     0.00    0.02    2.34    95.56
0       5       21      4200    20      0       0       0       20      0.00    0.00    0.00    99.94
0       6       6       4200    78      0       0       0       76      0.00    0.00    0.00    99.52
0       6       22      4200    2       0       0       0       2       0.00    0.00    0.00    99.99
0       7       7       4200    3       0       0       0       3       0.00    0.00    0.00    99.98
0       7       23      4200    1       0       0       0       1       0.00    0.00    0.00    99.99
1       8       8       4200    224     0       2       9       199     0.00    0.06    0.25    99.10
1       8       24      4200    91      0       0       0       79      0.00    0.00    0.00    99.37
1       9       9       4200    754     10      9       260     327     0.03    0.02    17.00   72.56
1       9       25      4200    348     2       3       9       252     0.01    0.04    0.36    92.90
1       10      10      4200    90      0       0       13      72      0.00    0.00    1.20    98.51
1       10      26      4200    130     0       0       7       118     0.00    0.00    0.61    99.03
1       11      11      4200    729     13      14      375     302     0.05    0.02    24.02   74.69
1       11      27      4200    270     0       4       13      248     0.00    0.10    0.83    98.26
1       12      12      4200    127     0       0       14      97      0.00    0.00    1.08    97.78
1       12      28      4200    293     0       28      13      247     0.00    1.29    0.33    97.40
1       13      13      4200    226     0       0       3       212     0.00    0.00    0.11    98.97
1       13      29      4200    397     0       9       40      339     0.00    0.23    1.81    97.37
1       14      14      4200    100     0       0       3       90      0.00    0.00    0.09    99.59
1       14      30      4200    75      0       0       0       66      0.00    0.00    0.00    99.56
1       15      15      4200    241     0       1       21      214     0.00    0.00    0.82    98.74
1       15      31      4200    167     0       1       0       156     0.00    0.08    0.00    99.60

FYI. What is actually output on Zen 4 using Kernel 6.10 and a 7950X3D. Also that data is wrong because there are actual tools out there that properly report CPU frequency because that's actually exported by sysfs.

[birdie-github]

I've not tried it on Zen 4, I've been using it on Zen 4 from day one.

RAPL is universal.

This is a software development issue and nobody is willing to maintain because it requires the end users to figure out mappings which are not immediately obvious (e.g. I struggled with it for a few days and made minimal progress.

RAPL works under Windows as well and it's here to stay. I've no idea what you're insinuating about.

Something about your system is terribly wrong, you may need to update your BIOS or something.

Here's my output for the Zen 4 based Ryzen 7 7840HS:

turbostat --quiet --num_iterations 1 --interval 1 --no-perf
Core    CPU Avg_MHz Busy%   Bzy_MHz TSC_MHz IRQ POLL    C1  C2  C3  POLL%   C1% C2% C3% CorWatt PkgWatt
-   -   101 3.75    2689    3793    6121    3   193 1328    4032    0.00    0.26    4.33    91.99   5.04    5.61
0   0   84  3.95    2137    3793    749 0   27  97  563 0.00    1.20    4.57    90.97   0.54    5.61
0   1   7   0.33    2174    3793    59  0   0   0   58  0.00    0.00    0.00    99.73
1   2   261 8.46    3082    3793    794 1   23  345 358 0.00    0.17    17.35   74.63   1.23
1   3   96  3.52    2739    3793    457 0   45  37  373 0.00    1.42    1.56    93.94
2   4   123 5.04    2441    3793    576 0   6   146 371 0.00    0.08    7.76    87.63   0.45
2   5   4   0.21    2095    3793    45  0   1   0   38  0.00    0.03    0.00    99.79
3   6   79  3.49    2273    3793    515 1   31  98  344 0.00    0.40    5.32    91.22   0.29
3   7   4   0.09    4307    3793    15  0   0   5   9   0.00    0.00    0.39    99.53
4   8   80  3.37    2372    3793    475 0   8   85  346 0.00    0.15    4.72    92.15   0.55
4   9   50  2.16    2312    3793    268 0   0   29  217 0.00    0.00    1.59    96.53
5   10  500 16.58   3015    3793    847 1   12  204 476 0.00    0.14    11.06   72.91   0.82
5   11  19  0.82    2315    3793    67  0   2   19  37  0.00    0.19    1.26    97.78
6   12  157 5.56    2829    3793    580 0   7   125 390 0.00    0.11    6.22    88.61   0.76
6   13  20  0.91    2150    3793    103 0   0   0   94  0.00    0.00    0.00    99.22
7   14  104 4.55    2283    3793    515 0   31  138 312 0.00    0.27    7.56    88.05   0.41
7   15  23  0.90    2558    3793    56  0   0   0   46  0.00    0.00    0.00    99.15

All exactly the same as in Windows 10/11 using HWiNFO64.

I've no idea why your output is missing CorWatt PkgWatt columns altogether.

[birdie-github]

A Ryzen 9 7950X's output (taken from https://www.realworldtech.com/forum/?threadid=210843&curpostid=210872 ):

turbostat version 21.05.04 - Len Brown
CPUID(0): AuthenticAMD 0x10 CPUID levels
CPUID(1): family:model:stepping 0x19:61:2 (25:97:2) microcode 0x0
CPUID(0x80000000): max_extended_levels: 0x80000028
CPUID(1): SSE3 MONITOR - - - TSC MSR - HT -
CPUID(6): APERF, No-TURBO, No-DTS, No-PTM, No-HWP, No-HWPnotify, No-HWPwindow, No-HWPepp, No-HWPpkg, No-EPB
CPUID(7): No-SGX
RAPL: 234 sec. Joule Counter Range, at 280 Watts
/dev/cpu_dma_latency: 2000000000 usec (default)
current_driver: acpi_idle
current_governor: menu
current_governor_ro: menu
cpu1: POLL: CPUIDLE CORE POLL IDLE
cpu1: C1: ACPI FFH MWAIT 0x0
cpu1: C2: ACPI IOPORT 0x414
cpu1: C3: ACPI IOPORT 0x415
cpu1: cpufreq driver: acpi-cpufreq
cpu1: cpufreq governor: schedutil
cpufreq boost: 1
cpu0: MSR_RAPL_PWR_UNIT: 0x000a1000 (1.000000 Watts, 0.000015 Joules, 0.000977 sec.)

Core    CPU Avg_MHz Busy%   Bzy_MHz TSC_MHz IPC IRQ POLL    C1  C2  C3  POLL%   C1% C2% C3% CorWatt PkgWatt
-   -   4531    99.65   4547    4500    1.02    52975   3   186 59  229 0.00    0.01    0.03    0.31    100.23  152.59
0   0   4590    99.88   4595    4500    0.85    1539    0   1   1   1   0.00    0.00    0.04    0.08    6.33    152.57
0   16  4576    99.59   4595    4500    1.07    1440    0   0   2   9   0.00    0.00    0.00    0.41
1   1   4535    98.69   4596    4500    1.07    2328    0   4   4   26  0.00    0.01    0.00    1.31    6.69
1   17  4552    99.06   4596    4500    1.02    2198    0   1   5   26  0.00    0.00    0.13    0.82
2   2   4594    99.99   4595    4500    1.04    1379    0   0   0   1   0.00    0.00    0.00    0.01    6.86
2   18  4585    99.79   4595    4500    1.15    1420    1   22  3   4   0.00    0.02    0.03    0.15
3   3   4594    99.97   4595    4500    1.05    1503    0   1   1   1   0.00    0.00    0.00    0.02    7.00
3   19  4589    99.87   4595    4500    1.03    1451    0   1   2   3   0.00    0.00    0.01    0.12
4   4   4585    99.78   4595    4500    1.03    1413    0   3   0   4   0.00    0.00    0.00    0.22    6.88
4   20  4594    99.98   4595    4500    1.01    1286    0   9   1   1   0.00    0.02    0.00    0.00
5   5   4594    99.98   4595    4500    1.09    1579    0   21  0   0   0.00    0.02    0.00    0.00    6.94
5   21  4593    99.95   4595    4500    1.11    1532    1   0   2   1   0.00    0.00    0.01    0.04
6   6   4586    99.80   4595    4500    0.97    1398    0   0   3   3   0.00    0.00    0.13    0.08    6.59
6   22  4595    100.00  4595    4500    0.99    1274    0   1   0   0   0.00    0.00    0.00    0.00
7   7   4568    99.40   4595    4500    0.94    2207    0   52  3   14  0.00    0.06    0.08    0.46    6.45
7   23  4575    99.58   4595    4500    1.04    2442    0   14  4   9   0.00    0.02    0.07    0.33
8   8   4499    100.00  4499    4500    1.00    1402    0   0   0   0   0.00    0.00    0.00    0.00    5.74
8   24  4474    99.45   4499    4500    0.98    1424    0   1   1   13  0.00    0.00    0.04    0.51
9   9   4490    99.80   4499    4500    0.80    2094    0   5   5   7   0.00    0.01    0.10    0.10    5.51
9   25  4499    100.00  4499    4500    1.04    1305    0   0   0   0   0.00    0.00    0.00    0.00
10  10  4491    99.82   4499    4500    1.04    1285    0   4   1   3   0.00    0.00    0.01    0.16    6.07
10  26  4496    99.94   4499    4500    1.00    3922    0   8   1   1   0.00    0.02    0.00    0.04
11  11  4497    99.95   4499    4500    0.94    1560    0   0   0   1   0.00    0.00    0.00    0.05    5.78
11  27  4496    99.94   4499    4500    0.96    1436    0   0   2   4   0.00    0.00    0.03    0.03
12  12  4458    99.09   4499    4500    0.98    1754    0   1   3   15  0.00    0.00    0.07    0.85    5.83
12  28  4464    99.22   4499    4500    0.97    1393    0   10  1   14  0.00    0.02    0.01    0.76
13  13  4399    97.77   4499    4500    1.14    1591    0   5   5   44  0.00    0.00    0.10    2.14    5.90
13  29  4484    99.67   4499    4500    1.25    1450    0   11  2   6   0.00    0.01    0.05    0.27
14  14  4499    100.00  4499    4500    1.10    1353    0   0   0   0   0.00    0.00    0.00    0.00    5.84
14  30  4467    99.29   4499    4500    0.88    1359    0   0   0   9   0.00    0.00    0.00    0.71
15  15  4494    99.89   4499    4500    1.02    1666    1   7   5   0   0.00    0.01    0.10    0.00    5.82
15  31  4482    99.63   4499    4500    1.03    1592    0   4   2   9   0.00    0.01    0.07    0.29

Core    CPU Avg_MHz Busy%   Bzy_MHz TSC_MHz IPC IRQ POLL    C1  C2  C3  POLL%   C1% C2% C3% CorWatt PkgWatt
-   -   3   0.10    3081    4500    0.57    2126    1   54  79  1917    0.00    0.03    0.07    99.82   0.09    23.95
0   0   6   0.18    3065    4500    1.66    62  0   29  5   52  0.00    0.07    0.02    99.73   0.01    23.96
0   16  4   0.10    3542    4500    1.01    48  0   0   1   45  0.00    0.00    0.01    99.90
1   1   3   0.11    2982    4500    1.09    47  0   0   4   39  0.00    0.00    0.14    99.76   0.00
1   17  1   0.02    2909    4500    0.43    22  0   0   1   19  0.00    0.00    0.00    99.98
2   2   4   0.12    2953    4500    0.21    161 0   0   1   157 0.00    0.00    0.00    99.90   0.01
2   18  2   0.05    2949    4500    0.19    62  0   0   1   60  0.00    0.00    0.00    99.96
3   3   2   0.07    2950    4500    0.17    45  0   0   2   41  0.00    0.00    0.02    99.92   0.00
3   19  2   0.08    2964    4500    0.21    106 0   13  10  87  0.00    0.57    0.65    98.72
4   4   6   0.20    3040    4500    1.89    51  0   9   0   41  0.00    0.21    0.00    99.60   0.01
4   20  1   0.03    3304    4500    0.18    54  0   0   0   51  0.00    0.00    0.00    99.98
5   5   1   0.02    2901    4500    0.25    29  0   0   3   25  0.00    0.00    0.03    99.96   0.00
5   21  1   0.02    2911    4500    0.22    28  0   0   0   27  0.00    0.00    0.00    99.99
6   6   16  0.48    3355    4500    0.39    170 0   0   13  145 0.00    0.00    0.13    99.41   0.01
6   22  1   0.03    3243    4500    0.35    25  0   0   0   23  0.00    0.00    0.00    99.98
7   7   2   0.07    3167    4500    0.65    57  1   0   0   51  0.00    0.00    0.00    99.94   0.00
7   23  4   0.12    3050    4500    0.82    50  0   0   0   31  0.00    0.00    0.00    99.89
8   8   3   0.11    2952    4500    0.28    59  0   0   0   57  0.00    0.00    0.00    99.90   0.00
8   24  1   0.05    2922    4500    0.47    49  0   0   4   44  0.00    0.00    0.08    99.89
9   9   9   0.31    2975    4500    0.24    211 0   0   5   190 0.00    0.00    0.05    99.68   0.02
9   25  7   0.25    2964    4500    0.12    112 0   2   19  88  0.00    0.01    0.98    98.79
10  10  1   0.04    2908    4500    0.23    54  0   1   1   51  0.00    0.01    0.02    99.95   0.00
10  26  1   0.05    2943    4500    0.28    38  0   0   0   36  0.00    0.00    0.00    99.97
11  11  4   0.14    2969    4500    0.15    146 0   0   3   139 0.00    0.00    0.03    99.86   0.01
11  27  2   0.06    2943    4500    0.18    67  0   0   1   65  0.00    0.00    0.01    99.95
12  12  1   0.02    2880    4500    0.35    37  0   0   1   31  0.00    0.00    0.00    99.99   0.00
12  28  2   0.05    2939    4500    0.66    33  0   0   0   32  0.00    0.00    0.00    99.96
13  13  2   0.05    2928    4500    0.79    53  0   0   0   51  0.00    0.00    0.00    99.96   0.00
13  29  1   0.05    2940    4500    0.22    73  0   0   2   70  0.00    0.00    0.02    99.94
14  14  2   0.05    3220    4500    0.42    50  0   0   0   49  0.00    0.00    0.00    99.96   0.01
14  30  2   0.06    3257    4500    0.23    61  0   0   2   58  0.00    0.00    0.06    99.89
15  15  1   0.02    3386    4500    0.18    27  0   0   0   26  0.00    0.00    0.00    99.99   0.00
15  31  1   0.04    3174    4500    1.17    39  0   0   0   36  0.00    0.00    0.00    99.96

[birdie-github]

I've been using Linux for almost 30 years. Never been known for BS'ing people.

[birdie-github]

Anyways, I couldn't care less about brain damaged ideas shared between and perpetrated by Linux developers.

You have secure boot on? No write access to MSR for you. Totally possible in Windows.

You wanna read CPU power consumption via lm-sensors? Well, let's remove the feature from the kernel.

You wanna read CPU power consumption via turbostat? Well, go throw several circles of hell, enabling read access to MSR and setting scarily looking flags on turbostat.

This is not a desktop OS, this is an abomination.

[Kraust]

Ended up figuring it out. This tool does everything but do temperature reading which is pretty nice, thanks.

[birdie-github]

@flightlessmango

Maybe you could consider implementing a turbostat solution. It works and it's there to stay for the foreseeable future.

[flightlessmango]

Maybe you could consider implementing a turbostat solution. It works and it's there to stay for the foreseeable future.

Like discussed previously in this thread, I don't like wrapping and depending on a application. We used to do this with intel_gpu_top and it caused more issues than it solved

[birdie-github]

@flightlessmango

Are you saying you won't even accept a merge request if someone implements the feature?