Question

5

How to enable AMD Turbo Core on Ubuntu 14.04?

rated 0 times [ 5] [ 0] / answers: 1 / hits: 16729 / 1 Year ago, thu, february 16, 2023, 3:06:33

I'm using Ubuntu 14.04 (3.13.0-24-generic kernel) on AMD A8-4500m based laptop, and I've recently noticed, that AMD Turbo Core is not working at all. Everything works on Windows, however, on Linux CPU frequency can't get past 1.9 GHz (checked with cpufreq-aperf).

Here is output of cpupower frequency-info, notice Active: no under boost state support:

analyzing CPU 0:

  driver: acpi-cpufreq

  CPUs which run at the same hardware frequency: 0

  CPUs which need to have their frequency coordinated by software: 0

  maximum transition latency: 4.0 us.

  hardware limits: 1.40 GHz - 1.90 GHz

  available frequency steps: 1.90 GHz, 1.80 GHz, 1.70 GHz, 1.60 GHz, 1.40 GHz

  available cpufreq governors: conservative, ondemand, userspace, powersave, performance

  current policy: frequency should be within 1.90 GHz and 1.90 GHz.

                  The governor "ondemand" may decide which speed to use

                  within this range.

  current CPU frequency is 1.90 GHz (asserted by call to hardware).

  cpufreq stats: 1.90 GHz:32,63%, 1.80 GHz:0,74%, 1.70 GHz:0,50%, 1.60 GHz:1,20%, 1.40 GHz:64,93%  (9287)

  boost state support:

    Supported: yes

    Active: no

    Boost States: 2

    Total States: 8

    Pstate-Pb0: 2800MHz (boost state)

    Pstate-Pb1: 2300MHz (boost state)

    Pstate-P0:  1900MHz

    Pstate-P1:  1800MHz

    Pstate-P2:  1700MHz

    Pstate-P3:  1600MHz

    Pstate-P4:  1400MHz

    Pstate-P5:  900MHz

More tests (ran with ondemand scheduler and 3 instances of ffmpeg decoding 1920x1080 H.264 video in background):

[m132@m132 turbostat]$ sudo cpupower monitor

[sudo] password for m132: 

     |Mperf               || Idle_Stats         

 CPU | C0   | Cx   | Freq || POLL | C1   | C2   

    0| 97,78|  2,22|  1821||  0,00|  0,00|  0,00

    1| 97,84|  2,16|  1821||  0,00|  0,00|  0,00

    2| 99,22|  0,78|  1807||  0,00|  0,00|  0,00

    3| 99,18|  0,82|  1808||  0,00|  0,00|  0,00

[m132@m132 turbostat]$ sudo ./turbostat 

cor CPU   GHz  TSC time

        1.81 1.90   5**

  0   0 1.81 1.90   5**

  1   1 1.81 1.90

  2   2 1.80 1.90

  3   3 1.81 1.90

EDIT: It seems like Trinity series processors have it's own BAPM switch in source. It's located in drivers/gpu/drm/radeon/trinity_dpm.c and it's (at the time of writing) enabled only for MSI boards, because of stability issues. To enable it, open this file, find this line:

    pi->enable_bapm = false;

Change false to true, then compile and install new kernel. You should get ~~a kernel panic~~ boost working now. Keep in mind that CPU power is also shared with GPU, so you'll almost never get the highest frequency available if GPU is also in use.

Answers

Only authorized users can answer the question. Please sign in first, or register a free account.

breadoules

Add To Favorites

Follow

Total Points: 212

Total Questions: 118

Total Answers: 120

Location: Dominica

Member since Mon, Jun 22, 2020

4 Years ago

answered 1 Year ago alkeake · Accepted Answer

[Update 2015]

Starting with Linux 3.16:

The value for bapm can be provided as a module parameter (see here).

The value of bapm is set to 1 by default for Kaveri, Kabini and desktop Trinity, Richland systems (see here), resulting in Turbo Core being enabled.

Which means that with Kernel 3.16 or later, Turbo Core should work right out of the box with radeon in many cases.

If you run Debian, or if you run Ubuntu and the above is not true for you, please refer to How to set up a Debian system (focus on 2D or console/server) with an AMD Turbo Core APU for maximum energy and computing efficiency?

[Update 2014-Aug-07]

I published a detailed analysis over at unix-stackexchange.com which contains (at least for me) a few surprises. Such as the ondemand governor potentially being worse than performance (at least with the Richland A10-6700) and fglrx apparently using some questionable tricks to improve short-term boosts, abruptly having to compensate for that in the long run. I'm glad I had a closer look at this.

[Original Answer]

I am in the process of configuring a system with an A10-6700 using Ubuntu 14.04 Server.

The output of cat /proc/cpuinfo does not, and will not, reflect any processor boost. The boost is an autonomous decision on the chip based on potentially complex criteria. Read about it here.

The output of cpufreq-info will not show available boost frequencies, but the output of cpupower frequency-info will. For the A10-6700, it should show 4300 MHz (Pb0), 4200 MHz (Pb1) and 3900 MHz (Pb2) in addition to the regular 3700, 3400, 2700, 2300 and 1800 MHz.

The output of cpufreq-aperf (after modprobe msr) is comparable to the output of cpupower monitor. The frequencies shown by these tools are supposed to reflect boosts.

With the standard video driver or even the radeon module in the kernel, I did not succeed to get the A10-6700 to use a boost frequency (using stress --cpu 2).

Withe the ATI/AMD fglrx module loaded into the kernel, the processor will boost two cores permanently (and apparently also the remaining cores temporarily). This was tested without X being used or even configured.

Remember, the decision for or against the boost is made by the APU autonomously. My guess at this point in time is that the APU cannot decide whether its current power dissipation leaves room for a boost unless the integrated graphics unit is adequately initialised.

The video output with the standard module differs from the output when fglrx is loaded. Therefore, we can only guess whether the kernel can put the APU into the presumably required "smart" (power loss aware) mode by itself, while still using the VGA module.

Until this gets sorted out, I will simply load fglrx. The downside is that building it it requires a lot of stuff which you don't want on a server and which needs to be deleted after compilation.

Also, fglrx is almost 9 MB in size. It's not a problem but knowing that it's mostly dead freight in this case doesn't make me too happy.

[Edit 2014-Jul-24]

I now reported this bug.

I noticed another bug has also been reported: If in fact you use fglrx and try to use the Dynamic Power Management mode.