Editing AGPM.kext

Guide moved to new site:

http://www.rampagedev.com/?page_id=311

61 Responses to Editing AGPM.kext

  1. Pingback: AGPM.Kext Guide Live | rampagedev

  2. Arkanis says:

    Hello Andrew, thank you for the guide, can this guide be applied to the GTX 680 family under Mavs with Mac Pro 6,1 definition ?

    • rampagedev says:

      Heuristic ID is different so in theory you could if you identify the older cards Heuristic ID. The code from the iMac with the GTX 680M should work.

      • Arkanis says:

        Andrew,

        Used the code from iMac 13, replaced the ID and done everything by your guide. PM seems to be working for GPU, GPU Mem and ROPs. But my score on Cinebench is only 48.52 which is under the GT 650M… not normal.

        Could you please enlighten me on which settings should I tweak to test results ?

        Thank you.

      • rampagedev says:

        Read what I posted:

        “I still do not have a full understanding on what values to change to improve performance so I am asking you the users to do some testing on my behalf and see what performance gains can be obtained.”

      • Arkanis says:

        Comparing with specs I can see that the Memory is reaching it’s peak speeds, but not the GPU who stays at 800Mhz maximum when the base clock is 1.1Ghz

  3. DKMN says:

    Is this framework a decent place to start with the NVIDIA GTX 670? Or do you have any values to use as a starting point?

  4. flmmkr says:

    Will this work with the Titan being that it’s so similar to the 780 which is in 7xx series?

  5. Alex Peacock says:

    Tested with GTX Titan. GPU clock changes like it should but Memory remains stuck at 6.01GHz. Also it doesn’t seem to be controlling the fan speed of the card correctly. HWMonitor shows the fan speed increasing all the way up to 2400RPM when under full load but I can hear that the fan isn’t actually spinning that fast.

  6. Rob van der Lans says:

    Interesting, is there also also a working solution for ATI cards like AMD Radeon HD 5870, probably yes, but how to figure out al these values compared to what? I have no clue.

    • rampagedev says:

      No idea what the heuristic ID would be if there is any.

    • Rob van der Lans says:

      Found that these ATI device ID’s are in (default) and it seems that ATI doesn”t make advantage of heuristic ID.

      iMac12,2
      Vendor1002Device6720 [AMD Radeon HD 6900M Series]
      Vendor1002Device6722 ?
      Vendor1002Device6740 [AMD Radeon 6600M]

      MacBookPro8,2
      Vendor1002Device6741 [AMD Radeon 6700M Series]
      Vendor1002Device6760 [AMD Radeon HD 6400M Series]

      Not in (Default)
      Vendor1002Device6898 [ATI Radeon HD 5800 and HD 6800 Series]

      • rampagedev says:

        Shame.I guess only the new ones do? Might be able to get away using them but not sure.

      • Rob van der Lans says:

        For info [in case of ATI HD 5870]

        Find out that ID 6722 is [AMD Radeon HD 6800] passed the content from Vendor1002Device6722 >>>> to Vendor1002Device6898 under (Mac-F60DEB81FF30ACF6). Set flag logIGPU [ON] in AICPMPI but unfortunately logIGPU will not work and IPG won’t show GPU P-states either. Annyone suggestions how to detect these GPU P-states.

      • rampagedev says:

        Thanks for the feedback.

      • Rob van der Lans says:

        All things set low temps but just don’t know how to detect (with which tool) to see those GPU P-states working, AICPMPI has a function (flag set logIGPU) to measure GPU P-states but in latest release v3.3 from @Pike this function is disabled. And I suppose IPG only detects GPU states from Intel chipset. (lol) 😉

      • Rob van der Lans says:

        Kernel log: [AGPM Controller] build GPUDict by Vendor1002Device6898

      • rampagedev says:

        Loaded at least 🙂

  7. Bill says:

    my gtx470 remains stuck at 50-67-270! no power up no another state…

  8. maroderdachs says:

    I followed the guide for a 760 GTX, but for some reason I can’t find AGPMEnabler in IOreg after rebooting. So I’m not sure whether it worked or not. Is there another way to check?

  9. maroderdachs says:

    Ok, after generating SSDT AGPMEnabler now shows up in IOreg. But now I get an AGPM: unknown device. I am 100% certain I followed your guide step for step. Crosschecked 1000 times.

    What I find weird, is that the ioreg doesn’t look anything like yours and also the control ID doesn’t fit (even so I added it correctly): https://www.dropbox.com/s/efoaia3fc8r5g1o/Screen%20Shot%202014-03-26%20at%2017.36.55.png

    Any ideas?

    • maroderdachs says:

      Device ID is wrong too – should be Vendor10deDevice1187, but it doesn’t show up.

    • rampagedev says:

      Device ID for you card is missing if it is unknown.

      • maroderdachs says:

        Thanks, I found the problem: I changed the smbios with all necessary information from iMac 14,2 to MacPro 6,1, but somehow there seems to be something cached. The Information showing in my IOreg is still from the iMac 14,2 part of the plist. I’ll have another look at this issue at the weekend.

  10. I tried this with a GTX 780 Ti (000a), but no discernible change in OpenCL or CUDA performance.

    • rampagedev says:

      Thats not the point. This needs to be tested and verified. Find the values which DO effect performance. Your completely missing the point of this guide. I even state that I do not know yet what values effect what.

      • I understood that, but I did not have time to do that. I did have time to test what you had and report back. I guess you think no information is better than some.

        Additionally, is a GTX 780Ti even being throttled in OS X? Other than the OpenCL issue where it gets stuck in second highest state?

      • rampagedev says:

        Not sure. The Retail drivers are the only ones that work properly so… could be a early adoption issue with the Ti. Nvidia should have that all worked out soon. Have you tried 10.9.3?

  11. 10.9.3 – 13d33 Nvidia drivers have changed 2 screens with gtx670 now working again no rollback needed.
    I have been playing with AGPM and appear to have stepping ((6) using 1189:11a3 Mac-FC02E91DDD3FA6A4 settings) BUT whenever I use Preview it pegs to flat out and does not return. I haven’t been able to figure out why?

  12. DKMN says:

    Sorry, figured that out just before I saw your email, using:
    iMac13,2:
    Mac-FC02E91DDD3FA6A4
    as a source, and
    MacPro6,1:
    Mac-F60DEB81FF30ACF6
    as the target I am configuring. Copied over the contents and put in the device ID for my cards 0x0fc6 (GTX 670).
    Still haven’t got AGPM to show up in IOREG with your latest SSDT and plain vanilla SSDT-1 from Pike’s latest script (which works otherwise).
    Tried a few permutations of GFX0, 1, 2 given that I have 2 cards (in NPE3 and NPE7, respectively).

    Any thoughts on what to try next?
    D

    • DKMN says:

      Huh, just had another thought. I don’t see X86PlatformPlugin, either.
      I wonder if using the vanilla script for Sandy Bridge is not generating an SSDT-1.aml that will load AGPM…

  13. Omar Mostafa says:

    I’ve done everything you said here, but I can’t find any AGPM in my IOReg.
    but the weird thing is, in System Profiler it says in Extensions that AGPM.kext is not loaded, what should I do?
    If you need me to add anything, just ask.

  14. Omar Mostafa says:

    Here’s my IOReg if you need to check anything: https://www.mediafire.com/?bylczc1xl5eod4v

  15. Comrad8g6 says:

    Dev: what is this App that you are editing info.pls in? is it x code?

  16. crushers1982 says:

    Does this work on the Yosemite on iMac13?I have GTX580 and I have problem after wake with my GPU.Thanks in advanced..

  17. crushers1982 says:

    whether it is possible to enable GPUPM!!!Did that follow your posts for AGPM and to knock the same numbers?

  18. hey rampagedev, is this still applicable for 10.10 and the nvidia maxwell cards (gtx 970, specifically)? I’m kind of confused and stuck, getting the dreaded “agpm unknown platform” message upon bootup. Not sure whether the nvidia web driver is handling the power management or there simply is none. Cuda-Z appears to show only a singular, high clock frequency. Cheers!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s