Jump to content

AMD issues with nVidia cards running Lion 64-bits/Mountain Lion [DON'T ASK HELP HERE, THE HELP TOPIC IS THERE FOR YOU]

theconnactic

By theconnactic,
in OS X 10.8 (Mountain Lion)

 Share
175 posts in this topic

Recommended Posts

spakk
spakk
Posted
Posted

@SS01: didnt work for me! Only two more errors!

 

 

So, I've made in the last week, some drivers adapt on my system:

First I reinstalled Mountain Lion and all non-relevant drivers for my hardware were placed away in a folder on desktop.

 

 

Then I installed the latest graphics driver from Mountain Lion 10.8.3 as follows:

 

the AppleGraphicsPowerManagement.kext (with plugins!) was for iMac13, 2 ​​with my

VEN ID and DEV ID modified. The PlatformSupport.plist was extended for iMac13, 2.

 

Driver Update from Mountain Lion 10.8.3

Geforce.kext (this driver is not changed and has been installed.)

GeForceGA.plugin (this driver is not changed and has been installed.)

GeForceGLDriver.bundle /System/Library/Extensions/GeForceGLDriver.bundle/Contents/Mac OS

The old libclh.dylib was with that file here http://rghost.net/39181502 exchanged!!

 

GeForceVADriver.bundle this driver is not changed and has been installed.

 

NVDAGF100Hal.kext (This driver has been adapted with my VEN ID and DEV-ID and has been installed.)

NVDAGF100Hal.kext (This driver has been adapted with my VEN ID and DEV-ID and has been installed.)

NVDAGK100Hal.kext (this driver is not changed and has been installed.)

NVDANV50Hal.kext (this driver is not changed and has been installed.).

NVDAResman.kext (This driver has been adapted with my VEN ID and DEV-ID and has been installed.)

NVDAStartup.kext (This driver has been adapted with my VEN ID and DEV-ID and has been installed.)

 

OpenGL.framework (is not changed and has been installed).

 

GPUSupport.framework (is not changed and has been installed).

 

OpenCL.framework \ Versions \ A \ OpenCL this framework seems to be modified see Netkas website (adaptation of OpenCL with the hex editor).

 

Then I installed this driver:

ATY_Init.kext adjusted with my VEN ID and DEV ID and installed.

 

at the end with the Disk Utility program repariern rights.

the smbios.plist for iMac13, 2 ​​has been placed in the folder Extra

 

Please check this if it works for you.

  • Like 1
Link to comment
Share on other sites
misterfrista
misterfrista
Posted
Posted

Coming from late 2012 update 10.8.2 your solution didn´t work for me!

 

With modified and unmodified 10.8.3 drivers I get instant reboot right after kext loading!

Maybe the difference is fermi - kepler as I need to modify NVDAGK100Hal.kext instead of NVDAGF100Hal.kext to get GTX660 to work!

 

I´ll try to modify NVDAGF100Hal.kext later and see if it brings any difference!

 

Anyway good work :)

 

Edit: Sorry it seems my hard drive is busted :-\ Need to run diagnostic tool so most likely my issues aren´t related to nvidia!

 

Edit2: Sooo my system is running again :D

 

Did the instructions from above, except the AGPM plugins (I dont know where those should be?!)

 

My system boots up as usual but doesnt start the desktop! As I don´t know if my modifications are correct I uploaded the modified plists with the windowserver log files!

 

All I can interpret from them is that the problem with the 10.8.3 is different to the one from 10.8.2!

 

Maybe someone has time to take a look in the plists if I done it wrong, my device id is 11c0:10de just for your information!

 

 

Greetings and good luck!

Modified plists.zip

Link to comment
Share on other sites
misterfrista
misterfrista
Posted
Posted

Yeah little but sweat success!

Did spakks Guide but with some modifications: I used an existing 12,2 imac profile for AGPM (maybe there is space for tuning) and deleted the whole block IOMatchProperties ( i think that is the Name as I am currently not at Home, anyway i will look it up tomorrow!) in nvdagk100hal! With the block in the plist i get a White Screen, without i See my driver loaded up two times and i get successfully to my Desktop and kextstat Shows my drivers as loaded! Furthermore i dont use aty_init.kext

 

There is a little Difference to the Version with deleted nvidia kexts : the Game Chess doesnt lag that hard and if i open a New tab in safari, the System doesnt freeze as it did before!

 

The System now detects my graphicscard correct with correct vram!

 

I will take a backup and play around with Andys amdautotool, maybe patching the binarys gives me access to qe/ci as I cant See an error in the logs and Calls for Intel CPU could hold back the potential of the driver maybe?!

 

I am open for ideas ;-)

 

 

Edit: Now successfully patched:

-/S/L/E

-/S/L/CoreServices

-/usr/lib

-/S/L/Frameworks

-/S/L/PrivateFrameworks

 

It brought me some improvment in network stability! My network connection with lnx2mac driver was a bit wonky but now it runs fine!

I played around a little bit with the kexts and I am able to boot into my desktop with loaded nvidia kexts without the need to patch AGPM!

 

Although the drivers are loaded and the card is recognized benchmarks like luxmark dont recognize the card with OpenCL! No need to tell OpenGL isn't running!

I tried to patch according to netkas guide but it seems something has changed in 10.8.3 opencl framework!

 

Edit2: I´m not that much into plist editing but would it be possible to inject a value into one of the nvidia plists to force load opengl extensions?

I found this https://3c.web.de/mail/client/dereferrer?redirectUrl=http%3A%2F%2Fwww.insanelymac.com%2Fforum%2Ftopic%2F283889-new-mountain-lion-1082-updater-late-2012-models-nvidia-drivers%2F&selection=tfol11a2f55e2046b4fe , there is a photo of a disassambled binary of the nvidia drivers with features/flags of the driver. Maybe something like NVDAgl=yes.. i don't know just thoughts ;)

  • Like 1
Link to comment
Share on other sites
Shaneee
Shaneee
Posted
Posted

I have made some progress on ML. I no longer get the black screen it's grey now, cursor comes and goes but no white square in top corner. I played around a bit more and I have got to the login screen with all the nVidia Kexts still in place. Problem is when I enter my password it goes right back to login, I don't see the desktop. I will keep at it and keep you all updated here.

 

Edit: If I enable Auto Login then I get the black screen inbetween the grey screen showing with cursor showing the com.apple.coreservices.appleid.authentication error.

 

Edit2: I am now recieving a new error while its stuck in this login loop with Auto Login on... "64 bit opemu wrong instruction"

  • Like 1
Link to comment
Share on other sites
theconnactic
theconnactic
Posted
  • Author
Posted

No, because oz ran a night-long testing session with the opemu, and found no issues at all. I'll take a look at the source code - but i did it already and found no wrong code. Perhaps it's not the opcode emulator itself, but something in the drivers that's not going along with it.

Link to comment
Share on other sites
theconnactic
theconnactic
Posted
  • Author
Posted

It's 64-bit kernel mode support, Shane! It's implemented the wrong way in the kernel. Thus all those issues, like nVidia and ATI bugs, crashing apps etc. They're consequences, different manifestations of a single problem. Solve the problem, all the other issues will disappear like magic.

 

Notice that in 32-bit kernel mode, even running 64-bit apps, these things don't happen.

 

Best regards!

Link to comment
Share on other sites
Andy Vandijck
Andy Vandijck
Posted
Posted

No, because oz ran a night-long testing session with the opemu, and found no issues at all. I'll take a look at the source code - but i did it already and found no wrong code. Perhaps it's not the opcode emulator itself, but something in the drivers that's not going along with it.

There is a possibilty:

The opcode emu works fine but the drivers use an instruction that isn't implemented in it.

SSE4.X instruction for example...

We could mod the kernel to show the instruction it doesn't emulate...

  • Like 3
Link to comment
Share on other sites
SS01
SS01
Posted
Posted

There is a possibilty:

The opcode emu works fine but the drivers use an instruction that isn't implemented in it.

SSE4.X instruction for example...

We could mod the kernel to show the instruction it doesn't emulate...

 

It can't hurt to try. IIRC K10+ supports SSE4.2 without emulation, however.

Link to comment
Share on other sites
Andy Vandijck
Andy Vandijck
Posted
Posted

 

 

It can't hurt to try. IIRC K10+ supports SSE4.2 without emulation, however.

You mean SSSE4a.

Thar is far from full SSE4.2...

  • Like 1
Link to comment
Share on other sites
SS01
SS01
Posted
Posted (edited)

You mean SSSE4a.

Thar is far from full SSE4.2...

 

So we need an SSE4.2 emulator?

 

!ping Sinetek

 

Or we could try porting the open-source nouveau driver and compile it without any offending instruction sets.

Edited by SS01
Link to comment
Share on other sites
SS01
SS01
Posted
Posted (edited)

Orig. by theconnactic:

...the kernels compiled with the current patch set don't play well with nVidia cards (and some ATIs) when running in 64-bit kernel mode on AMD CPUs (and only AMD: the same kernels will do just fine in, say, a Pentium D).

Pentium D does not support SSSE3, SSE4, or SSE4.2

 

I don't think the problem is here

Edited by SS01
Link to comment
Share on other sites
theconnactic
theconnactic
Posted
  • Author
Posted (edited)

You didn't get my point, SS01: Pentium Ds don't support SSSE3, SSE4 or SSE4.2, yet as far as i know, they run Lion 64-bit or Mountain Lion just fine with any graphics card without issues with the same patched kernels that make our system unusable with nVidia and some ATI cards. Obviously we need more relates to confirm this, but if it's really true, what does it prove? Well, if not anything else, it does prove that the opcode emulator is not flawed or malfunctioning in any way. That's what i'm trying to say: if it's true that Pentium Ds run the patched kernel without issues (and i guess it is) let's stop for good to aim the wrong target: our issue is somewhere/something else. And this somewhere/something is perhaps related to 64-bit support, but not with the opcode emulator, since 32-bit kernel mode runs well where it can (Lion), even supporting 64-bit processes (thanks to the opcode emulator) with few or even none of the issues we have with 64-bit kernel mode.

 

Best regards!

 

P.S.: by the way, i'm running Mountain Lion 10.8.3 (vanilla kernel, patched with Meklort's Chameleon autopatcher module) flawlessly on my 1201n, which is powered by an Atom 330 dual-core CPU, which by its turn doesn't support SSE4.2.

Edited by theconnactic
Link to comment
Share on other sites
misterfrista
misterfrista
Posted
Posted

Tell me if i am wrong but from my point of view the difference between Pentium Ds and AMD tells me: The Problem could lay in the way AMD and Intel handle the same thing in different ways!? That would explain why missing imstruction sets dont matter with Atom and Ds!

  • Like 1
Link to comment
Share on other sites
spakk
spakk
Posted
Posted (edited)

if we do not know the reason why the nvidia graphics card is not working correctly, the problem is probably on SSSE3emu

 

we should try as our Linux friends using the nouveau open source for use nvidia drivers for OSX , see Nouveau project:

http://linux.slashdo...ll-geforce-gpus

http://nouveau.freedesktop.org/wiki/

http://www.h-online.com/open/news/item/Open-source-Linux-driver-supports-3D-acceleration-with-all-GeForce-GPUs-1776089.html

 

is that possible?

Edited by spakk
  • Like 1
Link to comment
Share on other sites
instant idiot
instant idiot
Posted
Posted
is that possible?

Technically, it's quite possible that someone could write a completely perfect nVidia driver for OS X from scratch, and they might even be able to use some of the code from nouveau, but, even on Linux, nouveau is nowhere near a replacement for the proprietary drivers (Hardware acceleration is very limited.).

Link to comment
Share on other sites
SS01
SS01
Posted
Posted (edited)

Technically, it's quite possible that someone could write a completely perfect nVidia driver for OS X from scratch, and they might even be able to use some of the code from nouveau, but, even on Linux, nouveau is nowhere near a replacement for the proprietary drivers (Hardware acceleration is very limited.).

 

A while back when Steam for Linux was released, I put Fedora on a 9800GT-based rig using the nouveau driver. I must say, every game I tried played just fine. We would likely have to port mesa in addition to nouveau, however.

 

Some documentation:

 

http://nouveau.freed.../wiki/FrontPage

 

https://wiki.archlin...dex.php/Nouveau

 

http://www.mesa3d.org

 

irc.freenode.net #nouveau

 

it might be possible to install the proprietary FreeBSD driver on OS X? http://www.freebsd.org/doc/en/articles/compiz-fusion/nvidia-setup.html

Edited by SS01
Link to comment
Share on other sites
 Share
Go to topic listing
×
×
  • Create New...