AMD Driver Autodetect is a straightforward, yet efficient application designed to detect drivers for your AMD video card, as well as to download and install new updates on the spot. Dec 31, 2014 5) Since the latest software/drivers were not found with the AMD Driver Autodetect, I decided to manually download the latest Graphics card drivers from the AMD website for my graphics card and for my operating system (Windows 8.1 64-bit).
Finally succeed to mount a Aktio or a Sonnet Box (both with Nvidias GTX1080Ti) into my MBP mid 2015. Hot plug works but to disconnect we have to shutdown the machine.
Trying to eject from the widget will crash the machine. The method was to install new Nvidia web drivers for 10.13.4, Cuda drivers and NVDAEGPUSuppor-v8.zip. After exchange the AppleGPUWrangler.kext by the version from 10.14.4 beta 1. Finally use the maceditor suggested commands to repair permissions. $ sudo chown -R root:wheel /System/Library/Extensions/ $ sudo chmod -R 755 /System/Library/Extensions/ $ touch /System/Library/Extensions/ the last line was only successful by using sudo, like this: $ sudo touch /System/Library/Extensions/ After I've just restarted and things work as they used to work in 10.13.3. The only advantage now is that we can use the eGPU without the need to logout and log in again. I've noticed that there's some unsuitability with some apps, like the Luxmark benching app.
Agisoft Photoscan also works just fine but, after quit the 'send crash log to developer appears. However, despite this small issue, the app works just fine. Blender also sees the Nvidia and can use CUDA without problem. Waiting for the refining but, for now, seems to be fine.
I was able to simultaneously connect 2 NV 1080Ti eGPU's to the MBP 2015 but they seem to create some instability in several apps. They are recognized by the system but the do not seem to be able to work together. This is probably more a limitation of the TB2 controller of the MBP 2015 because, even before, it was impossible to have 2 eGPU working in this machine. Posted by: maceditor Could you test the latest commit at your leisure? Specifically the backup part (you can simulate an old version of macOS by modifying the manifest.wglr file, then attempting re-patch). I made another pull request on the GitHub repo. It seems to function really well.
I noticed you removed the auto reboot (maybe a silent option would run that but a normal patch could prompt the user Y/n defaulting to yet to reboot on user input) The sea is probably way simpler to implement and may for obvious reasons be more reliable in enforcing structure with the unknown. Changing the.wglr file to have 10.13.3 made it realize it needed to update the backup after that fix I committed. Re-running the patch seems to handle the existing backups.
The new version politely ignores the previous backup methods. I also validated the backup contained the TB3 string in its hex like the current stock binary has. I didn't check this between executions but changing the macOS version line to 10.13.3 that forced it to rerun could theoretically overwrite a valid backup however uninstalling and rerunning the patch seems to keep the correct version in the backup directory (given you set remove previous backups or modify the macOS version in the.wglr file so that the patch recreates the backup from an uninstalled(or unpatched) binary). All that to say if it is used without user modifications, it will be safe.
If you try to use it on an unsupported macOS version like 10.11 or something off like that it doesn't handle that. Everything else it handles and seems to check reliably.
Posted by: maceditor already merged it ? nice catch that one haha. The broader checks for different os versions and hardware will be handled in a bigger ‘super script’ so to speak, courtesy of. Also the.wglr extension and superuser permissions are in place to ‘dissuade’ people from modifying it.
Not that they can’t. Now we could check that too, but that’d be a bit much lol. I look forward to the 'Super Script' I had a feeling was working on something along those lines. Yeah if the end user modify's those backups lets just assume they either know what they are doing or are willing to accept fault. There is somewhat of a reason Apple puts a mask on all the magic behind the curtains and discourages everyone from touching things they shouldn't. Doesn't stop those who know there way around things.
I did not need to use the script. After the installation of the needed Nvidia files (Driver, CUDA and NVDAEGPUSupport-v8, I've just used the AppleGPUWrangler.kext from 10.13.4 beta 1 and did the maceditor suggested commands for permission repair. Closing the lid shuts the monitor light but will deactivate password protection. After reopen the lid the computer will be immediately available. Shutdown works fine but logout does not work. If we try to logout the monitor goes black and stays that way until a forced restart. All the rest it's working just fine.
HotPlug for itself it is a great add because sometimes I have the computer on, with work ongoing and now I do not need to log out and in anymore. Posted by: rhenriques11 HotPlug for itself it is a great add because sometimes I have the computer on, with work ongoing and now I do not need to log out and in anymore. Thank you for answering rhenriques11 and Sovereign. In fact, the hot plug feature was the main reason for me to test this thing because I never boot my mac (only for booting with bootcamp), just open/close the lid. I don't want to have to remember to shut down and boot every time (and switch on/off the eGPU PSU).
With hot-plug however I would plug the eGPU only when I wanted. I'll try if I can, anyway. Always is good to have the option to use the 1070 if I need it. Just upgraded my MacPro late 2013 and things are also working just fine with 2 external NV Geforce 1080Ti. The first time the machine start and we connect the GPUs, CUDA seems not to work correctly with some apps. After reinstalling CUDA and restart things are very smooth.
As says, do not distribute beta.kext from apple because is not allowed. I recommend you to first backup your machine (I've used Carbon Copy Cloner) and then upgrade.
This way you can get the AppleGPUWrangler.kext from the previous system and use the purposed methodology. As a final Conclusion, It's hard to understand why apple did this modification into AppleGPUWrangler.kext. It is not that hard for Apple engineers to automate the modification that we are doing for ourselves. It leads us to think that TB/TB2 deliberate blockage it's a planned obsolescence for sales increase.
Which is a very bad way of doing business and mines costumers trust. Posted by: rhenriques11 This is outstanding!! Check what appears in the menu bar eGPU widget if we startup with the eGPU connected to the monitor instead of the internal GPU! The name is recognized correctly. Dual screen output working? I have a Monitor directly connected to the CPU and another connected to one of the eGPUs.
Both working good. The other external eGPU is just for processing. Agisoft photoscan, for instance, can use all available GPUs for calculations. Here's how processing power is distributed accordingly to the processing power of each GPU. Notice how the once (in 2013) top notch AMD Fire Pro 500 now pales when compared to the Nvidias. Posted by: Sovereign It is in /System/Library/Extensions/AppleGraphicsControl.kext/Contents/PlugIns/ Thank you, Sovereign.
From my experience, the AppleGPUWrangler.kext from 10.13.3 seems to be a lot more stable than the one from 10.13.4 Beta1. You will notice, in some cases, a slower startup, but the overall stability seems to be a lot better after. I was experiencing, in the beginning occasional hangs during shutdown and, sometimes, the second eGPU failed to mount. Now it's solved with AppleGPUWrangler.kext 3.16.2 from MacOS 10.13.3.
I had the same problem. I ended up restoring the kext from 10.13.3 via Time Machine and rebuilt the kext cache manually from the command line. I think there might be an issue with one of the scripts that I used from here that did either incorrect tb1/2 detection patching and/or incorrect kext cache regeneration. I know I used the early version that would overwrite the original backup, so maybe that was part of my problem too. I’ve read In one of the posts of a user that used the original script prior to tb1/2 auto detect having to pass in tb2 on a tb1 Mac mini to get it working. With that switch removed and now relying on auto detect being the only option.
I hope if the above is true the guys providing us help can fix this before the next OSX update reverts these changes. Either way, thanks to all the folks in this forum for all the help and guidance this week.