euqlaog

curl -o ~/Desktop/automate-eGPU.sh https://raw.githubusercontent.com/goalque/automate-eGPU/7192c79bda052bff92428700b59b652b30e6ab4d/automate-eGPU.sh

This is a known issue. I’m discussing with @josho and @Simurgh5 about this, and will fix it. In the meantime, please use v0.9.6 for Nvidia dGPU equipped Mac (uninstall v0.9.7 with -uninstall option).

Only one day of waiting, now the web driver is available.

OS X 10.11 is not selectable from the dropdown at Nvidia's web site, but the script found it, this is not beta:[FONT=Menlo]*** automate-eGPU.sh v0.9.7 - (c) 2015 by Goalque ***[/FONT] [FONT=Menlo]-------------------------------------------------------[/FONT] [FONT=Menlo]Detected eGPU[/FONT] [FONT=Menlo] GM204 [GeForce GTX 980][/FONT] [FONT=Menlo]Current OS X[/FONT] [FONT=Menlo] 10.11 15A284[/FONT] [FONT=Menlo]Previous OS X[/FONT] [FONT=Menlo] 10.10.5 14F27[/FONT] [FONT=Menlo]Latest installed Nvidia web driver[/FONT] [FONT=Menlo] Version: 346.02.03f01[/FONT] [FONT=Menlo] Source: 3rd Party[/FONT] [FONT=Menlo] Install Date: 9/26/15, 4:57 PM[/FONT] [FONT=Menlo] [/FONT] [FONT=Menlo]You are running official Nvidia driver.[/FONT] [FONT=Menlo]Checking IOPCITunnelCompatible keys...[/FONT] [FONT=Menlo] [/FONT] [FONT=Menlo]Missing IOPCITunnelCompatible keys.[/FONT] [FONT=Menlo]Mac board-id found.[/FONT] [FONT=Menlo]Searching for matching driver...[/FONT] [FONT=Menlo] [/FONT] [FONT=Menlo]Driver [346.03.02f01] found from:[/FONT] [FONT=Menlo]http://us.download.nvidia.com/Mac/Quadro_Certified/346.03.02f01/WebDriver-346.03.02f01.pkg[/FONT] [FONT=Menlo]Do you want to download this driver (y/n)?[/FONT] [FONT=Menlo]y[/FONT] [FONT=Menlo] % Total % Received % Xferd Average Speed Time Time Time Current[/FONT] [FONT=Menlo] Dload Upload Total Spent Left Speed[/FONT] [FONT=Menlo]100 68.9M 100 68.9M 0 0 3500k 0 0:00:20 0:00:20 --:--:-- 3577k[/FONT] [FONT=Menlo]Driver downloaded.[/FONT] [FONT=Menlo]Removing validation checks...[/FONT] [FONT=Menlo]Modified package ready. Do you want to install (y/n)?[/FONT] [FONT=Menlo]y[/FONT] [FONT=Menlo]installer: Package name is NVIDIA Web Driver 346.03.02f01[/FONT] [FONT=Menlo]installer: Upgrading at base path /[/FONT] [FONT=Menlo]installer: The upgrade was successful.[/FONT] [FONT=Menlo]installer: The install requires restarting now.[/FONT] [FONT=Menlo]Checking IOPCITunnelCompatible keys...[/FONT] [FONT=Menlo] [/FONT] [FONT=Menlo]Missing IOPCITunnelCompatible keys.[/FONT] [FONT=Menlo]IOPCITunnelCompatible mods done.[/FONT] [FONT=Menlo]SetIOPCIMatch() set device ID 0x13C010DE in /System/Library/Extensions/NVDAStartupWeb.kext/Contents/Info.plist[/FONT] [FONT=Menlo]All ready. Please restart the Mac.[/FONT] Works great.

PCI device ID appending now allows to mix different family eGPUs, but the drawback is lost Nvidia dGPU acceleration. Fortunately it’s easy to uninstall v0.9.7 and switch back to v0.9.6. It was expected that Iris Pro won’t take over internal screen acceleration automatically. You can type "system_profiler SPDisplaysDataType" to see whether the internal screen is controlled by the Iris Pro or 750M. Even if Iris Pro worked properly, completely disabling 750M when not using the eGPU wasn’t my intended goal. I’ll send you PM so we can debug this further.

@danielbrown941: Seems that you installed the script with -url option. OS X 10.11 web driver is beta. What happens if you try the following: 1) sudo ./automate-eGPU.sh -uninstall 2) Restart 3) sudo ./automate-eGPU.sh -skipdriver 4) Restart

You could be the first to try, on September 30.

The problem is that you still have 750M activated for internal screen (maybe due to gfxCardStatus, please uninstall it) but as now NVDAStartup.kext is isolated to match only Nvidia eGPUs, the 750M won’t give any acceleration which appears as very slow internal screen. So you have to find a way to switch back to Iris Pro only (currently, the script doesn't change the gmux state). Your gpu-policy value <01> seems to be correct. What kind of custom connector? We cannot rule out that part yet. I’ve learned this a year ago. At that time, I tried also two PSUs, back-powered and non-back-powered risers - a wrong power configuration resulted in immediate system crash when launching Unigine Valley. Do you have the same problem in Windows Boot Camp? If "sudo ./automate-eGPU.sh -clpeak" command causes system crash too, the power issue is double confirmed.

@Muadi: Please read the first post of this thread how to download the script. Not necessary with the Nvidia eGPU. If you have an AMD eGPU, it should work fine with the M370X and you can mix eGPUs any way you want.

What's new in 0.9.7: SetIOPCIMatch() function which sets and appends device IDs (both the AMD and Nvidia) Automatic NVIDIA Driver Manager uninstalling Minor bug fixes 1) Download the new script 2) chmod +x automate-eGPU.sh 3) sudo ./automate-eGPU.sh -uninstall (Nvidia Driver Manager should prompt restart) 4) sudo ./automate-eGPU.sh 5) sudo ./automate-eGPU.sh -a 6) Restart the Mac Step 5 is optional. Let me know how it works with a 15” MBP 750M. Keep “Automatic graphics switching” turned on in Energy Saver settings.

Yes. Always use a single power source, and if you have to use a powered riser, you have to know whether it is back powered or not. Otherwise, you can damage the system.

@Simurgh5, @Mark, @josho, @soularchive and other 15” Nvidia dGPU equipped MBP owners. I got an idea to modify the web driver package purely for your external GPU, so that dGPU (such as 750M) will never utilize the web driver. This will not take long.

El Capitan GM is not the final release. The script will download the new official driver automatically when it’s available from Nvidia. For 10.10.5, the package was available in the next day. In the meantime, you can run the script with [-skipdriver], meaning that Nvidia web driver downloading part will be skipped and it modifies the previous package. Between OS X’s Codename change, the only way is to use [-url] option and beta driver. Someone at Nvidia leaks beta drivers purposely. The script downloads automatically only those official ones. AMD cards are easier since drivers are built-in by Apple and support for Fiji is coming. Official drivers for older Nvidia architecture is still there, so if you have Quadro NVS 295 or GTX 780 for instance, just run the script with [-skipdriver] option. Please don’t use Nvidia’s update prompt for installing the new driver. The preferred way is to install is just by running the script again.

I hope you are not using two PSUs. Maybe you could be in touch with the 750M users who don't have these problems. I would try setting the external monitor as primary. When successful booting procedure is found, it can be automated by the script.

@yocyoc Please give the command you typed and the output of the script when you uninstall. Your OS X might need reinstallation.

@soularchive: Thanks for the feedback. The best information you can give for troubleshooting is the script’s output. From the given information I picked up these: OS X 10.10 was released almost a year ago, when GTX 980Ti didn’t exist. Maxwell support was just in the beginning and HDMI port was not supported at all. The script downloads the correct old driver that matches the operation system build version. OS X 10.10.4 was the first build in which Nvidia decided to give support for dGPU graphics. The problem is not with the -a mode, that is confirmed by two 15” MBP 750M users. The latest OS X 10.10.5 is recommended. Avoid hot-plugging eGPU displays and run benchmarks on the eGPU screen. Most of the powered risers are bad quality and cannot provide ~40W-75W (depending on the GPU) for the x16 connector, plus ~15W-20W for one TB controller & PCIe board as back powered. Throughput of the DC jack is said to be max 120W, so you can use a barrel adapter from the ATX PSU 8-pin/6-pin.

“Selecting iGPU works” meaning that my guess was correct when “automatic graphics switching” is turned on + “integrated only” selected? So you can wake up the Mac without issues? Maybe a wrong combination causes unsuccessful booting too. In that case, forcing iGPU and checkbox checked before the user logs in would be an automated solution, or an event handler when the Mac is about to go sleep/restart.

I’ve no problems with iGPU only Macs. When awaken from sleep, the gmux device decides whether to use iGPU or dGPU for the internal retina screen. This is my best guess: since your MBP’s 750M dedicated graphics is beta supported by the driver, which simultaneously has to maintain eGPU connection, kernel panic can happen. There is a tool “gfxCardStatus” which calls AppleMuxControl’s methods from the user space. You can try forcing iGPU only mode and see if that helps, either “automatic graphics switching” turned on/off in Energy Saver settings. Would be possible to add similar functionality in script’s daemon background process.

@jackfsqyj: Early 2015 13” MBP is supported: https://support.apple.com/en-us/HT204990 A normal Boot Camp installation, a full ISO file method is better than upgrade from Win8.1. Then hot plug the eGPU, install Nvidia drivers, and reboot. If you see black screen or BSOD, boot into OS X and type “sudo ./automate-eGPU.sh -a”, select BOOTCAMP startup disk, and try again. Let me know if that works. @tranj10 said that Evo*’s “enabling Optimus” method is not necessary. In fact, it should be avoided with 2015 MBPs. In case you have done it -> https://support.apple.com/en-us/HT204904#erase

I didn’t say it’s impossible Apple is giving us little hints, they recently updated their FAQ: https://support.apple.com/en-us/HT204154 "Thunderbolt hot plug is supported under Windows 8 or 8.1 on all 2014 and later Mac computers. For all other Mac computers, Windows 7, 8, and 8.1 scans and activates Thunderbolt devices connected to Thunderbolt ports only during the Windows startup process. If your device was not plugged in at start up, Windows will not detect it without a restart.” Maybe I should update the BootCamp drivers since nothing happens when you plug or unplug with a Mid 2014 rMBP on Win8.1. The Late 2014 Mac mini was mentioned earlier, and I noticed this new booting process. Special drivers are needed when you unplug the eGPU - this applies both to OS X and to Windows.

@BastiFH: I just added 2015 Mac board-ids and Fiji codename, no actually new functionality. The last patch number should be incremented even if there are minor changes or bug fixes, but if every small change increases the number, this is going to be v1.0 too soon… @flyfree: I guess you mean the Thunderbolt cable and “hot plugging”? 2014 Mac mini and 2015 Macs do support hot plugging and removing the TB cable on Windows. Connect the TB cable and the eGPU pops up in device manager and gives the screen output - if not, switch the cable to another TB port. Technically it should be possible to do the same on OS X. The only exception are storage devices. Regarding any other PCIe devices, Thunderbolt technology allows unplugging at any time, but drivers have to be aware of it. However this is a step more complicated programming task - maybe possible with a kernel extension. We could just wait for the TB3 Script’s [-a] mode changes Intel’s Thunderbolt controller (156C) properties and is beneficial also when booting into Windows, no more black screens with 2015 Macs. But you have to circulate through OS X and turn on [-a] mode each time. Booting into OS X only preserves the value.

Small changes in v0.9.6: Support for 2015 Macs (-a mode is required for successful booting with a multi-slot enclosure) Prepared for Fiji architecture

@Kurbeco: The constant green and blue light is the only good news. Try other TB cables and ports. If no sign of the GPU under system report/graphics/displays, this is definitely a hardware issue. Going off-topic, let’s continue on that other thread. I hope you get it fixed!

The problem might be AKiTiO itself. I would try with a ATX PSU + paper clip trick + molex-to-barrel adapter. Confirm that you see constant green and blue light when the TB cable is connected. If a blinking light, your TB controller is broken.

The script tries twice to detect the eGPU. There is 4 second delay. Plug in TB cable, wait about 10 seconds and run the script. If still the same, you have a hardware issue. Your graphics card is not properly powered.