Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Building external GPU's on Mac (eGPU)
- Thread starter Synchro3
- Start date
- Sort by reaction score
I have such an eGPU setup, too. I opted to run an ITX GTX750Ti inside the AKiTiO case, as it fit's in perfectly and does not need any additional power supply (you only need to upgrade the 5A brick to a 8A one) as it can run from the 75W PCIe port.
The thing that concerns me is the heat generation. I thought about observing the GPUs temperature from one of those many apps out there, but at last with my old NVIDIA 8500 GT test GPU it does not work. Anybody has any knowledge wether it works with newer cards?
The thing that concerns me is the heat generation. I thought about observing the GPUs temperature from one of those many apps out there, but at last with my old NVIDIA 8500 GT test GPU it does not work. Anybody has any knowledge wether it works with newer cards?
eGPU in OS X 10.11 El Capitan
The steps for making an eGPU (like my GTX 960) work with my late 2014 Mac Mini in OS X El Capitan have changed a little from those in OS X Yosemite. It is necessary, coming from an earlier installation of OS X Yosemite, to start with a fresh installation of OS X El Capitan (not just the upgraded version), to make sure that no traces of earlier modified kexts are present.
NVidia has stopped checking whether NVidia hardware is actually present, so installation of the latest WebDriver for OS X El Capitan will succeed as long as it matches the installed OS X version and build number.
Kext modifications
Kexts in the /System/Library/Extensions folder can now only be modified and loaded when System Integrity Protection (SIP) is switched off. This can be done in Recovery Mode with the terminal command csrutil disable followed by a reboot.
The NVidia-related kexts to be modified are:
<key>CFBundleIdentifier</key>
<string>...</string>
and add to those sections (inside the same <dict></dict> tags) the following tag/value pair
<key>IOPCITunnelCompatible</key>
<true/>
A similar change to AppleHDAController.kext (inside AppleHDA.kext) can be omitted since HDMI sound does not work.
The eGPU graphics output can be selected for late 2014 Mac Mini without any further steps. Finally, rebuild system caches with the Terminal commands
sudo touch /System/Library/Extensions/
sudo kextcache -system-caches
and wait a few minutes before rebooting.
The above changes must be redone after an upgrade of OS X or of the NVidia WebDriver package. Resetting PRAM requires that SIP be disabled again in Recovery mode.
Enjoy!
The steps for making an eGPU (like my GTX 960) work with my late 2014 Mac Mini in OS X El Capitan have changed a little from those in OS X Yosemite. It is necessary, coming from an earlier installation of OS X Yosemite, to start with a fresh installation of OS X El Capitan (not just the upgraded version), to make sure that no traces of earlier modified kexts are present.
NVidia has stopped checking whether NVidia hardware is actually present, so installation of the latest WebDriver for OS X El Capitan will succeed as long as it matches the installed OS X version and build number.
Kext modifications
Kexts in the /System/Library/Extensions folder can now only be modified and loaded when System Integrity Protection (SIP) is switched off. This can be done in Recovery Mode with the terminal command csrutil disable followed by a reboot.
The NVidia-related kexts to be modified are:
- IONDRVSupport.kext
- NVDAStartup.kext
- NVDAStartupWeb.kext
- GeForce.kext
- NVDAResman.kext
<key>CFBundleIdentifier</key>
<string>...</string>
and add to those sections (inside the same <dict></dict> tags) the following tag/value pair
<key>IOPCITunnelCompatible</key>
<true/>
A similar change to AppleHDAController.kext (inside AppleHDA.kext) can be omitted since HDMI sound does not work.
The eGPU graphics output can be selected for late 2014 Mac Mini without any further steps. Finally, rebuild system caches with the Terminal commands
sudo touch /System/Library/Extensions/
sudo kextcache -system-caches
and wait a few minutes before rebooting.
The above changes must be redone after an upgrade of OS X or of the NVidia WebDriver package. Resetting PRAM requires that SIP be disabled again in Recovery mode.
Enjoy!
Last edited:
Is there an automated way to edit those kext files? I know Thunderbolterizer can edit kext files for Thunderbolt eGPUs, but I'm not sure if it works in El Capitan.
Last edited:
Not as far as I know. There is no need for fancy programmes to edit kext files, just open a Terminal and use the "nano" text editor that already comes with OS X (or whatever text editor you fancy).
Maybe due to the fact that AKiTiO is a single slot enclosure and electrically x4. I haven’t noticed any stability issues with a multi-slot enclosure & AMDs, they show up as PCIe lane with x8 (even though TB2 connection restricts the speed to x4).
*** automate-eGPU.sh v0.9.7 - (c) 2015 by Goalque ***
-------------------------------------------------------
Detected eGPU
Curacao XT [Radeon R7 370 / R9 270X/370 OEM]
Current OS X
10.11.1 15B42
Previous OS X
10.10.5 14F27
Latest installed Nvidia web driver
Version: 346.03.03f01
Source: 3rd Party
Install Date: 10/22/15, 9:19 PM
Checking IOPCITunnelCompatible keys...
IOPCITunnelCompatible mods are valid.
IOPCITunnelCompatible mods done.
All ready. Please restart the Mac.
The script makes it easy, AMD or Nvidia or mix of them.
Which GPU would you use externally? I created a poll: https://forums.macrumors.com/threads/poll-which-gpu-would-you-use-as-a-thunderbolt-egpu.1932396/
I’ve recently learned that a Thunderbolt eGPU can’t use a MacBook’s internal screen in OS X. Could this problem be circumvented by using this headless video accelerator: http://eshop.macsales.com/item/NewerTech/ADP4KHEAD/
As far as I know this is only possible with Nvidia Optimus driver running and 'cloning' the external display output to the internal display. Possible in Windows, but not in OS X. And you will get a significant loss of performance.
Better using an eGPU with an external display connected directly to the graphics card.
Better using an eGPU with an external display connected directly to the graphics card.
You can't use the internal display when your Mac had a discrete GPU. But if you only have Intel Graphics you can.
Setting up my own egpu with the instructions above but having an issue with the set up.
Got the Akitio box w/ 8A power supply and updated my 2014 Mac mini to the most current 10.11 and Nvidia web drivers. I modified the files specified in post #28 above and installed an old GT 210 I had laying around. Fired it all up and it is working fine with a 23" and 30" cinema display hooked up. I ordered a GT 740 and swapped it in for the old 210. On restart, no video is out put to the displays. Hooked one of the displays in the mini directly and checked the system. The system profile shows the GT 740 along side the integrated intel graphics but no display is shown connected to it even though one monitor is still hooked up. Swapped the old GT 210 back in and display output works fine. Also borrowed a 750ti to test and it showed the same issue. No video output on any of the ports, even though the card is detected correctly in system profile. Odd that it works with the older GT 210 but not the newer 700 series cards.
Not sure if I missed a file to modify or something. Any advice would be appreciated.
Got the Akitio box w/ 8A power supply and updated my 2014 Mac mini to the most current 10.11 and Nvidia web drivers. I modified the files specified in post #28 above and installed an old GT 210 I had laying around. Fired it all up and it is working fine with a 23" and 30" cinema display hooked up. I ordered a GT 740 and swapped it in for the old 210. On restart, no video is out put to the displays. Hooked one of the displays in the mini directly and checked the system. The system profile shows the GT 740 along side the integrated intel graphics but no display is shown connected to it even though one monitor is still hooked up. Swapped the old GT 210 back in and display output works fine. Also borrowed a 750ti to test and it showed the same issue. No video output on any of the ports, even though the card is detected correctly in system profile. Odd that it works with the older GT 210 but not the newer 700 series cards.
Not sure if I missed a file to modify or something. Any advice would be appreciated.
64 watts should be fine for a GT 740, but yeah, a 76 watt PSU might not be sufficient.
If Blender can select or adress the GPU directly it should be possible! Octane Render can do that! If there is a way that I could test that for you, let me know. Maybe a Blender benchmark?
Edit: Just tested it and it worked great. Ran the BMW27.blend test with stock settings in 01:37:67 on the eGPU.
Last edited:
I recently got an eGPU. I noticed that Minecraft seems to only use my CPU even though I have the eGPU plugged in. Does anyone here know of a program which will allow me to force Minecraft to use the dedicated eGPU instead of just using the processor? Would GFXCardStatus suffice?
synchro3 says in the OP...
2) Edit three KEXT's for tunneling Nvidia drivers PCI Express functionality to Thunderbolt:
/System/Library/Extensions/NVDAStartup.kext
/System/Library/Extensions/IONDRVSupport.kext
/System/Library/Extensions/AppleHDA.kext/Contents/PlugIns/AppleHDAController.kext
Look for sections beginning with <key>CFBundleIdentifier</key> and add just before </dict>, the two following lines :
<key>IOPCITunnelCompatible</key>
<true/>
... etc ...
Thunderbolt to PCI Express adapter
- When I began my first eGPU the only affordable adapter was the PE4L (PCIE-expresscard):http://www.hwtools.net/Adapter/PE4L V2.1.html together with Sonnet Echo ExpressCard Prohttp://www.sonnettech.com/product/echoexpresscard34thunderbolt.html
Does anyone know if these exact same kext edits and procedure would work with the PE4L (PCIE-expresscard)? Or are they thunderbolt specific? How might I get the computer to recognize the expresscard eGPU without the Thunderbolt Sonnet adapter? The OP never mentions this guide in particular works just that when he first began he used the PE4L. Did he ever use the PE4L with El Capitan? Thanks is advance!
2) Edit three KEXT's for tunneling Nvidia drivers PCI Express functionality to Thunderbolt:
/System/Library/Extensions/NVDAStartup.kext
/System/Library/Extensions/IONDRVSupport.kext
/System/Library/Extensions/AppleHDA.kext/Contents/PlugIns/AppleHDAController.kext
Look for sections beginning with <key>CFBundleIdentifier</key> and add just before </dict>, the two following lines :
<key>IOPCITunnelCompatible</key>
<true/>
... etc ...
Thunderbolt to PCI Express adapter
- When I began my first eGPU the only affordable adapter was the PE4L (PCIE-expresscard):http://www.hwtools.net/Adapter/PE4L V2.1.html together with Sonnet Echo ExpressCard Prohttp://www.sonnettech.com/product/echoexpresscard34thunderbolt.html
Does anyone know if these exact same kext edits and procedure would work with the PE4L (PCIE-expresscard)? Or are they thunderbolt specific? How might I get the computer to recognize the expresscard eGPU without the Thunderbolt Sonnet adapter? The OP never mentions this guide in particular works just that when he first began he used the PE4L. Did he ever use the PE4L with El Capitan? Thanks is advance!
