Two new eGPU solutions on macOS 10.13.4: pure EFI and hybrid
 
Notifications
Clear all

Two new eGPU solutions on macOS 10.13.4: pure EFI and hybrid  

 of  68
  RSS

goalque
(@goalque)
Noble Member Admin
Joined: 4 years ago
 

@eightarmedpet, @technomad, @daniel_varnai and others who have enabled FileVault - turn it off, or do full erase + reinstall.

Otherwise, EFI's LoadImage() method returns "Not Found" for \System\Library\CoreServices\boot.efi even if the volume path is correct.

We've concluded that this was the common factor, thanks to @daniel_varnai's debug logs.

automate-eGPU EFIapple_set_os.efi

Mid 2015 15-inch MacBook Pro eGPU Master Thread

 
2018 13" MacBook Pro [8th,4C,U] + Radeon VII @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 1809 [build link]  


ReplyQuote
Technomad
(@technomad)
Eminent Member
Joined: 3 years ago
 

Thanks for the info - unfortunately, turning off FileVault isn't an acceptable option 🙁

Mid-2017 MBP 15”/3.1GHz, Razer Core v2, EVGA GTX1080Ti, LG 34UM95, HTC Vive, Samsung Odyssey, MS Hololens.


ReplyQuote
frank böhle
(@frank_bohle)
New Member
Joined: 2 years ago
 

Hi, 
so there are two options now for me to use a egpu. This way and the purge-wrangler, here in another post. 
I wonder, what your recommendation is, what I should try first. 
My set up is as follow: 
Macbook pro 11,3, Retina 15" Late 2013, 2,6 Ghz and 10.13.4 with NVIDIA GeForce GT 750M 2048 MB and Intel Iris Pro 1536 MB
Next Days I will get a Apple Thunderbolt 3 to 2 Adapter, Akitio Node and a AMD Radeon RX 480

Thanks for your Help

Macbook pro 11,3, Retina 15" Late 2013, 2,6 Ghz and 10.13.4 with NVIDIA GeForce GT 750M 2048 MB and Intel Iris Pro 1536 MB
Apple Thunderbolt 3 to 2 Adapter, Akitio Node and a Red Devil AMD Radeon RX 480


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 4 years ago
 

@technomad That's how it has been for a long time, with the old automate-eGPU.sh script as well. One downside of using these Nvidia workarounds. AMD is always better option Smile I will let you know if there is another way.

automate-eGPU EFIapple_set_os.efi

Mid 2015 15-inch MacBook Pro eGPU Master Thread

 
2018 13" MacBook Pro [8th,4C,U] + Radeon VII @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 1809 [build link]  


ReplyQuote
Technomad
(@technomad)
Eminent Member
Joined: 3 years ago
 

The old script did work for me, but always kernel panic'd on logout/shutdown, no matter what I did. Starting to look as though I need to ditch my GTX1080Ti for a Vega RX 64 and then try to work out how to accelerate tensorflow without CUDA 🙁

Mid-2017 MBP 15”/3.1GHz, Razer Core v2, EVGA GTX1080Ti, LG 34UM95, HTC Vive, Samsung Odyssey, MS Hololens.


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 4 years ago
 

Yes, scripts may work with FileVault enabled but lots of users reported problems years ago so the recommendation is not to turn FileVault on.

I got an idea. Maybe the installer could make a copy of \System\Library\CoreServices\boot.efi and "EFI Boot" would load it from an unencrypted USB stick. But I cannot promise anything...

automate-eGPU EFIapple_set_os.efi

Mid 2015 15-inch MacBook Pro eGPU Master Thread

 
2018 13" MacBook Pro [8th,4C,U] + Radeon VII @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 1809 [build link]  


ReplyQuote
Daniel Varnai
(@daniel_varnai)
Eminent Member
Joined: 3 years ago
 
Posted by: Technomad

The old script did work for me, but always kernel panic'd on logout/shutdown, no matter what I did. Starting to look as though I need to ditch my GTX1080Ti for a Vega RX 64 and then try to work out how to accelerate tensorflow without CUDA 🙁

Even tensorflow needs a workaround on Mac to work and it's about 30% slower. It also keeps crashing my mac unless I regularly restart the Jupyter notebook after training a network ... I got nvidia for mostly the same reason as well: cuda, but imo you might be better off installing Linux on your Mac for tensorflow

MBP 2017 15", MacOS 10.15 19A526h, GTX1070 EGPU


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 4 years ago
 

With 2018 MBPs and iMac Pro, you need to "Allow booting from external media"

https://support.apple.com/en-us/HT208198#externalboot

You also may need to change the security level.

automate-eGPU EFIapple_set_os.efi

Mid 2015 15-inch MacBook Pro eGPU Master Thread

 
2018 13" MacBook Pro [8th,4C,U] + Radeon VII @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 1809 [build link]  


ReplyQuote
Technomad
(@technomad)
Eminent Member
Joined: 3 years ago
 
Posted by: Daniel Varnai
Posted by: Technomad

The old script did work for me, but always kernel panic'd on logout/shutdown, no matter what I did. Starting to look as though I need to ditch my GTX1080Ti for a Vega RX 64 and then try to work out how to accelerate tensorflow without CUDA 🙁

Even tensorflow needs a workaround on Mac to work and it's about 30% slower. It also keeps crashing my mac unless I regularly restart the Jupyter notebook after training a network ... I got nvidia for mostly the same reason as well: cuda, but imo you might be better off installing Linux on your Mac for tensorflow

Thanks for the heads-up - I've been following some of the source forks for tensorflow, and am coming to the same conclusion. I might be better off dedicating a lower-power Mac to that, given that most of the heavy lifting would be done by the GPU. Or send it to a cloud.

Mid-2017 MBP 15”/3.1GHz, Razer Core v2, EVGA GTX1080Ti, LG 34UM95, HTC Vive, Samsung Odyssey, MS Hololens.


ReplyQuote
Daniel Varnai
(@daniel_varnai)
Eminent Member
Joined: 3 years ago
 
Posted by: Technomad
Posted by: Daniel Varnai
Posted by: Technomad

The old script did work for me, but always kernel panic'd on logout/shutdown, no matter what I did. Starting to look as though I need to ditch my GTX1080Ti for a Vega RX 64 and then try to work out how to accelerate tensorflow without CUDA 🙁

Even tensorflow needs a workaround on Mac to work and it's about 30% slower. It also keeps crashing my mac unless I regularly restart the Jupyter notebook after training a network ... I got nvidia for mostly the same reason as well: cuda, but imo you might be better off installing Linux on your Mac for tensorflow

Thanks for the heads-up - I've been following some of the source forks for tensorflow, and am coming to the same conclusion. I might be better off dedicating a lower-power Mac to that, given that most of the heavy lifting would be done by the GPU. Or send it to a cloud.

https://github.com/zylo117/tensorflow-gpu-macosx

this is the one that works for me at the moment but I got so fed up with my mac crashing all the time that I just ended up using AWS EC2 p2.xlarge instance with deep learning AMI. it would be awesome if tensorflow worked with opencl. utilising the iris pro chip, the radeon pro 555 and my egpu at the same time would probably melt the chassis of my mbp

MBP 2017 15", MacOS 10.15 19A526h, GTX1070 EGPU


ReplyQuote
 of  68