support for eGPU via TB3 on macos systems
Afficher commentaires plus anciens
Hello,
I am planning to buy an external GPU for my 2016 MacBook Pro, and I am wondering whether Matlab can take advantage of it for high-peformance computing, Machine Learning and Deep Learning tasks? Any hints on a particular enclosure or GPU to maximize compatibility? I prefer an nvidia GPU to run my other CUDA-based programs. My first guess is GTX-1080Ti or Titan-X on an Akito Node enclosure via Thunderbolt 3 (USB-C).
Alternatively I would consider connecting the eGPU to a linux host.
Which of both configurations would maximize matlab compatibility and performance?
Thank you in advance,
7 commentaires
Walter Roberson
le 26 Nov 2017
Interesting, literally just yesterday I started reading about this possibility, as I am considering switching to a (used) iMac without NVIDIA from my current older MacBook Pro 2012 that has NVIDIA.
I had not gotten very far with my reading, but one thing I noticed at https://developer.apple.com/development-kit/external-graphics/ is "Apps that use Metal, OpenCL, and OpenGL can now take advantage of the increased performance that external graphics processors can bring.", which is possibly hinting that CUDA might not have support.
If you do consider Mac for this, then https://egpu.io/bootcamp-setup-guide-tb3-macbook-pro/ has some hardware related information that is relevant. That page is about setting up Bootcamp to access eGPU, but the parts about which ports to use and what to disable is relevant.
Joss Knight
le 27 Nov 2017
Modifié(e) : Joss Knight
le 19 Déc 2017
We have qualified the Razer Core system on Windows and are aware that it works when you have the latest Mac OS and a Thunderbolt 3-compatible USB-C connector. I can't speak for your particular choice, but we are aware these things do work.
aa
le 28 Nov 2017
Walter Roberson
le 28 Nov 2017
The GPU drivers are available on OS-X / MacOS, including the RCCN. Joss indicates the device is accessible to MATLAB with that enclosure.
Joss Knight
le 19 Déc 2017
Sorry, I made an error in my comment which I have just corrected - we have qualified the Razer Core on Windows, not Mac, but we understand that it works and will be qualifying it soon once we have upgraded our systems to the latest Mac OS.
Whether or not MATLAB can use a device is entirely down to whether the NVIDIA driver can use it. If you can run any CUDA code on the card then so can MATLAB.
Shen Zhao
le 15 Sep 2018
Is there any update for macOS now?
Has anyone tested macOS with Nvidia eGPU CUDA acceleration?
Joss Knight
le 20 Sep 2018
I have mixed reports. Some of our engineers have successfully got this working and others have failed. The biggest sticking point is the need to disable certain kernel protections.
Réponses (0)
Catégories
En savoir plus sur Parallel and Cloud dans Centre d'aide et File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!