GPU installation

GPU installation NVIDIA TITAN Xp Graphics Card (900-1G611-2530-000) (12GB) MacOS 10.13.6 (17G65)

2018-10-02 https://github.com/learex/macOS-eGPU/tree/master#macos-egpush I previously ran the macos-egpu install script and installed everything. Current output:

~/Documents/4-PhD/Other/worklog$ macos-egpu

macOS-eGPU.sh (v1.6)

Accept license terms…                                                                                          [done] Killing all other running programs…                                                                              [OK] Internet connection established…                                                                                [YES] Fetching system information…    macOS info                                                                                                    [done]    system integrity protection                                                                                   [done]    thunderbolt version                                                                                           [done]    GPU information                                                                                               [done]    installed eGPU software                                                                                       [done]    installed patches                                                                                             [done]    installed programs                                                                                            [done] Setting internal switches… Checking for incompatibilies and up to date software…    NVIDIA drivers                                                                                    [skip, up to date]    NVIDIA eGPU enabler                                                                             [skip, incompatible]    AMD legacy drivers                                                                                            [skip]    T82 unblocker                                                                                                 [skip]    NVIDIA dGPU deactivator                                                                                       [skip]    macOS 10.13.4/.5 NVIDIA patch                                                                   [skip, incompatible]    macOS 10.13.4+ thunderbolt 1/2 unlock                                                           [skip, incompatible]    CUDA software       CUDA drivers                                                                                   [skip, up to date]       CUDA developer driver                                                                          [skip, up to date]       CUDA toolkit                                                                                   [skip, up to date]       CUDA samples                                                                                   [skip, up to date]    thunderbolt daemon                                                                                            [skip]    IO PCIE Tunnelled patch                                                                    [skip, already installed] Checking if SIP is sufficently disabled…                                                                         [OK]

Download external content…

Uninstalling… Installing… Patching… deactivating auto-updates…

Finish…    cleaning system                                                                                               [done] Nothing has been changed.

============================ Now how do I find out if it’s working?  https://apple.stackexchange.com/questions/277356/machine-learning-on-external-gpu-with-cuda-and-late-mbp-2016 The accepted answer has almost the same configuration as mine, except using an Akitio Node enclosure instead of a Sonnet Box. But as I understand it, Sonnet Box has the best support for MacBooks. 

My CUDA version: (via /usr/local/cuda/bin/nvcc —version) nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2018 NVIDIA Corporation Built on Sat_Aug_25_21:08:56_CDT_2018 Cuda compilation tools, release 10.0, V10.0.130

============================ OK, looks like I’ve got CUDA and CUDNN installed. But now how to get tensorflow with GPU support? Tried Tensorflow’s docker image. No luck.  Tried installing tensorflow-gpu with python 2 and 3. There was always some library incompatibility. 

Looks like I’ll have to compile it myself. 

But not today.  Stopping work on this for now.

================================= October 2019.

Trying to get the egpu working with Zuz’s windows machine. The conclusion for now is that we got the egpu working, and the Titan Xp recognized, but keep getting error 43 saying the device was stopped because it reported problems. We can’t find a version of the Nvidia webdriver that works, and installing various versions cause the built-in GPU to stop working too. Our latest attempts were installing the studio version v431.86 and the gaming version v436.48.

— MOVED TO DRIVE —