CUDALink and eGPU for Mac?

Question

I have:

• MacBook Pro mid 2014 13" (i.e. just an onboard Intel graphics card)
• NVIDIA GeForce Titan X
• Akitio Node
• Thunderbolt 2 Cable
• Thunderbolt 2 to Thunderbolt 3 Adaptor

I have installed the NVIDIA Drivers for the GPU and CUDA (and my MacBook Pro now recognizes the GPU):

also:

ls /Library/Frameworks | grep CUDA
CUDA.framework/

So now what?

Needs["CUDALink`"]
CUDAInformation[]

CUDAInformation::invdrivver: CUDA was unable to determine the CUDA driver version installed on the system. Refer to CUDALink System Requirements for system requirements.

CUDAQ[]

False

CUDAResourcesInstall[Update -> True]

PacletUninstall::nodelete: The paclet named CUDALink at location /Applications/Mathematica.app/Contents/SystemFiles/Links/CUDALink cannot be uninstalled because it does not reside in the standard paclet repository. You will have to delete the paclet's files manually, such as by using DeleteDirectory with DeleteContents->True. PacletUninstall::nodelete: The paclet named GPUTools at location /Applications/Mathematica.app/Contents/SystemFiles/Links/GPUTools cannot be uninstalled because it does not reside in the standard paclet repository. You will have to delete the paclet's files manually, such as by using DeleteDirectory with DeleteContents->True. General::stop: Further output of PacletUninstall::nodelete will be suppressed during this calculation.

CUDAResourcesInformation[]

{{"Name" -> "CUDAResources", "Version" -> "10.5.0", "BuildNumber" -> "", "Qualifier" -> "OSX", "WolframVersion" -> "10.5+", "SystemID" -> {"MacOSX-x86-64"}, "Description" -> "{ToolkitVersion -> 7.5, MinimumDriver -> 300.0}", "Category" -> "", "Creator" -> "", "Publisher" -> "", "Support" -> "", "Internal" -> False, "Location" -> "/Users/sumner/Library/Mathematica/Paclets/Repository/\ CUDAResources-OSX-10.5.0", "Context" -> {}, "Enabled" -> True, "Loading" -> Manual, }}

So any ideas as to what I should do to troubleshoot this?

Answer 1

I was just able to get CUDALink to detect my eGPU. I have a Titan XP in an Akitio Node 3 with Thunderbolt 2-> 3 adapter running on a Retina 15-inch Late 2013 Macbook Pro with High Sierra 10.13.2 and MMA 11.2.0.0

I followed the instructions here: https://egpu.io/forums/mac-setup/wip-nvidia-egpu-support-for-high-sierra/

Two wrinkles:

1. Be sure you download the NVIDIA drivers and the NVDA support installer for the exact build of your OS (there are multiple builds with the same version number). To find the build go into System Report... on your Mac and click on Software.

2. If you have FileVault enabled, in order to use the eGPU I had to take the following steps:

   1. Shut down your Mac and make sure the eGPU is disconnected.
   2. Restart your Mac and login.
   3. Shut down your Mac
   4. Restart your Mac and enter your password to start FileVault decryption and login.
   5. As the progress bar starts moving, plug in the eGPU

When I check CUDAInformation[] in MMA 11.2.0.0, I see this:

1 -> {"Name" -> "TITAN Xp", "Clock Rate" -> 1582000, "Compute Capabilities" -> 6.1, "GPU Overlap" -> 1, "Maximum Block Dimensions" -> {1024, 1024, 64}, "Maximum Grid Dimensions" -> {2147483647, 65535, 65535}, "Maximum Threads Per Block" -> 1024, "Maximum Shared Memory Per Block" -> 49152, "Total Constant Memory" -> 65536, "Warp Size" -> 32, "Maximum Pitch" -> 2147483647, "Maximum Registers Per Block" -> 65536, "Texture Alignment" -> 512, "Multiprocessor Count" -> 30, "Core Count" -> 960, "Execution Timeout" -> 1, "Integrated" -> False, "Can Map Host Memory" -> True, "Compute Mode" -> "Default", "Texture1D Width" -> 131072, "Texture2D Width" -> 131072, "Texture2D Height" -> 65536, "Texture3D Width" -> 16384, "Texture3D Height" -> 16384, "Texture3D Depth" -> 16384, "Texture2D Array Width" -> 32768, "Texture2D Array Height" -> 32768, "Texture2D Array Slices" -> 2048, "Surface Alignment" -> 512, "Concurrent Kernels" -> True, "ECC Enabled" -> False, "TCC Enabled" -> False, "Total Memory" -> 12884705280},

2 -> {"Name" -> "GeForce GT 750M", "Clock Rate" -> 925500, "Compute Capabilities" -> 3., "GPU Overlap" -> 1, "Maximum Block Dimensions" -> {1024, 1024, 64}, "Maximum Grid Dimensions" -> {2147483647, 65535, 65535}, "Maximum Threads Per Block" -> 1024, "Maximum Shared Memory Per Block" -> 49152, "Total Constant Memory" -> 65536, "Warp Size" -> 32, "Maximum Pitch" -> 2147483647, "Maximum Registers Per Block" -> 65536, "Texture Alignment" -> 512, "Multiprocessor Count" -> 2, "Core Count" -> 64, "Execution Timeout" -> 1, "Integrated" -> False, "Can Map Host Memory" -> True, "Compute Mode" -> "Default", "Texture1D Width" -> 65536, "Texture2D Width" -> 65536, "Texture2D Height" -> 65536, "Texture3D Width" -> 4096, "Texture3D Height" -> 4096, "Texture3D Depth" -> 4096, "Texture2D Array Width" -> 16384, "Texture2D Array Height" -> 16384, "Texture2D Array Slices" -> 2048, "Surface Alignment" -> 512, "Concurrent Kernels" -> True, "ECC Enabled" -> False, "TCC Enabled" -> False, "Total Memory" -> 2147024896}}

I am using this setup only for CUDALink, so I have not tested whether an external monitor attached to the Titan XP would work, nor have I tried to accelerate the internal display.