Steps necessary to setup RTX A5000 on Matlab 2019.b or 2020.a

Hi,

Just installed an RTX A5000 Ampere (24 Gb memory) for GPU computing including machine learning and deep learning.

I have several toolboxes (Parallel computing, Statistics and machine learning, Deep learning and others)

I'd like to know all the necessary steps to properly use the GPU to it's maximum capacity. Through several searchs, answers are not clear.

Do I need to update some librairies (Cuda or else), environement variables or else?

Here are the properties (CUDADevice):

CUDADevice with properties:

Name: 'NVIDIA RTX A5000'

Index: 1

ComputeCapability: '8.6'

SupportsDouble: 1

DriverVersion: 11.4000

ToolkitVersion: 10.1000

MaxThreadsPerBlock: 1024

MaxShmemPerBlock: 49152

MaxThreadBlockSize: [1024 1024 64]

MaxGridSize: [2.1475e+09 65535 65535]

SIMDWidth: 32

TotalMemory: 2.5757e+10

AvailableMemory: 2.4256e+10

MultiprocessorCount: 64

ClockRateKHz: 1695000

ComputeMode: 'Default'

GPUOverlapsTransfers: 1

KernelExecutionTimeout: 1

CanMapHostMemory: 1

DeviceSupported: 1

DeviceSelected: 1

Thanks,

Michel

Answer 1

[David Willingham]

Hi Michel,

Full support for Ampere is only available as of R2021a:

https://mathworks.com/help/parallel-computing/gpu-support-by-release.html

David

[Michel Rathé]

Thanks for the answer David,

I run 2019.b. gpuDevice shows ComputeCapability 8.6 which seems to be conform to the cc8.x on the Gpu support page and the Toolkit version at 10.1 also seems to be ok, BUT I still get the message: "Warning: The CUDA driver must recompile the GPU libraries because your device is more recent than the libraries."

Should I install the toolkit (from Nvidia) and recompile? Or this warning irrelevant.

Thanks,

Michel

[David Willingham]

Hi Michel, upgrading to the latest NVIDIA toolkit might cause older versions of MATLAB to not work. If it works now, I'd leave it, but you won't get the maximum performance.