Does MATLAB support NVIDIA Ada Lovelace cards for GPU computation?

NVIDIA has released a GPU architecture called "Ada Lovelace". Does MATLAB support using these GPU cards to accelerate deep learning training or other GPU computation?

Example card names: GeForce RTX 40 series (e.g. 4070 etc), RTX 6000 Ada, L40

Answer 1

[MathWorks Support Team]

For releases R2010b to R2020b support for Ada Lovelace will be via NVIDIA's forwards compatibility mode. Optimized device libraries must be compiled at runtime from an unoptimized version. Support can be limited and you might see errors and unexpected behaviour.

Forward compatibility from CUDA version 10.0–10.2 (MATLAB versions R2019a, R2019b, R2020a and R2020b) to Ada Lovelace (compute capability 8.9) has only limited functionality.

Full Ada Lovelace built-in binary support within MATLAB is available from R2021a.

 

You may wish to be aware of the following bug report affecting the Ada Lovelace support in MATLAB: https://www.mathworks.com/support/bugreports/2905390

For more details please see 

https://www.mathworks.com/help/parallel-computing/gpu-computing-requirements.html

.

Starting in R2020b, forward compatibility for GPU devices is disabled by default. Support can be enabled for the current MATLAB session by executing:

parallel.gpu.enableCUDAForwardCompatibility

Enabling forward compatibility can result in wrong answers and unexpected behaviour during GPU computations. In some cases, forward compatibility does not work as expected and recompilation of the libraries results in errors.