GPU MEX not compatible with GPU with compute capability '3.0'?

Hi, I'm trying to set up the prerequisites for the GPU Coder. In the MATLAB Command Window, I entered:

gpuEnvObj = coder.gpuEnvConfig;
gpuEnvObj.BasicCodegen = 1;
gpuEnvObj.BasicCodeexec = 1;
results = coder.checkGpuInstall(gpuEnvObj)

and got the following results:

Compatible GPU           : FAILED (The compute capability '3.0' of the selected GPU '0' is not supported by GPU Coder. Execution of the generated GPU MEX will not be available.)
CUDA Environment         : PASSED 
	Runtime   : PASSED 
	cuFFT     : PASSED 
	cuSOLVER  : PASSED 
	cuBLAS    : PASSED 
Basic Code Generation    : PASSED 

results = 

  struct with fields:

                 gpu: 0
                cuda: 1
               cudnn: 0
            tensorrt: 0
        basiccodegen: 1
       basiccodeexec: 0
         deepcodegen: 0
        deepcodeexec: 0
    tensorrtdatatype: 0
           profiling: 0

My GPU is NVIDIA Quadro K420 (Driver version: 465.89, CUDA version 11.3). My questions are

• Is there really no way to execute GPU Coder on my GPU? In the MATLAB documentations, I have not found descriptions about the requirement of the compute capability.
• The result doesn't say it's impossible to generate the MEX code. Is it possible to generate the MEX code on my PC and then execute the generated code on a cluster node if the node is equipped with a better compute capability?

I appreciate any help.

Answer 1

[Walter Roberson]

https://www.mathworks.com/help/releases/R2021b/parallel-computing/gpu-support-by-release.html

R2020b was the last release with support for Kepler

[Daigo]

I'm using R2020a so it should support for Kepler (cc3.0), right?

[Walter Roberson]

Yes if you have installed CUDA Toolkit 10.1

https://www.mathworks.com/help/releases/R2020a/parallel-computing/gpu-support-by-release.html

[Daigo]

Thank you, Walter. I installed CUDA Toolkit 10.1 and changed an enviroment variable to

• CUDA_PATH = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1

and set the path to to the CUDA executables (C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin)

I restarted my PC and confirmed that the CUDA Toolkit is correctly installed by entering `nvcc --version` on command prompt:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Fri_Feb__8_19:08:26_Pacific_Standard_Time_2019
Cuda compilation tools, release 10.1, V10.1.105

However, still, I get the GPU compatibility error. Here is the input of the GPU Environment Check App:

and here is the result I got:

Does the driver API version also need to be 10.1? According to this post, CUDA has 2 primary APIs: Runtime API and Driver API. Currently, my Runtime API version (result of nvcc --version on cmd0) is 10.1 but my Driver API version (result of nvidia-smi on cmd) is 11.3. In the same post, it is discussed that this kind of inconsistency won't cause any problem in general, but I'm not sure that is the case for the use of GPU Coder.

[Walter Roberson]

Sorry, I do not know.