What does the error message "Requested code-table for unsupported compute-capability: 2" mean?

I am using a workstation with two Nvidia Tesla C2075 GPUs, as well as a Quadro 4000 graphics card. All are CUDA compute capability 2.0. Matlab version is 2017b, OS is Ubuntu 18.04 LTS, and Nvidia driver version is 390.48, Cuda toolkit version is 9.1.

When I run the following script:

N = 10000;
s1 = gpuArray.rand(N);
s2 = gpuArray.rand(N);
s3 = gpuArray.rand(N);
[o1, o2] = arrayfun(@aFunction, s1, s2, s3);

function [o1, o2] = aFunction(a, b, c)
o1 = a + b;
o2 = o1 .* c + 2;
end

I get the following error message:

Error using gpuArray/arrayfun Requested code-table for unsupported compute-capability: 2

Any advice on where the problem might lie? Note that the MATLAB GPU Benchmark code (gpuBench.m) runs without any problems, so there does not seem to be a fundamental setup problem.

Answer 1

[Joss Knight]

It means you've hit a bug for Fermi class GPUs in R2017b and need to download the latest update.

[Paul Schaberg]

Thanks. When I "Check for Updates" I am informed that all of my 2017b products are up to date. (I went back to 2017b from 2018a because the latter requires Cuda compute capability of 3.0 or higher.)

[PaulS]

Installed Update 6 and now the code runs. Thanks very much.

[Joss Knight]

That's a pity, that it didn't tell you there was an update available. I hope that isn't a bug.