Why comm.gpu.AWGNChannel is not faster than comm.AWGNChannel?

I changed the code to the following,

M = 8;
modData = zeros(8192,1,8);
SNR = randn(8,1);
for i=1:8
modData(:,:,i) = pskmod(randi([0 M-1],8192,1),M,pi/M);
end
channelop = awgntest(modData,SNR);
t = @() awgntest(modData,SNR);
gputimeit(t) % For GPU
timeit() % for functions other than GPU

Now the modData has 8 sets of PSK Modulated data. (3-D)

function awgntest() is,

function channelOutput = awgntest(modData,SNR)

for i=1:8
% channelOutput = awgn(modData(:,:,i),SNR(i));
% gpuChannel = comm.gpu.AWGNChannel('NoiseMethod','Signal to noise ratio (SNR)','SNR',SNR(i));
% channelOutput = gpuChannel(modData(:,:,i));
Channel = comm.AWGNChannel('NoiseMethod','Signal to noise ratio (SNR)','SNR',SNR(i));
channelOutput = step(Channel,modData(:,:,i));
end

• comm.AWGNChannel = 0.0086s
• comm.gpuAWGNChannel = 0.4172s (First run); 0.4181 (Second run)
• awgn function = 0.0027s

Still, the time taken for gpu AWGN channel is more than the others.

The main Problem is, I want to call this function from Simulink so I thought I will test for it performance from the MATLAB and then use this function in Simulink. But it seems it's not fast.

What do you think is the reason?

And also is it practical to use gpu functions if I want to Speed up the Simulation execution time of my Simulink model? If not other than this and parallel computing Toolbox, what other Options are there for using the GPU Cores for parallel processing.?