Impact of setting KERNEl_RADIUS in convolution separable when trying with larger input image size

Could someone explain to me the impact of kernel_radius value in convolution separable code when working with larger input size (eg: Image Width x Height = 12800 x 12800)? I don't see any variation even when setting the value to KERNEL_RADIUS=8 or setting it to KERNEL_RADIUS=1 in this case.

Here is the output for the above image size with Kernel_radius = 8

Image Width x Height = 12800 x 12800

Allocating and initializing host arrays...
Allocating and initializing CUDA arrays...
Running GPU convolution (16 identical iterations)...

convolutionSeparable, Throughput = 21993.2377 MPixels/sec, **Time = 0.00745 s**, Size = 163840000 Pixels, NumDevsUsed = 1, Workgroup = 0

when kernel_radius = 1

Image Width x Height = 12800 x 12800

Allocating and initializing host arrays...
Allocating and initializing CUDA arrays...
Running GPU convolution (16 identical iterations)...

convolutionSeparable, Throughput = 21968.3556 MPixels/sec, **Time = 0.00746 s**, Size = 163840000 Pixels, NumDevsUsed = 1, Workgroup = 0

Is there any specific reason why it is not getting affected?
Also, it would be great if someone could tell me how to choose the kernel size (or any other parameter which needs to be set apart from it) while performing convolution separable on a larger input size.

Thanks,
Vidya.

Start with the convolution separable sample and its KERNEL_RADIUS setting, using the reported 12800x12800 runs as the comparison case. Document the radius/kernel-size relationship and any other parameters needed for larger inputs; done means the sample guidance explains why the radius-1 and radius-8 results differ so little.