ippiDilateBorder slow on Xeon E5-2680 v2 (IPP 2017 U3)

Hi,

I use ippiDilateBorder_8u_C1R on a 1000x1000 image (attached) with a 50x50 mask.

On Xeon E5-1650 v4 it takes ~48ms. On Xeon E5-2680 v2 it takes ~3.8s. Both use Windows 10.

Using IPP 2017 update 3.

Is that a known issue? What can I further check?

Thanks,

Shai