Tegra K1 vs Tegra X1 vs Tegra X2 cuDNN convolution speeds

Hi all, just a quick, general question about the difference in performance of the 3 current Jetson boards. Based on a cuDNN convolution for input with 512x512x3 channels, convolved with kernel (3x3x256 channels, stride 1 for both U,V, no padding).

I’m curious as I have a Jetson TK1, and wanted to know what kind of performance gains going up with the TX1 and TX2.