I would use Jetpack to install cudnn.
When you say “I did install Jetpack 3.2 on Jetson TX2 board.” I would interpret as you got Jetpack 3.2 installed at Host PC and from Host PC you have flashed Jetson with Jetpack 3.2
“I wonder if I installed Jetpack3.2, i dont need to copy and paste cuDnn from Nvidia Homepage.” Why would you copy paste?
“On desktop, installing cuda and cudnn is different step. So i have a conflict.” CUDA can be installed to desktop [Host PC] either standalone or by Jetopack. CUDNN will be a separate standalone step, in my opinion. What conflict do you have?
“Someone said if you check the cudnn installation done, comment $ cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2.” If you have checked cudnn installation while flashing Jetson you should have CUDNN installed to Jetson already. Unless you need custom configuration you do not need to manually configure CUDNN at Jetson, in my opinion.
“But Jetson didn’t say anything. Because i didnt copy and paste cudnn files on the path.” To keep an eye on path matters is always a good practice, in my opinion. Houwever, you wouldn’t need to add cudnn to path if you have installed it to jetson from Jetpack while installing OS via flashing Jetson from Host OS installed Jetpack, as it seems to me.
JetPack has various things it can do, all of which are selectable (or deselectable) via a checkbox. You can disable flashing and just check install of the software you want (for software package steps you’d want your Jetson connected to the Ubuntu host PC over wired ethernet…the micro-B USB is only used for flashing).
JetPack will download the correct dependencies automatically. Individual packages are not available (apparently part of the reason the packages are only available through JetPack is for dependency resolution in a complex set of package dependencies). The components manager is just one of the GUI screens you get to when running JetPack.