I have install AGX Xavier by SDK_manager, now I have a TX2 board. Can I install TX2 environment on this machine? does they have conflicts?
I don’t know your exact problem. You need to install TX2 environment on a TX2 board.
when I use SDK_manager-0.9.13, I can choose which target hardware I use, like Xavier or TX2.I have choosed Xavier and install host machine environment. Now I want to choose TX2. I upload SDK_manager picture as attachments.
Yes, you need to choose the target hardware to TX2 if you want to flash TX2 board.
As 3.png, when I choose TX2, the HOST COMPONENTS have installed, because I have installed Xavier before. And TARGET COMPONENTS need to be downloaded, which includes Jteson OS ect. But I have installed Xavier like 4.png, which include Jteson OS too. Does They have conflicts? Can I use like 3.png?
If you want to flash through sdkmanager, the answer is no.
However, if you want to flash with the flash script directly, then actually I am not sure either.
You could go to Linux_for_Tegra folder and you will see TX2 config is also there too.
You’ll find the host has only one (same) set of components regardless of whether you select TX2 or Xavier. The components specific to TX2 or Xavier have no effect on host. Now if you were to add components from two different JetPack release versions, then the host would get different components from the different JetPack versions.
But I wouldn’t make that assumption going forward. You are living dangerously then.
On the host CUDA of different revisions can coexist. For example, my Fedora system has:
/usr/local/cuda-8.0 /usr/local/cuda-9.0 /usr/local/cuda-10.0
It happens that “/usr/local/cuda/” is a symbolic link to “/usr/local/cuda-10.0/”, and thus a default (if you don’t specify version) for this system is cuda-10.
The distinction is that the Jetson can only function with the single version of CUDA it was designed for.
I can’t tell you what the future is, but the current system is such that the host won’t care about mixes of versions, although you might need to specify a specific version in some cases.
CUDA versioning is a bit complex but the bottom line is that different versions of CUDA can support the same GPU.
Yes, you can lay down whatever user run-time you want. So in that regard, I agree with you - it’s just libraries. But that doesn’t mean your application will work! Caveat emptor.