The Tegra X1 is a system on a chip (SoC). Normally it can only be purchased permanently soldered to a small circuit board (the Jetson TX1 module). That module needs to be seated on something similar to a motherboard specifically designed for that module (system on a module, SoM, gets mounted on a carrier board). Several people sell various carrier board, the one sold by nVidia is designed for developers to have good access to most features, but is perhaps not good for a field device (e.g., it’s a bit big and has things you don’t need if using it in a vehicle or aerial drone). When you purchase the developer kit you get the module pre-mounted on the developer carrier board.
FYI, the Jetson TX1 uses the Tegra 210 chip, which is the tegra21x series. Everything else around it is the module board, carrier board, or board support package. When using the module on the developer carrier board you’ll see some references to “p2597” and “2180”. This is the identification of the module’s circuit board and the carrier board circuit board. References to tegra210 are references to the specific chip, references to tegra21 or tegra21x are references shared in common to all variants of the Tegra TX1 chip.
Jetson in general is a reference to GPU SoC products sold by nVidia based on those Tegra chip series which support GPU and which have some sort of board support (something like an older Tegra3 only has board support from third party vendors and won’t use “Jetson” in the name). So a Jetson TX1 implies the tegra210 on its module and developer carrier board. The older Jetson TK1 is the tegra124, or tegra12x series. Note that the Jetson TX1 developer kit has a camera which comes with the developer carrier board.
The Jetson TX1 developer kit’s module can be unscrewed from the developer board and seated on other boards. You could test things out on a developer board and then later plug it in to a third party carrier board. Part of the board support package of software would require changing, though the module portion would remain constant. The whole purpose of making this as a module which can be separated from the carrier board is for the exact reason of making it possible to use the same module both for development and for use with a custom carrier board. The carrier board can evolve while keeping the module constant.
I don’t know how to compare neural network performance on the TX1 with other solutions, but what I’ve seen is that other platforms are typically used to train neural networks, and the TX1 is good at using those pre-trained networks with anything you might want to put on a drone or vehicle.