I have a whole other set of parts that I’m going to assemble, and I’ll see how that goes. Unopened Jetson, A603, and NVMe. If it goes okay, I will consider systematically swapping parts to identify which one is bad, but this would be a huge time sink. More likely, I will get an NVMe-to-USB adapter and just clone the working drive.
Honestly this just exposes how brittle and user-unfriendly these tools are, more than anything else. The whole process should be performing more checks, provide more insight into what’s going on, perform more automatic troubleshooting, and there shouldn’t be so many points of failure. The tools should work across more versions of Linux (and other host platforms), they should work in VMs, they should be better about telling you which dependencies are missing, etc. Considering the vast number of complains I see here and in Nvidia’s forums, whoever developed these seriously dropped the ball.
I’ve heard rumors that AMD is developing a competing product. Perhaps this will light a fire under Nvidia and get them to develop a better product. I can’t begin to tell you just how many problems there are even with fully-working Jetsons. Want a GPIO to connect to something a meter away? Have to use an amplifier in between. Want to toggle a GPIO faster than 60KHz? Not possible. Want to consistently perform SPI transactions more often than every 22ms? Again, not possible.