Micro SD for Jetson TX2

Hi there,

I am writing regarding how to integrate a micro SD card with the Jetson TX2 for the purpose of expanding storage that we can write to disk. I’m quite new to this so any help would be greatly appreciated.

Our application is to stream and record data of 6 synced cameras to disk - ideally recording data frame by frame in a RAW format or a video encoded format.

We have the following carrier board from Leopard Imaging: LI-TX1-CB-6CAM | Leopard Imaging Inc. which only states it has a micro SD interface without many details.

From this thread TX2 SD card slot - #7 by Andrey1984 it seems Jetson TX2 only supports UHS I cards.

A few questions I have which I hope you could clarify:

• Can Micro SDXC cards be used and what format does it need to be to work with Jetson TX2?
• Does class A1 vs A2 make a difference?
• Do you think V30 will be good enough for this application?
• Are there any Micro SD card specifications you can recommend based on the application we want?

Any insights would be much appreciated.

Kind regards,

Josiah

Faster SD cards work, but they fall back to the slower speed standards. I don’t have any particular recommendations, other than I’ve had good luck with Samsung micro-SD cards.

FYI, I doubt you’ll have good luck with being able to support the required bandwidth on SD. I’ve never used NVMe or SSD (either SATA connector or m.2) with a Jetson (I don’t have any to try), but these are much faster than SD. You could just mount this on some special mount point where you want the data to appear, and then you wouldn’t need to deal with learning to make alternate boot media work as a boot drive.

Thanks for this linuxdev. Yes, I’ve heard that you can use newer cards and they should be backwards compatible with the older interface (at the slower speeds). I’ll have a look around.

Re: SATA or m.2 these definitely would be faster and more suitable for our application, however except for USB A and Micro USB i don’t see how we could attach an external SSD or NVMe to our carrier board? Am I not fully understanding the hardware? Could you explain the process of creating this mount point?

Again, really appreciate the reply!

Perhaps you are referring to the developer board which comes with an m.2 connector and pcie x4? Unfortunately I believe the Leopard imaging carrier board does not have these connector slots…

You are correct that being able to use those types of devices depends on the carrier board. It is probably overly complicated, but if your device is on a local LAN with good throughput speeds, and if you have a server somewhere on that LAN, then you could actually mount the drive to the server and export the drive as iSCSI. Then the device would appear just like any other SATA drive, but instead of it coming from a local PHY it would exist over ethernet. I find gigabit ethernet to be a fairly high performance environment for iSCSI, although newer drives like an SSD can now exceed gigabit. Depending on needs though I’d think gigabit for iSCSI would be sufficient.

Do note though that there is a lot of learning curve to use iSCSI. I don’t necessarily suggest you use this, but it is one of the few external media options if you don’t have any kind of SATA or m.2 or other suitable interfaces. Otherwise you might need a different model of carrier board.

Any kind of partition can be mounted on any directory on the filesystem. As an example, if you have “/dev/sdb1” (adjust for your device), then you can mount it on “/usr/local” and then all content there goes through sdb1. If you umount sdb1, then the original content shows up again. You can selectively copy what is already on “/usr/local” so that the old content does not appear to go away upon mount. Most cuda content is there. For details, see:
https://forums.developer.nvidia.com/t/how-to-install-linux-on-system-disk-and-cuda-on-ssd/188834/2