Using Xaxier AGX in place of Jetson nano

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hi All,

can you please point me to any documents or links that explain how to use Jetson xavier AGX by Jetson Nano ?

thanks

2

You should probably use either the Xavier or Nano forum to ask, but you will want to include more information and details as to what you mean by “how to use Jetson xavier AGX by Jetson Nano” (anything about your use case would help). For example, if you mean replacing one with the other, or using both together (communicating), or differences in carrier board design (I see the “board-design” tag). FYI, the main documents download is here (you may need to go there, log in, and then go there a second time):
https://developer.nvidia.com/embedded/downloads

3

Thank you very much for the reply.

Apologies for the lack of clarity in the note. I am using Jetson Nano at present and wish to replace it with Xavier AGX hence can you please suggest if there is any document that I can refer to for identifying the changes that I need to make ?

regards

4

The main documents URL is here:
https://developer.nvidia.com/embedded/downloads

The Xavier development kit (with some specs) is here:
https://developer.nvidia.com/embedded/jetson-agx-xavier-developer-kit

In terms of software, the Xavier CUDA compile uses Volta architecture:
-gencode arch=compute_72,code=sm_72
…while the Nano uses:
-gencode arch=compute_53,code=sm_53

You might find this of interest regarding compute compile options:
http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/

The above probably isn’t something to help choose products, but it does imply that if the same CUDA release is available on the different Jetsons, then you should have no issue compiling software for both. The specification of the gencode would need to change between the two. You can compile for both at the cost of a larger file. This compile change is to allow the code to work on either GPU architecture.

The Nano uses an older GPU architecture, and in fact is the same architecture as the TX1. The TX1 had gone end of life for development quite some time back, and although the hardware continued to be available, it wasn’t until the Nano came into existence that various software feature upgrades started to occur. The TX1 could use whatever Nano could use, so why not? However, if you want the more recent architecture, then you need the NX or Xavier.

If you need more RAM, then Xavier is the undisputed King. In terms of CUDA and AI, a GPU depends on physical RAM, and adding swap won’t help directly (swap can allow other programs to give up some physical RAM, but swap won’t be directly useful to a GPU). The development kit I have has 16GB of RAM, which means it can work with much larger CUDA models than can any other kit.

The Xavier later added 32GB RAM modules, double that of the 16GB model I have (my Xavier is from the very first batch). I’m not positive, and someone from NVIDIA may be able to confirm one way or another, but the dev kits now seem to all be 32GB RAM models, but you would have to verify this from wherever you purchase.

Note that carrier boards from dev kits are not available separately. However, the carrier board of the b01 release of Nanos and NX are interchangeable. The NX could offer you somewhere between the full Xavier and Nano, using the newer architecture, and compatible with the b01 Nano carrier boards (including the dev kit if your revision is b01 and not a02). The NX has 8GB of RAM. Between the extra RAM and the newer GPU architecture the NX is quite a step up from the Nano while maintaining its footprint. The Xavier is as much of a step up over the NX as the NX is over the Nano. Just be careful to consider carrier boards if you go to purchase.

Here are NVIDIA’s spec pages. Note that in some cases you will get a unit requiring a microSD card (dev kit versions), but other models (individually purchased modules) have eMMC. The Xavier always comes with eMMC, other non-Xavier dev kits tend to require microSD, while lone non-Xavier module purchases tend to provide eMMC:
Xavier
NX
Nano

5

Thank you very much for the reply.

It will be very helpful if you can provide some mapping between the pins of Nano versus Xavier NX/AGX based upon the functionality.

For examples : I2C0_SCL is pin 185 on Nano and on Xavier NX/AGX its pin …

regards

8

hello siddharthm,

please download Developer Kit User Guides for comparison.
you may also check GPIO header pinout, for example, NX J12 Header Pinout, Nano J41 Header Pinout for reference.
thanks

9

thanks for the reply,

but i feel the headerpinout doesn’t give pin information about a lot of other pins like USB, CAN, ETHERNET etc…

can you please help to point out any comprehensive comparison list ?

10

hello siddharthm,

may I know what’s your use-case?
you may access Product Design Guides for the detailed information on the capabilities of the hardware modules.
thanks

11

thanks for the reply.

As of now I am using Jetson Nano carrier board schematic and layout with some modifications. So i am using all the pins of the 260 pin connector of Nano including interfaces like USB, Ethernet, MIPI, ISC, SPI …

So now i want to make this design compatible with Xavier AGX/NX as well so that in future if i need to replace the nano module with Xavier that should be possible.

Hence I need to know what changes will i have to make in my jetson nano carrier board based schematic to make the compatible with xavier module.

12

hello siddharthm,

there’s an application note that you should start with, NX and Nano Interface Comparison and Migration Application Note .
thanks

13

Thank you very much for the note.

Is there any similar note that compares Nano to Xavier AGX ?

14

Also in the note shared I see this projection in Nano which is marked in red.

Can you please let me know the purpose for the same ?

image.png

15

hello siddharthm,

please check the developer kits page, please refer to NX developer kit and also Nano developer kit.
you may switch the modules, both of them were using the same carrier board,
thanks

16

Thank you very much for the note.

Is there any similar note that compares Nano to Xavier AGX ?

17

Nano and NX share form factors and (assuming b01 revision) can share carrier boards. There is no such comparison for the Xavier AGX versus NX/Nano because Xavier has a different connector type. There is the reference carrier board schematic, but no information that I know of for directly migrating from NX/Nano to Xavier. Anyone migrating to the Xavier from Nano/NX will probably need a complete carrier board redesign.

18

Ok thanks,

can you please let me know what is the worst case and typical power consumption of Jeston Nano & NX separately ?

19

Someone else will have to help with that, which is something I am sure is in one of the docs, but I don’t know without research.

20

thanks for the reply. I see 2 modes mentioned but I am not very clear on that.

Can you please suggest the link where I can post this query ?

21

hello siddharthm,

you may refer to developer guide, Clock Frequency and Power Management.
please check the [Supported Modes and Power Efficiency] session of different Jetson platforms; there’s a table for NVPModel clock configuration to indicate the power budget of default configuration.
for example,
you may have comparison with NX and Jetson AGX Xavier, and Nano and TX1.

BTW,
instead of Q&A in the comments, suggest you may check forum pages to searching your inquiries or you may initial a new topic at proper categories; we should keep using single discussion thread for tracking a question, this may help other developers to find answers more easily and consistency.
thanks

22

Thanks for the reply.

can you please let me know the link to post the hardware and PCB related queries