Hi, I ran a Linux program to increase the memory from 4 to 8 GB to a Jetson Nano Developer Kit. Afterwards, I tried to attach a VANSUNY SSD (120 GB External SSD, USB 3.0 400MB/s High-Speed Read Write Portable Aluminum SSD External Hard Drive, Mobile External Solid State Drive) to further increase the memory. However, the memory only went from 8 to 9.9 GB. What am I doing wrong?
Which memory type are we talking about? It sounds like you wanted to add swap space for virtual memory. If so, then it is easy to create either a swap partition or a swap file on the new SSD, but be warned this may not work for what you are interested in.
Beware that the GPU (and thus CUDA) can only use physical RAM. Normal user space software can end up being a lot slower, but giving up some physical RAM and indirectly helping make more physical RAM available to the GPU.
A typical rule of thumb is that once your swap is twice the physical RAM, then you’ll be grinding your system to a halt by using more swap.
If you are only speaking of more disk space, then you are in luck and this is easy to work with. Can you give more detail on exactly what you want to expand?
Thank you for your reply. Yes, I want to add SWAP space for virtual memory. I want to create a SWAP file on the new SSD, and not more disk space. A friend suggested this because my Python program runs on my 16 GB laptop, but it stops on this Jetson Nano.
Assuming your Nano has 4GB of RAM (there is also a 2GB model), then you will be very disappointed in performance if you swap out more than 8GB. So I’ll suggest trying 4GB, and if you need more, then try 8GB. While doing this keep in mind what was mentioned earlier: Your CUDA and other programs needing memory for the GPU cannot use swap, so if your AI models need more memory, then you are out of luck unless the swap frees some other program’s physical RAM requirements and ends up leaving more RAM for the GPU to use, it won’t help.
There will also be warnings that a solid state drive, despite being fast, can wear out sooner. Your drive could fail much earlier than you expect.
The instructions won’t be any different on a Jetson versus a desktop PC. Here’s one article which seemed good for the topic, and included some tuning tips:
Appreciate your help. Do you know what this error message I’m getting means? If it’s hardware related, then the only recourse may be to upgrade the developer kit. Be safe.Error Message
Sorry, I don’t know enough about CUDA/Python code to say. I do believe though that such errors are highly unlikely due to hardware failure, although there could be some content simply not compatible with the hardware (e.g., a wrong version). In that case you’d probably be better off starting a new topic which mentions the particular software and use-case.
Thank you. I’ll start a new topic in case someone on the forum knows.