HATs and other add-ons • Re: AI Kit + NVMe SSD ?

Well you asked "What kind of frame rates can I expect", that post above yours clearly shows that going down to gen2 gives big drop on framerate. That alone answers your question. Also if you want any performance then when everything is started and running is should better run all from RAM as much as possible so disk i/o should be minimal and NVME vs microSD should better not matter. If it does, then you already have suboptimal system no matter which storage type you use. Or do you have reasons to expect that the system you described should/will constantly access the storage while running?

Thats precisely my question, ie if a MicroSD card becomes a bottleneck for my application. I do have 3 to 4 codes running parallel and updating 10~20 sharable variables based on the real-time location of face center wrt to screen center.
I am planning to buy Pi HAT+ w/ 26TOPS Hailo(https://www.raspberrypi.com/news/raspberry-pi-ai-hat/). Thought in my mind is to use this "expensive" hardware stack for other performance-intensive use cases like running LLM, faster object detection in running car or training models. I am not sure if going with MicroSD would become a bottleneck or it's only the RAM that matters. Is there a middle path b/w MicroSD card and an NVME? How about USB? Is it more faster and reliable than MicoSD?

Updating 10 to 20 variables should not require anything greater than punch cards or paper tape. In particular, don't write the contents of the variables to the SD card (or paper tape) to share them.

To share the contents of the variables use either shared memory

https://www.man7.org/linux/man-pages/ma ... iew.7.html

or

https://www.man7.org/linux/man-pages/ma ... ipc.7.html

Alternatively, mount a tmpfs in RAM

https://www.man7.org/linux/man-pages/man5/tmpfs.5.html

so you can still use a filesystem interface to share information about the variable states.

Statistics: Posted by roco2025 — Wed Nov 06, 2024 5:32 pm