Is there an NVIDIA Jetson ORIN or IGX Orin which is available as an edge device, like rugged IOT but with BMC. Ideally, it would be small enough to be mounted on a DIN rail, for example, and also expandable in memory (SSD, NVE, etc.) if more data is to be stored on the edge device.
I had already asked the support once and was referred to the forum, where only rack hardware was suggested to me, since no further answer was added when I formulated my question more precisely, I am now opening a new blog entry. (Unfortunately, centralized training is not possible)
Jetson Orin Hardware useable with Canonical MAAS
BMC should be able to be addressed in encrypted form, like Redfish or Intel AMT.
Furthermore, network boot with encrypted data transfer should be possible, as with IPXE Crypto or UEFI HTTPS. It should be possible to provision new images, whereby cloud-init or other preseed packages can also be used here.
Tools for this are for example Tinkerbell, Foreman, Openstack Ironic (Canonical MAAS unfortunately only unencrypted so far? HTTP boot in MAAS - Users - Canonical MAAS | Discourse)
Training tasks for predictive maintenence and predictive production as well as inference should then take place on these devices as Kubernetes workers, whereby the devices train a model in a federated manner.
It can of course also be other hardware, which is CUDA compatible and small (Edge iot) and equipped with BMC?
In terms of price, it would be wonderful if something like this were available for less than 1000 or 1500 dollars.
Perhaps someone already has experience of which devices are available with which functions, or what hardware could be used.