Well there is the asus AI Accelerator card but that’s just 8 coral TPUs. I think the real reason why we don’t see large TPUs is because nvidia cards has had tensor cores built into the architect since Volta and with a GPU you don’t have to worry about system memory speed since you have 80Gb of HBM.

because studio equipment uses 3.5mm or other standard jacks (XLR for microphones for example) as they cause the lowest interference.

Digital signaling is not susceptible to interference like analog signaling. Comparing three analog connectors to a digital signal is a false comparison. With a digital signal unless the interference is large enough to sway the voltage to the wrong side of the threshold it doesn’t matter as it will still register a one or a zero. Analog signaling on the other hand is very susceptible to interference unless you use balanced connections which uses wave interference to remove the added noise.

Avarage load for me is around 300w running two T320s, a R510, and a NUC. The T320s are clustered running plex, 'arrs, pihole/unbound, game servers, and odoo. The nuc serves two purposes, firstly to keep quorum in the cluster and secondarily as a low power device to run a secondary pihole/unbound instance incase the power goes out as it’s the only server that will stay on UPS power until the battery runs out. The R510 is my storage server with around 56TB and growing.

I am planning on adding a GPU server with a few tesla P40s as I’ve been using my workstation for these tasks which makes it difficult to use it for work.

ZFS not have access to smart only works up until a drive starts acting up. Without SMART ZFS can’t accurately determine if a drive is failing and lock the pool in order to prevent further data loss.

The best way to do this is to run a truenas VM within proxmox and passthrough your HBA into the truenas VM. That will give truenas full control over any drive connected to that HBA. The performance overhead isn’t that much so don’t worry about it.