Tayschrenn
Young grasshopper
I've been running my 4 cameras with BI on an Optiplex 7070 MT with an i5-9500 and 16gb ram, with an Intel PCI nvme for docker (AI), DB, alerts and first storage.
What I've found is that really, I need a GPU.
This means I'll have to either upgrade my m.2 to something more write intensive (not an issue really) and move first storage to my 10tb HDD, and use the pci slot for a gpu...
Or... And this might be crazy but I wanted to get the communities thought on it: repurpose an old Precision laptop that has an i7 or Xeon cpu with an on board quadro to handle the AI, and possibly more. I could even use an enterprise SATA SSD for the higher write endurance in it for initial storage.
Then I could turn my Optiplex into more of a proxmox or vmware host with network storage to offload to just like I do now from the pci nvme to my HDD. It'd be slightly slower, but I do have a use case for that box by throwing a 2.5gbps card into it and running pfsense to replace my aging ubiquiti USG (which can't handle my 1.2gbps internet anyways)
The main reason I'm considering this is that we're going to be replacing A LOT of Precision laptops over the next 12-18 months at my office, and the vast majority will have decent specs by BI standards (we're just moving to a firm 5-year life cycle on all hardware and I'm pushing to allow employee purchase vs recycling). Plus, the optiplex has a 75w max pci slot with no gpu PSU cable, and a proprietary PSU... Which I found out after getting an older quadro being recycled that wouldn't work in it, so I'd be limited to maybe a 1650 anyways.
I can't find much in regards to supported models when it comes to Quadro for Codeproject.ai or many running them, but it feels like they'd likely be great for the task. The xeons we run all have the iGPU still as well to offload the BI encoding there.
What I've found is that really, I need a GPU.
This means I'll have to either upgrade my m.2 to something more write intensive (not an issue really) and move first storage to my 10tb HDD, and use the pci slot for a gpu...
Or... And this might be crazy but I wanted to get the communities thought on it: repurpose an old Precision laptop that has an i7 or Xeon cpu with an on board quadro to handle the AI, and possibly more. I could even use an enterprise SATA SSD for the higher write endurance in it for initial storage.
Then I could turn my Optiplex into more of a proxmox or vmware host with network storage to offload to just like I do now from the pci nvme to my HDD. It'd be slightly slower, but I do have a use case for that box by throwing a 2.5gbps card into it and running pfsense to replace my aging ubiquiti USG (which can't handle my 1.2gbps internet anyways)
The main reason I'm considering this is that we're going to be replacing A LOT of Precision laptops over the next 12-18 months at my office, and the vast majority will have decent specs by BI standards (we're just moving to a firm 5-year life cycle on all hardware and I'm pushing to allow employee purchase vs recycling). Plus, the optiplex has a 75w max pci slot with no gpu PSU cable, and a proprietary PSU... Which I found out after getting an older quadro being recycled that wouldn't work in it, so I'd be limited to maybe a 1650 anyways.
I can't find much in regards to supported models when it comes to Quadro for Codeproject.ai or many running them, but it feels like they'd likely be great for the task. The xeons we run all have the iGPU still as well to offload the BI encoding there.