Working combo, Asrock Deskmini A300 (AMD) and m.2 Coral TPU

honkkis

n3wb
Joined
Apr 9, 2023
Messages
4
Reaction score
9
Location
Finland
Installed yesterday single TPU m.2 Coral into my Asrock Deskmini A300 cameraserver, which has A+E m.2 slot for wifi. I used to have earlier USB-version of Coral, but this is even better! Getting around 75ms object recognition (average) on low power Ryzen system in mini-STX form factor computer is quite cool, even more so with very little energy consumption ;) Running several 3MP and 4MP cameras and single 8k, motion analysis from substream (it does little miracles for cpu load) without problems. Ryzen 3 3200G, 32Gbb RAM, 128Gb m.2 system disk and 500GGb 7200rpm conventional hard disk for recordings, Win11 Pro and newest BI, CodeProject.AI (Coral-module / small).
 

honkkis

n3wb
Joined
Apr 9, 2023
Messages
4
Reaction score
9
Location
Finland
And one big benefit is stability: USB-version of Coral used to crash CodeProject.AI server (2.1.11 beta) at least daily, usually many times a day, but m.2 version of Coral is really stable, no more AI crashes.

Availability and prices of m.2 Coral are good at least in Europe, I got mine in a week at recommended price from German retailer. For USB-version, I had to wait couple months :rolleyes: So, if you have m.2 slot in your computer,m.2 Coral is definitely worth considering.
 

harleyl7

Pulling my weight
Joined
Jun 4, 2021
Messages
260
Reaction score
223
Just purchased an m.2 in the US via eBay for 69 bucks. Birthday gift from someone else so they didn't mind paying the high fee. Excited to test it out with CPAI on my unRAID server via a VM vs adding it to my BI baremetal server
 

honkkis

n3wb
Joined
Apr 9, 2023
Messages
4
Reaction score
9
Location
Finland
After having M.2 Coral for more than five months now, my experience is that interference times have improved a lot as CP.AI gets new releases! In my place, we have usually lot of snow (2023 in November!), that's variable, but improvement was seen even before that. Earlier, there used to be some +100ms and lot of 60ms...90ms interferences, now it seems to be generally below 20ms as reported by CP.AI server tools. I think it's about same that used to be reported for upper tier nVidia display cards while ago. Of course, models are smaller, but so is the price and power consumption! CP.AI and M.2 Coral is definitely heading to right direction! And it does its work with high confidence.
 
Top