So we got great news from one of our Alpha testers:
They are using the DepthAI: System on Module (SoM) in their own custom board.
To de-risk the development of their custom board, they implemented the board to support our SoM and also the incumbent Myriad X PCIE board (the manufacturer of which will rename nameless here).
What did they find?
Running the exact same neural inference over USB3, our module uses 1W (yes, 1 Watt!) less power than the incumbent board. They also had to work around instability with the incumbent board, where it would regularly hard-crash, requiring full-reboot of their host Linux system.
This crash would manifest in two ways:
First, if they started performing full-bore neural inference within 1 minute of system powerup, it would crash almost every time.
To solve that, they implemented a delay running neural inference until 1 minute after startup (which was materially detrimental to their use case).
But worse, even with this delay, the system would randomly crash within a couple hours, requiring a full system reboot.
This was killing them.
They spent a long time debugging it, assuming that this was an error on their custom hardware design or software flow.
It surely couldn’t be the fault of this large not-to-be-named OEM who made the incumbent Myriad X PCIE card, right?!
Eventually they gave up on this incumbent PCIE board and swapped in our DepthAI SoM in its place.
Zero crashes, ever.
To this date they haven’t seen a single crash with DepthAI, where they were seeing them hourly with the incumbent!
So in the end; 1W less power, and infinitely better stability. And this is for the Alpha version of DepthAI, literally the very first boards we ordered (and we’re on board order 6 now, with improvements each time).
We were super excited to hear this, and figured you all would be too.
Brandon & the Luxonis team.
An open video development board in a PCI express form factor that supports overlaying content on encrypted video signals. Let's bring open video to the digital age!
A high-performance SDR seamlessly integrated with state-of-the-art deep learning hardware