A few days after Raspberry Pi “leaked” the existence of a Raspberry Pi 500 via a Git commit, we seemingly have another new product seeing the light of day some point soon.
A new Hailo 8 (this one omits the L which is found on the Raspberry Pi AI Kit) overlay has been spotted in a new Git commit which would imply that we’re going to have a new HAT that uses the more powerful Hailo 8 module, rather than the cut-down Hailo 8L.
Table of Contents
What’s The Difference? Should I care?
The main difference between the 8 and the 8L is that the Hailo 8L has “half of the compute elements”, so it looks like we’ll be getting a HAT with a beefier NPU in the near future as the Hailo 8 has 26 TOPS onboard, as opposed to the 13 on the 8L.
The other difference is in the form factor, as the Hailo 8 only comes in an M-Key form factor, though it remains to be seen if Raspberry Pi is simply going to re-use the AI Kit’s M.2 HAT, or if there’ll be some special sauce for the newer model. It would be nice for users if they could simply grab the new module but it will likely only come in kit form I imagine.
Pricing?
Given it’s going to be packing twice as many TOPS under the hood, and the modules themselves are considerably more expensive, the pricing is going to be an interesting topic.
The Raspberry Pi AI Kit retails at $70 + taxes/shipping (USD) depending on where you are, so it’s safe to say that we’ll see a fairly hefty price difference between the 2, unless Raspberry Pi can use its scale to negotiate some nice volume discounts.
Thoughts on the Raspberry Pi Hailo 8
Overall it seems nice that we’ll be getting additional options, though I’m curious to see what kind of use cases people are buying these for. I’ve done some basic Frigate work, and even the Coral TPUs have been sufficient for me so I have no horse in this race.. What are people looking to use this much power for?
I’m serious, and for once I’m not being snarky! I’m oblivious to most of this ecosystem and genuinely want to know what people are using these for that current solutions for the Pi 5 can’t provide. Please help this idiot out..
1 comment
Use case most interesting to me would be to assist ROS 2 vSLAM (visual simultaneous localization and mapping) with RTABMap processing 3D Stereo depth camera images from the 4TOP neural processor in my Oak-D-W97 camera .
Right now my Raspberry Pi5 barely keeps up with processing the images into a 3D map and recognizing when it has seen that area before (localization). Most folks are doing vSLAM off-board the mobile robot where they have a strong GPU. I always program my robots to be truly autonomous – no distributed processing, no cloud processing. Power is getting a bit tight though – I’m only getting 2 hours “playtime” hours with the “33Wh battery” (actually only get 21Wh out of it) for the Pi5, Oak-D-W camera, and LIDAR.