Comply with ZDNET: Add us as a most well-liked supply on Google.
ZDNET’s key takeaways
- The upgraded AI HAT+ 2 boosts efficiency to 40 TOPS.
- The board comes with 8GB of onboard RAM.
- The ban might have a big affect on companies and jobs.
For those who’re into Raspberry Pi computer systems and AI, then likelihood is that you just’re already utilizing the AI HAT+ that was launched again in 2024. This was a game-changer when it comes to offloading AI workloads onto the Pi 5. The thought is that this can be a low-cost, data-private approach to course of edge AI workloads with no need to make use of third-party cloud-based AI companies.
Properly, in the present day sees that board get a big improve with the discharge of the AI HAT+ 2.
Two variations
The AI HAT+ got here in two variations — a regular 13 TOPS (tera-operations per second) model powered by a Hailo-8L AI accelerator chip, and a 26 TOPS model powered by the Hailo-8 AI accelerator. The brand new AI HAT+ 2 boosts this efficiency to 40 TOPS because of using the Hailo-10H chip backed up by 8GB of onboard RAM.
This efficiency increase permits the AI HAT+ 2 to deal with giant language fashions (LLMs), vision-language fashions (VLMs), and different generative AI workloads domestically, unbiased from the Raspberry Pi 5 board. It could possibly additionally deal with vision-based AI workloads similar to object recognition and pose estimation at about the identical 26 TOPS efficiency as that of the Hailo-8-powered AI HAT+.
Those that have already constructed initiatives across the AI HAT+ and need to reap the benefits of the efficiency increase that the AI HAT+ 2 has to supply, concern not, as a result of based on Naush Patuck, Senior Principal Engineer at Raspberry Pi, “transitioning to the AI HAT+ 2 is generally seamless and clear.”
And for these needing inspiration or data on tips on how to make use of all this AI energy, Hailo’s GitHub repo has every little thing you’ll want to get began with generative AI. There are additionally a collection of LLMs out there to put in at launch, together with Llama3.2, DeepSeek-R1-Distill, and Qwen2, with extra out there after launch.
Availability
This board works with all Raspberry Pi 5 boards (from 1GB all the best way to 16GB), connects by way of the onboard PCI Specific interface, and integrates immediately with Raspberry Pi’s digital camera software program stack, avoiding further configuration trouble.
The AI HAT+ 2 board is shipped with an non-compulsory heatsink, however based mostly in your expertise of the earlier AI HAT+, I would advocate becoming this, particularly if operating demanding duties or benchmarking, as a result of the processor can get fairly scorching beneath heavy loading.
The board is offered instantly for $130, though on the time of writing ZDNET solely discovered inventory at PiShop (this outlet additionally sells a whole AI HAT+ 2 starter equipment).
