TL;DR

Raspberry Pi introduced the AI HAT+ 2, a $130 add-on for the Raspberry Pi 5 that brings 8GB of onboard RAM and a Hailo 10H accelerator rated at 40 TOPS. The board can run and fine‑tune small generative AI models locally, though independent testing found a stock Pi 5 sometimes outperforms the HAT due to power limits.

What happened

Raspberry Pi unveiled the AI HAT+ 2, an upgraded accessory for the Raspberry Pi 5 priced at $130. The module adds 8GB of dedicated RAM and includes a Hailo 10H AI chip with a stated 40 TOPS of inference performance. When attached, the add-on can take on AI workloads to free the Pi’s main Arm CPU for other tasks. Raspberry Pi demonstrated the HAT+ 2 running small generative models — including Llama 3.2, DeepSeek‑R1‑Distill and multiple Qwen variants — to generate scene descriptions from camera input, detect people in video feeds, and perform French-to-English translation with Qwen2. The company also said larger models are being prepared for future updates. However, reviewer Jeff Geerling reported that a standalone Raspberry Pi 5 with 8GB RAM often beat the HAT+ 2 on supported models, a gap he links to the HAT’s lower power ceiling compared with the Pi 5.

Why it matters

  • Enables local execution of small generative AI models on a low-cost single-board computer.
  • Offloads AI processing from the Pi 5’s CPU, potentially leaving the main board available for other tasks.
  • Signals continued focus on edge AI hardware and software support for model deployment on tiny systems.
  • Performance and value depend on power limits and system RAM; a higher‑RAM Pi may be a better option for some users.

Key facts

  • Product: AI HAT+ 2 add-on for Raspberry Pi 5.
  • Price: $130.
  • Hardware: Hailo 10H accelerator rated at 40 TOPS and 8GB of onboard RAM.
  • Supported models (announced): Llama 3.2, DeepSeek‑R1‑Distill and a series of Qwen models, including demonstrations with Qwen2.
  • Capabilities: Can run, train and fine‑tune small generative AI models locally.
  • Demo use cases shown: text descriptions of camera streams, person‑presence detection, and French→English translation.
  • Independent testing by Jeff Geerling found a Raspberry Pi 5 with 8GB RAM usually outperformed the HAT+ 2 on supported models.
  • Performance gap attributed to power: Pi 5 can draw up to 10W, while the AI HAT+ 2 is limited to 3W.
  • Previous model: original AI HAT focused on image processing and started at $70.

What to watch next

  • Availability and shipping dates for the AI HAT+ 2: not confirmed in the source.
  • Release of the “larger” models Raspberry Pi says are being readied for post‑launch updates.
  • Independent benchmarks comparing HAT+ 2 performance to higher‑RAM Pi 5 configurations across more models and workloads.

Quick glossary

  • HAT: A Hardware Attached on Top: a standard form factor for Raspberry Pi add‑on boards that expand capabilities such as sensors or accelerators.
  • TOPS: Trillions of operations per second, a unit used to express the raw inference throughput of AI accelerators.
  • Generative AI: AI systems that produce content — such as text, images or code — from learned patterns in data.
  • Fine‑tuning: The process of adjusting a pre‑trained model on a smaller, task‑specific dataset to improve performance on that task.

Reader FAQ

What does the AI HAT+ 2 cost?
$130.

Which Raspberry Pi is it designed for?
Designed as an add‑on for the Raspberry Pi 5.

Can it run popular small generative models?
The company demonstrated support for Llama 3.2, DeepSeek‑R1‑Distill and several Qwen models, and said more models are being prepared.

Does the HAT+ 2 outperform a Pi 5 with more RAM?
Independent testing reported a stock Pi 5 with 8GB sometimes outperformed the HAT+ 2 on supported models; reviewers noted the HAT’s 3W limit as a factor.

When will it be available to buy?
not confirmed in the source.

NEWS AI GADGETS Raspberry Pi’s new add-on board has 8GB of RAM for running gen AI models The Raspberry Pi 5 can shift AI-related workloads to the $130 AI HAT…

Sources

Related posts

By

Leave a Reply

Your email address will not be published. Required fields are marked *