Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
For most Raspberry Pi 5 projects that need local language-model (LLM) or vision-language-model (VLM) inference, choose a Raspberry Pi 5 with the Raspberry Pi AI HAT+ 2. It combines a Hailo-10H accelerator rated at 40 TOPS (INT4) with 8 GB of onboard memory, and Raspberry Pi documents support for local models up to approximately 6 billion parameters. The original AI HAT is a different product for supported vision workloads and does not support LLMs or VLMs.
The right setup still depends on your workload. The AI HAT+ 2 is intended for bounded, privacy-sensitive, latency-sensitive tasks—not as a replacement for a cloud service with broad world knowledge or long-context reasoning. Use the guide below to match the hardware, cooling, power, software and model ecosystem to what you actually want to run.
As an Amazon Associate I earn from qualifying purchases.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
At a glance
| Name | Best for | Key spec | Platforms | Checked |
|---|---|---|---|---|
| Raspberry Pi 5 + AI HAT+ 2 | Local LLM and VLM inference | Hailo-10H, 40 TOPS INT4, 8 GB onboard memory; vendor-stated support up to approximately 6 billion parameters | Raspberry Pi 5 with Raspberry Pi OS and Hailo-supported runtime/models | November 2026 |
| Raspberry Pi 5 + original AI HAT+ | Object detection, camera processing, robotics and other supported vision workloads | Dedicated AI accelerator for supported vision tasks; not compatible with LLMs or VLMs | Raspberry Pi 5 with the original AI HAT software stack | November 2026 |
| Raspberry Pi 5 without an AI HAT | General-purpose computing or investigation of CPU-based inference | No added AI accelerator; no CPU performance baseline or best model is established | Raspberry Pi 5 and your chosen operating-system/software path | November 2026 |
Which Raspberry Pi 5 setup should you choose?
Choose the AI HAT+ 2 if the project must run an LLM or VLM locally on a Pi 5. Choose the original AI HAT+ only when the job is a supported vision pipeline such as detection or camera post-processing. Choose no HAT when the Pi is primarily a general-purpose computer and AI is experimental rather than a defined requirement.
This distinction matters because the names are deceptively similar. Raspberry Pi’s AI HAT documentation explicitly separates the products: AI HAT+ 2 is the generative-AI option for LLMs and VLMs, while the original AI HAT+ is not an LLM/VLM device. Buying the cheaper or more readily available original board for a chatbot project will not produce the supported software path you expect.
Popular on Amazon
- 【90 Degree Full Function USB C Adapter】Supports the latest USB4 standard for up to 40Gbps of bandwidth, 8K video support, and up to 240W Fast charging; Support USB PD 3.1 Extended Power Range (EPR) including 140W, 180W and 240W. Your productivity is easily doubled. *Warm Tip: For the best fit, please make sure the USB-C port is fully exposed; thick or bulky cases may affect compatibility.
- 【Reliable Quality】Made from high-quality, scratch-resistant aluminum alloy, delicate metal frame. get excellent wear resistance and heat dissipation. Non-Slip Design for easier insertion and removal.
- 【Small and portable】Ultra thin size, ultra compact design, a petite physique that matches the minimalist design of your modern laptops. Coming with a slim figure that won't block the port next to the one it's plugged into.
- 【Compatibility】Widely work with most usb c, thunderbolt 4/3 devices. Compatibilit for ROG Ally, Steam Deck, Switch, MacBook Pro, Quest 3 VR headset, Surface Book 2, iPhone 18/17/16/15, Samsung Galaxy S23/S22/S21, Note 20/10, Google Pixel, ChromeBook, etc.
- 【Excellence Customer Support】If you experience any issues or cable failure, simply reach out to our dedicated customer service team. We provide a prompt response within 18 hours, providing you with the necessary technical support to resolve any concerns.
Match the board to the job
| Workload | Recommended choice | Why | Boundary to plan for |
|---|---|---|---|
| Private document question-answering over a small, fixed collection | Pi 5 + AI HAT+ 2 | Local processing can keep bounded content on the device and reduce network dependence | Knowledge is limited by the model and your retrieval/application design; broad world knowledge is not implied |
| Camera scene description or a multimodal assistant | Pi 5 + AI HAT+ 2 | The HAT+ 2 is the documented VLM-capable path | Use a model and runtime listed as supported; image size, context and response quality vary by model |
| Object detection, robotics perception or camera post-processing | Pi 5 + original AI HAT+ | It is designed for supported vision workloads | It does not support LLMs or VLMs |
| Desktop, gateway, sensor hub or software development | Pi 5 without a HAT | Fewer components and no accelerator requirement | No documented CPU-only speed or model recommendation is established here |
| Cloud-connected assistant with very long context | Pi 5 as a client, not the sole inference machine | Cloud systems are better suited to extensive context and knowledge-heavy reasoning | Local privacy and offline operation are reduced when data leaves the device |
Details checked November 2026.
What does the AI HAT+ 2 actually add?
The AI HAT+ 2 supplies a Hailo-10H neural accelerator and its own 8 GB of memory. Raspberry Pi quotes 40 TOPS at INT4 and says the board can support local models up to approximately 6 billion parameters. Those figures describe the documented hardware envelope, not a guarantee that every model of that size will load, run correctly or provide a particular response speed.
The Pi 5 remains the host computer. It runs the operating system, application logic, input and output handling, networking and coordination with the accelerator. The HAT, Hailo runtime and a supported model provide the inference path. This division means the Pi’s own RAM is still important for the operating system, camera buffers, document indexing, user interface and any other services running alongside the model.
Free tools Windows power users keep installed
One-click scans. No signup required.
Raspberry Pi’s introduction to the board contrasts typical edge models of roughly 1–7 billion parameters with much larger cloud models. Parameter count is only one consideration: quantisation, architecture, context length, tokenizer support and the Hailo software version also determine whether a model is usable. Treat “up to approximately 6 billion parameters” as a vendor-stated scale for supported models, not as a universal compatibility promise.
Why is the original AI HAT+ not the right choice for an LLM?
The original AI HAT+ targets neural-network vision jobs, including object detection, camera processing and robotics applications. Raspberry Pi states that it does not support LLMs or VLMs. It can therefore be the correct accelerator for a camera that identifies objects, but it is the wrong purchase for a text-generation assistant or an image-and-text model.
Do not infer capability from a similar name, a shared connector or the presence of an AI accelerator. Before buying, check that the model category is explicitly listed in the current AI HAT documentation and that the software guide names the runtime you intend to use. If the project requirements include “chat,” “summarise,” “answer questions” or “describe this image,” start with AI HAT+ 2 compatibility rather than the original HAT.
Can a Pi 5 without an AI HAT run local AI?
A Pi 5 without an HAT can still be useful as a general-purpose computer, and you can investigate CPU-based inference separately. However, the available Raspberry Pi material does not establish a CPU performance baseline, a preferred CPU-only model or a token-rate expectation. Do not choose this route when a predictable LLM/VLM deployment is a hard requirement.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallAlso popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
A HAT-free Pi makes sense when the AI component is optional, when the device mainly collects sensor data or serves a small application, or when you want to prototype the user interface before committing to an accelerator. Keep the architecture modular so an accelerator can be added later, and avoid promising offline response times until the exact CPU software and model have been validated by your own project.
What hardware do you need for an AI HAT+ 2 build?
Start with these components:
- Raspberry Pi 5: the official HAT+ 2 setup is built for Pi 5. Raspberry Pi lists 1 GB, 2 GB, 4 GB, 8 GB and 16 GB Pi 5 memory configurations. Select host RAM for the operating system and concurrent services; it does not replace the HAT’s separate 8 GB.
- AI HAT+ 2: it includes the Hailo-10H accelerator and its onboard memory. Use the model and runtime combinations supported in Raspberry Pi’s software documentation.
- Power: Raspberry Pi recommends a high-quality 5 V/5 A USB-C supply for Pi 5 and gives its 27 W USB-C supply as an example. Account for the HAT, storage, cameras, displays and USB peripherals when checking the assembled system.
- Cooling: the Pi 5 product documentation says the computer performs best with active cooling. For intensive AI workloads, Raspberry Pi recommends the Pi Active Cooler and fitting the HAT+ 2 heatsink supplied with the HAT.
- Storage: use the storage required by the operating system, models, documents and logs. Pi 5 exposes a PCIe 2.0 x1 interface, so an M.2 HAT or another adapter can add NVMe storage, but NVMe is optional for the HAT+ 2 software path.
- Mechanical hardware: both AI HAT models connect through the Pi 5 PCIe port with the supplied ribbon cable and mounting hardware.
There is no single correct Pi RAM size for every application. A minimal chatbot host may need less host memory than a device that also runs a web interface, camera pipeline, vector index and database. The accelerator’s 8 GB remains separate in all cases.
How do you physically install the HAT?
- Shut down and disconnect power. Never attach or remove the ribbon cable while the Pi is powered.
- Prepare the Pi 5 and HAT. Identify the Pi 5 PCIe connector, the supplied ribbon cable, spacers and mounting hardware. Keep the HAT heatsink available for installation.
- Connect the PCIe ribbon cable. Follow the orientation shown in Raspberry Pi’s AI HAT installation documentation. A reversed or incompletely seated cable can make the accelerator appear absent.
- Secure the boards. Fit the supplied mounting hardware so the HAT is supported and does not press against the Pi or other components.
- Install cooling. Fit the Pi Active Cooler and the HAT+ 2 heatsink as recommended for sustained workloads. Leave airflow around the assembly rather than enclosing it in an unventilated case.
- Connect the approved power supply and boot. Include any camera, display, USB device or storage adapter that the final application will use, then check that the supply is adequate for the total system.
The original AI HAT+ follows the same broad PCIe-and-mounting approach, but its supported workloads and software are different. Do not use the assembly instructions as evidence that the two boards have interchangeable model support.
How do you install the local-AI software?
Use Raspberry Pi’s AI software guide as the authoritative sequence because package names, operating-system releases and model support can change. The documented flow is:
- Update the operating system. Bring Raspberry Pi OS up to date and reboot when the guide requires it. Starting from a current system avoids mixing an older kernel or library with the HAT software.
- Install the documented dependencies. The guide lists the packages and configuration needed for the Hailo software path. Use those commands rather than substituting an unrelated community runtime.
- Install the Hailo Ollama server. This is the documented server layer for running supported language models through the HAT+ 2 setup.
- Select a model from Hailo’s Gen-AI Model Zoo. Begin with a model explicitly listed for the accelerator and runtime. A model that works with a desktop Ollama installation is not automatically compatible with the Hailo path.
- Reboot and verify. Follow the guide’s verification steps to confirm that the Hailo device is detected and that the selected model can be loaded before integrating cameras, documents or a custom application.
- Add a user interface only if needed. Open WebUI is optional. In the Raspberry Pi OS Trixie instructions, it runs in Docker because the interface is incompatible with Python 3.13. Installing it is therefore an additional service with its own storage, ports and maintenance requirements.
Keep the host application, Hailo server and model files on versions covered by the same documentation set. If a model disappears from the model zoo or a package changes, select another documented model instead of assuming that a generic conversion will work.
How should you plan models and memory?
Start with the smallest supported model that can answer the application’s questions. Smaller models generally leave more headroom for the operating system and other services, while a larger model may offer better language coverage but consume more memory and impose stricter compatibility requirements. The vendor’s approximately 6-billion-parameter figure is an upper scale, not a target that every project should attempt.
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Separate three kinds of memory in your design:
- HAT memory: the AI HAT+ 2 has 8 GB for the accelerator’s supported inference path.
- Pi RAM: this holds the operating system, server processes, application code, camera frames, document indexes and optional web interface.
- Storage: this holds the operating-system image, model files, documents, logs and container data. Storage capacity does not increase inference memory.
For a document assistant, keep the knowledge collection bounded and curated. For a camera assistant, define the image resolution, capture rate and response format before selecting a model. For either workload, test the exact model, prompt format and context length in the intended software stack; parameter count alone cannot predict quality or successful loading.
When are privacy and latency good reasons to use local AI?
Local inference is attractive when data must remain on the device, network access is intermittent or a response must not wait for a round trip to a remote service. Raspberry Pi describes the AI HAT+ 2 as a fit for “latency-sensitive, privacy-critical, knowledge-confined applications.” Examples include a fixed set of internal documents, a local equipment monitor or a camera workflow that should not upload frames.
Those benefits come with a capability boundary. Raspberry Pi also says applications requiring general world awareness, continuous learning, extensive context or knowledge-heavy reasoning are better suited to cloud systems. A Pi can still act as a local sensor, filter or interface while delegating selected questions to a cloud model, but that hybrid design changes the privacy and connectivity assumptions.
Write the boundary into the product specification: which data is allowed to leave the device, whether the system must work offline, how much context a request contains and what happens when the local model cannot answer. These decisions often determine the hardware choice more clearly than a headline TOPS number.
What cooling and power problems should you prevent?
AI inference is a sustained workload rather than a short boot-time task. Use the recommended 5 V/5 A USB-C supply, install active cooling on the Pi 5 and fit the HAT+ 2 heatsink. Place the assembly where air can move around it, and include the power draw of external drives, cameras and USB devices.
If the system resets, reports undervoltage, becomes unreliable after several minutes or slows under a continuous workload, check power and cooling before changing models. A supply that is adequate for a bare Pi may not be adequate for the Pi, HAT, NVMe adapter, camera and display together. Thermal and power symptoms can resemble software failures, so change one variable at a time and consult the hardware guidance linked above.
Recommended Free Tools
Also popular on Amazon
- Portable and powerful USB-C HUB: BENFEI USB Type-C HUB, with super-soft and knot-free silicone woven design cable, meets most mobile office needs. Compact, lightweight, stylish, and powerful portable USB C Hub equipped with 1 x HDMI port, 1 x 100W charging, and 3 x USB ports. 18-month warranty, 24-hour response, to ensure you feel at ease when using our product.
- Design centered on comfort and reliability: Thanks to BENFEI's end-to-end in-house cable production capability, in-house PCBA and assembly capability, using the industry's most advanced silicone woven design and process, 20cm cable in length, no knots, super-soft, the HUB is easy to use in all scenarios: laptop, tablet, stand etc. Super-soft, 25000+ life cycles, to meet your daily carrying and office needs.
- 100W Charging: Support up to 90W USB C pass-through charging via Type-C port to keep your laptop powered. 10W is reserved for other interface operations. No data and video function on the Type-C port.
- 4K HDMI Display: The HDMI port supports media display at resolutions up to 4K 30Hz, keeping every incredible moment detailed and ultra vivid. Please note that the C port of the Host device needs to support video output.
- Transfer Files in Seconds: Transfer files and from your laptop at speeds up to 10 Gbps with USB A 3.2 port. Extra 2 USB A 2.0 ports are perfectly for your keyboards and mouse.
Is NVMe necessary, and when is it useful?
No. NVMe is an optional expansion path, not a prerequisite for the HAT+ 2 software installation. Pi 5’s PCIe 2.0 x1 interface can connect to a separate M.2 HAT or adapter, which is useful when model files, document collections, logs or containers need more capacity or faster storage than your existing boot media.
No particular NVMe capacity, drive or best-value combination is established. Choose an adapter that is explicitly compatible with Pi 5, allow for its power and cooling needs, and verify that your operating-system boot and model-storage plan supports it. Do not buy NVMe merely because the AI HAT has a PCIe connection.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How do you troubleshoot a failed setup?
The HAT is not detected
Power off, reseat the ribbon cable at both ends, confirm its orientation and check that the board is mounted without mechanical stress. Verify that you are using a Pi 5 and the current software instructions for the correct HAT model. A cable or installation problem should be resolved before changing models.
The model will not load
Check that it is listed in Hailo’s supported Gen-AI Model Zoo path and that the Hailo Ollama server is installed as documented. A model can be below the approximate six-billion-parameter scale and still be unsupported because of its architecture, quantisation or runtime requirements. Try a smaller documented model and verify the accelerator before adding application code.
The system runs out of memory
Reduce the model or context size, stop unnecessary services and review host-RAM use by the operating system, Open WebUI, document indexing and camera buffers. Remember that the Pi’s RAM and the HAT’s 8 GB are separate resources; adding storage will not fix an inference-memory limit.
The device becomes unstable during long runs
Confirm the quality and rating of the USB-C supply, install the recommended active cooling and HAT heatsink, and improve airflow. Disconnect optional USB devices temporarily to isolate total-system power demand.
Also popular on Amazon
- 【Read before You Buy】 Note: This adapter is USB 2.0 version and cannot transmit HDM, VGA, or other video signals. Nevertheless, they are fully capable of charging devices,duplicating files, and can be seamlessly integrated with USB-C earphones, wireless network adapters, and SD card adapters.
- 【Easy to Use and Safe】Thanks to modern manufacturing techniques, we've made these adapters sufficiently compact so that they can remain inserted in your device without noticeable protrusion or inconvenience. This means you don't have to constantly plug and unplug them, reducing the likelihood of wear and misalignment, and making them more convenient to carry around.
- 【Reliable Quality】Engineered for enhanced durability, durable aluminum alloy shell material,get excellent wear resistance and heat dissipation,the male USB C and female USB connectors & the male USB and female USB C connectors are designed to be plugged and unplugged up to 20000 times. Provides data transfer, device charging, memory space expansion, and additional computer peripherals.
- 【Think more for you】Do you need to connect a USB-C-enabled device to an odd device that only has USB-A ports? Then these USB-A male to USB-C female adapters are made for you.Simply insert the adapter into the USB-A port on the device, and you can instantly connect your USB-C device to the device that originally didn't have the correct connection port.
- 【Trusted xiwxi Quality and Service】As a responsible brand seller, you will receive xiwxi's 12-month Quality assurance and lifetime friendly customer service, you're not just buying a product,you're investing in peace of mind with all-round quality and exceptional customer service.
Open WebUI fails on Raspberry Pi OS Trixie
Use the documented Docker deployment. The Raspberry Pi guide notes that the interface is incompatible with Python 3.13 in that configuration, so a direct Python installation is not the supported route described by the guide.
How should you budget and check availability?
Regional pricing and stock for the Pi 5, AI HAT+ 2, cooler, power supply and optional storage are not fixed by the supplied evidence. Check the current Raspberry Pi 5 product information and local distributor listings immediately before purchase. Budget for the complete system rather than the HAT alone: power, cooling, a case or mounting hardware, storage and any camera or display can materially change the total.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsAvailability can also affect the decision between an official HAT path and a HAT-free prototype. If the AI HAT+ 2 is temporarily unavailable, do not silently substitute the original AI HAT+ for an LLM project. Either wait, redesign around a cloud service or prototype the non-AI parts on a Pi 5 and add the correct accelerator later.
What is the simplest decision process?
- Name the inference type: text generation, image-and-text generation, object detection or no defined AI workload.
- Set the privacy and connectivity rule: fully offline, local by default with optional cloud escalation, or cloud-connected.
- Choose the supported software path: AI HAT+ 2 with Hailo Ollama and a model from the Gen-AI Model Zoo for LLM/VLM work; original AI HAT+ software for its supported vision workloads.
- Size the host: select Pi RAM for the operating system and concurrent services, then add active cooling and a quality 5 V/5 A supply.
- Plan storage: use existing storage unless model files, documents or containers justify a separate M.2/NVMe adapter.
- Validate the exact model: confirm support, memory fit and application context before buying additional peripherals or building a user interface.
- Document the fallback: define what the application does when the local model cannot answer, the HAT is unavailable or the network is down.
How we picked these recommendations
The choices follow Raspberry Pi’s documented product roles and setup path. The comparison prioritises workload compatibility (LLM/VLM versus vision), supported runtime and model availability, privacy and connectivity requirements, sustained power and cooling, host-memory needs, optional storage and the uncertainty around regional price and stock. No independent token-rate, response-time, energy-use or quality benchmark is available, so none is presented as a promise.
Bottom line: which setup is right for you?
For local LLM or VLM inference on a Raspberry Pi 5, buy the AI HAT+ 2 and build around its Hailo-10H accelerator, 8 GB onboard memory and supported Hailo software/model ecosystem. Add the recommended active cooling, HAT heatsink and 5 V/5 A USB-C supply, and size Pi RAM for the rest of your application. Use the original AI HAT+ for supported vision tasks only, and choose a HAT-free Pi when AI is not yet a defined requirement.
Frequently Asked Questions
Can I run a local LLM on a Raspberry Pi 5 without an accelerator?
You can investigate CPU-based inference, but the supplied Raspberry Pi material does not define a CPU performance baseline or a recommended CPU-only model. Treat a HAT-free design as an experiment or general-purpose build rather than a documented, predictable LLM deployment.
Does the AI HAT+ 2 increase the Raspberry Pi 5’s system RAM?
No. The HAT has its own 8 GB of memory for the accelerator’s inference path. Pi 5 RAM remains available for the operating system, applications, camera buffers, document indexes and optional interfaces.
Can the AI HAT+ 2 run any model that works with Ollama?
No. The supported path uses the Hailo Ollama server and models from Hailo’s Gen-AI Model Zoo. Model architecture, quantisation and runtime support matter in addition to parameter count, so generic Ollama compatibility is not enough.
Do I need an NVMe drive for local AI on Pi 5?
No. NVMe is optional expansion through a separate M.2 HAT or adapter. Add it when model files, documents, logs or containers need more capacity or faster storage, not as a requirement for the AI HAT+ 2 installation.

