Combined workload sizing calculator
Home Assistant + Frigate AI Hardware Calculator
This calculator is the decision engine for the cluster. It does not assign one mini PC to everyone. It scores the Home Assistant footprint, Frigate camera/decode/detector load, local voice and Ollama model class, then adds memory, storage and virtualization headroom before recommending a host tier and acceleration architecture.
Quick answer
What this combined workload needs
The result should tell you whether an appliance-class N100/N150 is enough, whether N305/Core-class hardware is safer, or whether local AI has moved the design into 32–64GB Core Ultra/Ryzen AI or GPU-capable territory. It also identifies when splitting Ollama onto a second system is the more reliable answer.
Live Amazon hardware
Current products that fit this decision
Compare current Amazon listings relevant to this guide. Product availability and prices can change.
Buying decision
Match the hardware to the bottleneck
Use the calculator before shopping. Then compare the live products that match the resulting tier. After deployment, replace every planning assumption with measured camera bitrate, detector latency, RAM usage, Ollama memory/VRAM allocation, disk growth and wall power.
Interactive decision tool
Home Assistant + Frigate AI Hardware Calculator
Size the combined workload before comparing products. Results are planning tiers, not benchmark guarantees. Verify current Home Assistant installation support, Frigate detector/device access and Ollama hardware support for the exact system.
Compatibility checklist
Four checks before you purchase
Assign every accelerator a job
Video decoding, Frigate detection, enrichments and Ollama inference are separate workloads. Do not assume one NPU/GPU automatically covers all of them.
Protect Home Assistant headroom
Size the combined host for peak overlap so camera or LLM activity cannot starve automations, databases or radio services.
Verify device access
USB radios, iGPU/NPU/GPU and Hailo devices must be visible to the correct host, container or VM before their benchmark performance matters.
Keep data roles separate
Use fast SSD space for system/models and size Frigate recordings independently so retention cannot crowd the smart-home environment.
The calculator begins with the workload mix
Home Assistant-only, Home Assistant plus Frigate, and the full Home Assistant + Frigate + local AI stack have completely different resource envelopes. Selecting the workload mix changes the CPU, memory and accelerator baseline before any camera or model inputs are applied.
This prevents a local-AI requirement from being hidden behind a generic smart-home recommendation.
Camera count is adjusted by activity and video complexity
Frigate camera count matters, but detector demand follows simultaneous activity and the video path depends on codec, resolution and hardware decoding. The calculator increases the tier for higher activity and 4K-heavy deployments instead of assuming all cameras cost the same.
Use the camera manufacturer’s real configured stream settings after installation to refine the estimate.
Detector selection changes CPU and interface requirements
A supported Hailo/OpenVINO/GPU detector can move inference away from general CPU cores. That reduces CPU pressure but creates a slot, passthrough and driver requirement.
The result therefore includes a detector direction rather than simply adding more CPU every time the camera count grows.
Local AI is scored from model class and responsiveness target
A tiny local model used occasionally is fundamentally different from a larger conversational model expected to answer voice queries quickly. The calculator uses model class, context/usage and whether GPU acceleration is desired to raise the memory and host tier.
The output is intentionally a planning category, not a promise of tokens per second. Ollama performance changes with model, quantization and exact hardware.
Voice AI does not automatically mean large LLM
The tool separates local voice pipelines from full Ollama conversation workloads. A household can keep speech-to-text and text-to-speech local while using deterministic Home Assistant intents, avoiding the memory cost of a large model.
Choose the local LLM option only when you actually want open-ended conversation or tool-assisted reasoning.
RAM is calculated from the peak overlap
Home Assistant, Frigate and Ollama can all be active during the same event. The calculator therefore leaves a reserve instead of assigning every byte to the model.
Virtualization and many extra containers add more headroom. The result may recommend 64GB even when the model alone appears to fit inside 32GB.
Storage is split into fast system/model space and recordings
The calculator estimates a system/model SSD tier independently from Frigate recording capacity. That keeps Ollama downloads and Home Assistant updates from being squeezed by a rolling NVR volume.
Use the dedicated Frigate storage calculator for exact retention TB; this combined tool focuses on the host-side SSD requirement.
Radio requirements influence deployment, not CPU
Zigbee, Thread and Z-Wave do not meaningfully raise CPU tier, but they affect USB passthrough and failure domains. The calculator flags when virtualization plus multiple USB coordinators may make a network-radio strategy easier.
That is a hardware recommendation even though it does not involve a faster processor.
Networking is evaluated at the combined-server level
The host may serve camera live views, NAS recordings, model downloads, backups and Home Assistant clients at once. A 2.5GbE recommendation appears when those aggregate flows justify it.
It is not presented as a requirement for ordinary Home Assistant traffic or individual IP cameras.
Virtualization adds a resource tax and device map
Proxmox or another hypervisor adds host memory and makes iGPU/GPU/USB ownership explicit. The calculator raises memory targets and may recommend separate accelerators when one physical GPU cannot be shared cleanly.
If device passthrough is the hardest part of the build, the tool can recommend bare-metal Docker instead of simply choosing a more expensive mini PC.
Power target can change the preferred architecture
A low-power goal favors integrated media engines, NPUs or dedicated accelerators over a large discrete GPU when the workload allows it. A performance-first local AI goal may justify a GPU and higher idle draw.
The correct answer is based on annual usage and required responsiveness, not a moral preference for low or high power.
Production measurements should replace the estimate
After the system has run through busy camera periods and real local-AI sessions, inspect CPU, memory, GPU/NPU usage, detector latency, disk growth, network throughput and wall power.
If one resource is clearly underused, do not upgrade it. If one is saturated, the calculator’s architecture makes it easier to target that bottleneck instead of replacing the entire stack. Save a baseline after commissioning, then compare it again after adding cameras, voice features, larger models or new Home Assistant apps. Capacity planning is most useful when it becomes a repeatable measurement process instead of a one-time purchase guess. Record the measured baseline with the server model, firmware, Frigate version, Ollama model and Home Assistant workload so later changes can be compared against real evidence.
Questions people ask
Home Assistant + Frigate AI Hardware Calculator questions
What does the calculator size?
CPU/host tier, RAM, detector direction, GPU/NPU need, system SSD, networking and deployment architecture for the combined Home Assistant, Frigate and local AI workload.
Does it calculate exact Frigate recording terabytes?
It estimates the host/storage tier and points to the dedicated Frigate storage calculator for detailed retention math.
Does it guarantee Ollama speed?
No. Model, quantization, context and exact GPU/CPU determine real performance. The output is a planning tier.
Can it recommend no GPU?
Yes. A small Home Assistant/Frigate setup can use iGPU + dedicated detector, and light local AI can run on CPU.
Can it recommend a separate Ollama server?
Yes when local AI demands would make the always-on smart-home/NVR box too expensive, hot or complex.
Why does virtualization increase RAM?
The hypervisor and guest operating systems consume memory, and Frigate GPU passthrough benefits from predictable reserved memory.
Why does it ask about radio devices?
Radios affect USB/network passthrough and placement even though they do not change the CPU tier much.
When does 2.5GbE become useful?
When the combined host moves recordings, backups, model files and several client streams rather than only Home Assistant traffic.
Should I use the maximum recommended tier?
Not automatically. Choose the smallest tier that has comfortable headroom and the interfaces you need.
What should I measure after deployment?
CPU, RAM, decoder usage, detector latency, GPU/NPU utilization, model memory, disk growth, network load and wall power.
Official references and methodology
Verify current Home Assistant, Frigate and Ollama support
Cloudzat treats Home Assistant, Frigate and local AI as separate workloads that share one hardware envelope. Home Assistant installation guidance, Frigate device-access and detector guidance, and Ollama GPU/context documentation are used as the technical baseline. Calculator outputs are planning tiers rather than benchmark guarantees, and live Amazon listings come only from this sprint's dedicated catalogue.
- Home Assistant - Installation
- Home Assistant - Ollama integration
- Home Assistant - Fully local voice assistant
- Frigate - Recommended hardware
- Frigate - Installation and Proxmox guidance
- Frigate - Home Assistant integration
- Ollama - Hardware support
- Ollama - Context length and memory
- Ollama - Docker GPU setup
As an Amazon Associate, Cloudzat may earn from qualifying purchases. Prices, seller terms, exact configurations and Home Assistant, Frigate or Ollama hardware support can change.