Beelink SER8 Review: Specs, Price & 32GB vs 64GB (2026)

• Buy 32GB/1TB if: your target is a 7B–8B model, you run Home Assistant + Frigate + Whisper alongside it, and you want the lowest price that still does the job comfortably.
• Buy 64GB/1TB if: you plan to run 13B–14B models regularly and want them to feel as responsive as the 7B tier feels on 32 GB, or you want extra RAM headroom for running multiple services (HA, Frigate, a vector database, Ollama) without swapping.
• Both configs share the same Radeon 780M iGPU and DDR5-5600 memory bandwidth — 64 GB does not make inference faster, it only lets larger models fit.
• The 64 GB SKU currently costs roughly $400–450 more than the 32 GB SKU — worth it only if the larger model tier is the actual goal, not "just in case."

• No discrete GPU: all AI inference runs on the CPU and Radeon 780M iGPU sharing system RAM.
• Dual SO-DIMM slots are user-replaceable. Beelink's own spec sheet lists a 256 GB maximum; in practice most buyers reach 64–96 GB with commonly available kits, and higher-density kits are not independently verified here — treat the manufacturer figure as a ceiling, not a tested number.
• Idle power draw: roughly 8–12 W by most independent reviews (not independently re-measured for this update) — cheap to leave on continuously as a home server.

• Ollama, LM Studio, and llama.cpp all run on the SER8; the 780M is used through Vulkan or ROCm backends depending on the tool.
• A 7B–8B model at Q4 is the comfortable sweet spot on either RAM config — this is the workload the SER8 is genuinely good at.
• For 14B models: use 64 GB if you want the model to feel as responsive as the 7B tier does on 32 GB; it works on 32 GB but competes with Home Assistant and Frigate for memory.
• 32B-class models fit only tightly even at 64 GB and run slowly — if 32B is your actual target, buy a discrete-GPU system instead (see the GPU-server comparison).
• The 16-TOPS NPU is not used by mainstream LLM runtimes (Ollama, llama.cpp) in 2026 — see the NPU note below.
• See Ollama on Home Assistant to connect the model, and the local LLM hardware guide for VRAM depth on other platforms.

• What actually determines local-LLM speed on this hardware: the Radeon 780M's compute throughput and DDR5-5600 memory bandwidth, since the iGPU and CPU share system RAM.
• Don't buy the SER8 (or any Ryzen AI mini PC) expecting NPU-accelerated chat — buy it for the CPU/iGPU/RAM combination, which is the part that actually runs your model.
• This also means the newer SER9/SER10's higher NPU TOPS ratings do not translate into meaningfully faster local-LLM inference for this exact use case — a genuine reason the older SER8 remains a fair value pick.

• Home Assistant OS or Home Assistant Container runs comfortably alongside the other services — CPU and RAM headroom are the bottleneck, not raw compute.
• Frigate (camera object detection) benefits from a hardware accelerator — the SER8's iGPU can decode video, but a dedicated accelerator like the Hailo-8L (see the bundle below) is the better fit if you run more than 2–3 cameras.
• Whisper (speech-to-text) for voice commands runs fine on the CPU for a single-household setup; it does not need the iGPU.
• Ollama serves the local LLM (7B recommended) that Home Assistant's Assist pipeline can call for natural-language automations and Q&A.
• Build this exact setup: see Connect Ollama to Home Assistant and Running Your Smart Home on a Local LLM.

• Mini PC: SER8 32GB/1TB — the hub for Home Assistant, Ollama, Whisper, and Frigate.
• Zigbee coordinator: a USB dongle for Zigbee devices (locks, sensors, bulbs) — the SONOFF ZBDongle-E below is the one already recommended on this site's smart-home hub roundup.
• PoE cameras: Reolink RLC-810A is a straightforward starting point for Frigate — PoE means one cable per camera for power and network.
• Frigate accelerator: if you run more than 2–3 cameras, add a Hailo-8L M.2 module — it offloads object detection so the SER8's CPU stays free for Home Assistant and the LLM.
• Additional storage: either M.2 2280 PCIe NVMe SSD fits the SER8's second slot (up to 8 TB per Beelink's spec) — pick capacity based on how many days of Frigate footage you want to retain.
• Software: Home Assistant and Ollama are both free and open-source — no subscription cost beyond the hardware.

• 32 GB DDR5 (upgradeable to 64 GB): enough to run Home Assistant, Frigate, Whisper, and a 7B model together without constant memory pressure.
• User-replaceable RAM: unlike soldered-RAM competitors, you can start at 32 GB and move to 64 GB later without buying a new machine.
• 2.5GbE networking: matters for a home server pulling camera streams and serving Home Assistant's dashboard to multiple devices — a 1GbE port would be the bottleneck first.
• Two M.2 slots: run the OS on one drive and dedicate the second entirely to Frigate footage, so recordings don't compete with the OS drive for I/O.
• Radeon 780M iGPU: strong enough for 7B-model inference and Frigate's video decode without a discrete GPU's power draw or noise.

Why I like it

• Best value for a capable local-AI box — $799–849 for a machine that comfortably runs a 7B model plus a full smart-home stack
• User-replaceable RAM and two M.2 slots mean the machine grows with you instead of forcing a replacement
• 2.5GbE and USB4 cover the actual bottlenecks of a home server (network throughput, external storage) that cheaper boxes skip
• Runs at 65 W under load and roughly 8–12 W at idle — inexpensive to leave on 24/7

• GEEKOM A9 Max — the headroom pick: 128 GB RAM ceiling and Wi-Fi 7, but now $1,099–1,799 (2026 Edition, Ryzen AI 9 HX 370/HX 470) — more than double, often triple the SER8's price.
• Minisforum UM890 Pro — similar Radeon 780M performance, but adds dual 2.5GbE, Wi-Fi 6E, and an OCuLink eGPU port for a modest premium.
• Beelink EQ14 — budget Intel N150 box for Home Assistant plus a tiny model only; skip if you want a 7B assistant.
• GMKtec G3 Plus — another budget N150 option with upgradeable RAM.
• Beelink's own SER9 and SER10 Pro/Max (launched CES 2026, Ryzen AI 9 HX 470, 10GbE on some SKUs) are the current flagship line — worth it only if 10GbE or the newer NPU generation matters to you specifically; for HA + Frigate + a 7B model, they do not change the recommendation.
• Still comparing? Start from the best mini PCs for Home Assistant + local AI roundup.

• Beelink official store: direct manufacturer warranty and support; occasionally the highest price of the two channels.
• Amazon: usually the more competitive price, plus Amazon's standard return policy — the practical default for most buyers.
• Confirm the RAM SKU (16, 24, 32, or 64 GB) and the CPU (8845HS, not 8745HS) on the listing before buying — Beelink sells multiple variants under the same "SER8" name.

• RAM upgrade to 64 GB: a DDR5 SO-DIMM kit sized for the SER8's two slots — check the current kit price and confirm DDR5-5600 compatibility before buying.
• Additional NVMe SSD: any M.2 2280 PCIe 4.0 drive fits the second slot — useful for dedicating storage to Frigate footage separately from the OS drive.
• Zigbee coordinator: needed if you add Zigbee devices (locks, sensors, bulbs) beyond what Wi-Fi/Matter devices already cover.
• Hailo-8L accelerator: add this once you have more than 2–3 Frigate cameras and object-detection starts competing with the LLM for CPU time.

• This total excludes PoE switch/injector and cabling, which vary by home and are usually a one-time cost shared across other network devices.
• Skip the Hailo-8L if you run 1–2 cameras only — the SER8's iGPU handles light Frigate loads without it.

• Build a GPU workstation instead if 32B+ is a hard requirement: see the local-AI workstation build guide for a discrete-GPU system spec'd for that tier.
• For everything else on this page — Home Assistant, Frigate, a 7B–14B assistant — the SER8 is the better value and the quieter, cheaper machine to run continuously.

• US: Section 301 tariffs on Chinese electronics remain in force in 2026; the separate 2025 "IEEPA" tariffs were struck down by the Supreme Court in February 2026 and replaced by a temporary, capped Section 122 tariff. The sub-$800 duty-free "de minimis" exemption for direct imports has also ended.
• EU (affects DE/FR): there is no broad EU tariff on finished mini PCs, but from July 2026 the €150 duty-free threshold on low-value direct-from-China parcels was removed and a small per-parcel handling fee added.
• Assumption (macro trend, not a per-product fact): redirected Chinese export capacity has kept availability of these brands high in the EU and US, which broadly supports competitive pricing.
• Net effect: verify the current price at the retailer before buying — the ranges on this page are indicative and date-stamped August 2026.

• If the SER8 32GB/1TB is priced within ~$100 of a GEEKOM A9 Max sale price, the A9 Max's 128 GB ceiling and Wi-Fi 7 likely justify the difference for anyone who might grow into larger models.
• If the SER8 and SER9 are priced close together, check whether the SER9's spec improvements (verify current SER9 pricing and specs at purchase time) matter enough to you — otherwise the SER8 remains the value pick.
• Don't buy on brand momentum alone — the recommendation on this page is tied to the 32GB/1TB config at roughly $799–849; re-check before purchase.

• Best Mini PCs for Home Assistant + Local AI — the roundup this review belongs to
• Local-AI Workstation Build Guide — the GPU-server path for 32B+ models
• Local LLM Hardware Guide — cross-cluster: VRAM and quantization depth
• Running Your Smart Home on a Local LLM — what the box actually runs
• Connect Ollama to Home Assistant — wire the model in
• Best Hardware for a Local Smart Home — Pi vs mini PC vs NAS

• YES, buy the 32 GB/1 TB SER8 if: you want Home Assistant, Frigate, and a 7B–8B local assistant on one quiet, inexpensive box — this is the default recommendation for most readers of this page.
• YES, but choose 64 GB/1 TB if: 13B–14B models are your actual target and you want them to feel responsive, not just technically loaded.
• NO if: 32B or 70B models are your primary use case — buy a discrete-GPU system instead; see the local-AI workstation build guide.