Self-Hosting

Pick a mini-PC home server by workload

Choose an Intel N-series mini-PC by what it'll actually run — N100/N150 (4-core, ~6W) for Pi-hole/Home Assistant + light transcoding, N305/N355 (8-core) for heavier multi-stream — plus the single-channel-16GB ceiling and the QuickSync AV1 decode-not-encode catch.

Problem summary

Picking a mini-PC home server is a workload-fit decision, not a benchmark race. The Intel N-series (Alder Lake-N) dominates the low-power tier: ~6W, a QuickSync iGPU that makes media transcoding nearly free. The real forks are core count (4 vs 8), the single-channel memory ceiling (Intel validates 16GB; >16GB is board-dependent), NIC count/speed (single 1GbE vs dual 2.5GbE — a board choice, not a CPU spec), and whether QuickSync covers your codecs. N100/N150 (4-core) handle Pi-hole, Home Assistant, a handful of containers, and 1–2 Plex/Jellyfin transcodes; N305/N355 (8-core) suit heavier multi-container/multi-stream. QuickSync on this generation decodes AV1 but does not encode it (encode is H.264/HEVC).

Operator snapshotEvidence first

First proof

List the services you'll actually run and their concurrency.

Screen to open

ls -l /dev/dri

Expected signal

A concrete list (e.g. Pi-hole + HA + 6 containers + 2 transcodes).

Stop boundary

Don't assume >16GB as a spec guarantee.

Layer path

1Choosing a mini-PC home server is a workload-fit decision, not a benchmark race. The Intel N-series (Alder Lake-N) owns the low-power tier: ~6W, cheap, with a QuickSync iGPU that makes transcoding nearly free.

2The real forks are core count (4 vs 8), the single-channel 16GB validated memory ceiling, NIC count/speed (a board choice, not a CPU spec), idle power, and whether QuickSync covers the codecs you serve.

3N100/N150 (4-core) handle Pi-hole, Home Assistant, a handful of containers, and 1–2 transcodes; N305/N350/N355 (8-core) suit heavier multi-container/multi-stream. N150 is a clock-bump refresh of N100, not a tier jump.

4QuickSync on this generation decodes AV1 (plus H.264/HEVC/VP9) but does not encode AV1 — encode is H.264/HEVC; AV1 hardware encode arrived with Arc/later Xe.

Runbook

Step-by-step runbook

Start here. Do each check in order, compare it to the expected result, and stop when the evidence explains the failure or the safe stop point applies.

Write the workload list

Check: Enumerate services and peak concurrency, including transcodes.

Expected result: A concrete demand profile.

If not: Don't shop before this exists.

Pick the core tier

Check: Map the list to 4-core (N100/N150) or 8-core (N305/N355).

Expected result: A right-sized CPU tier.

If not: Many transcodes/heavy containers → 8-core.

Check codecs against QuickSync

Check: Confirm decode/encode needs vs the iGPU's support.

Expected result: Your transcodes are hardware-supported.

If not: AV1 encode needs Arc/later Xe, not Alder Lake-N.

Size RAM and NICs

Check: Confirm 16GB (or a verified 32GB board) and the NIC count/speed.

Expected result: Memory and networking match the workload.

If not: Treat >16GB as board-dependent; verify per unit.

Validate the iGPU passthrough

Check: After build, confirm /dev/dri and intel_gpu_top show transcoding.

Expected result: Hardware transcoding is actually engaged.

If not: If software fallback, fix the device passthrough or codec choice.

Re-test under peak

Check: Drive the real peak (max streams + containers) and watch CPU/GPU/RAM.

Expected result: Headroom remains at peak.

If not: If saturated, step up a tier rather than overclocking a 6W part.

Decision tree

If: Pi-hole + Home Assistant + a few containers + 1–2 transcodes

Then: Squarely a 4-core N100/N150 workload.

Action: Pick N100/N150; QuickSync handles the transcodes nearly free.

If: Heavy multi-container or many simultaneous transcodes

Then: You need the 8-core tier's non-iGPU headroom.

Action: Pick N305/N350/N355; the media engine class is the same.

If: You need lots of RAM

Then: Single-channel 16GB validated is the ceiling.

Action: Verify a specific board takes 32GB before relying on it; otherwise size down the stack.

Safe stop: Don't assume >16GB as a spec guarantee.

If: You want the box to also route/VLAN

Then: NIC count/speed matters.

Action: Choose a board with dual 2.5GbE; confirm on the board spec.

If: You expect AV1 hardware encode

Then: Alder Lake-N decodes AV1 but doesn't encode it.

Action: Plan H.264/HEVC encode, or step to Arc/later Xe if AV1 encode is required.

Safe stop: Don't buy an N-series box expecting AV1 encode.

Evidence

Evidence table

Symptom	Evidence to collect	Likely layer	Next action
Transcodes stutter under load on a 4-core box	Simultaneous transcode count + non-iGPU CPU usage	4-core ceiling for many concurrent streams	Step up to the 8-core N305/N355 tier.
Out-of-memory / heavy swapping	RAM installed vs the service set	Single-channel 16GB validated ceiling	Trim services or verify a 32GB-capable board.
AV1 'transcode' falls back to CPU	Whether the job is AV1 encode	No AV1 hardware encode on this iGPU	Encode to H.264/HEVC, or use Arc/later Xe for AV1 encode.
Can't run a second network segment	NIC count on the board	Single-NIC board	Choose a dual-2.5GbE board.
N150 transcodes slower than expected	The box's sustained power limit (PL1)	Power-capped enclosure	Pick a box with a higher sustained power limit.

Reference

Commands and settings paths

Confirm the iGPU render node exists for QuickSync

ls -l /dev/dri

Where: On the running host (Linux)

Expected: A renderD128 node is present for hardware transcoding.

Failure means: Missing /dev/dri means the iGPU isn't exposed (BIOS/driver).

Safe next step: Enable the iGPU and pass /dev/dri into the Plex/Jellyfin container.

Check memory channels/size actually in use

sudo dmidecode -t memory | grep -E 'Size|Configured Memory Speed|Locator'

Where: On the running host

Expected: Reported size matches what you installed; single channel as expected.

Failure means: Wrong size means a stick isn't seated/recognized.

Safe next step: Reseat/replace; remember Intel validates 16GB on one channel.

Verify hardware transcoding is in use

intel_gpu_top # watch the Video/Render engines during a transcode

Where: On the running host while transcoding (install intel-gpu-tools)

Expected: The Video engine shows activity = QuickSync is doing the work.

Failure means: Flat Video engine + high CPU means it fell back to software.

Safe next step: Confirm the codec is hardware-supported (AV1 decode yes, encode no).

Hardware boundary

Hardware and platform boundary

Change only when

Step from the 4-core to the 8-core tier when simultaneous transcodes or container count saturate the CPU.
Move to Arc/later Xe only if you genuinely need AV1 hardware encode.

Evidence that matters

Core count matched to peak concurrency, and QuickSync codec coverage for your transcodes.
16GB single-channel (verified 32GB only if the board supports it) and the right NIC count/speed.
A box whose sustained power limit lets the part perform.

Evidence that does not matter

Raw benchmark scores — the iGPU media engine, not CPU GHz, drives transcoding.
ECC/AV1-encode features this tier doesn't offer.

Avoid

Buying an N-series box expecting AV1 hardware encode.
Assuming >16GB RAM as a guaranteed spec.
Naming a single 'best' box — match the tier to the workload instead.

Related tool

Use the linked tool to turn this runbook into a guided check for your exact setup.

NAS setup planner

Last reviewed

2026-06-03 · Reviewed by HomeTechOps. Built from 2026-06 research verified against Intel ARK (N100), the Intel N-series comparison chart, and Intel's Media Capabilities documentation. The operator differentiators are choosing by workload tier rather than benchmarks, the single-channel-16GB ceiling, and the QuickSync 'AV1 decode yes, encode no' catch — all spec-verifiable, with no product picks or prices.

Sources/assumptions

Spec figures (cores/threads, ~6W TDP, 16GB validated max, single memory channel) are from Intel ARK / the Intel N-series comparison chart for N100, N150, N200/N250, and the 8-core i3-N305/N350/N355.
The QuickSync 'AV1 decode yes, AV1 encode no' split for Alder Lake-N is well-supported but should be verified against Intel's Media Capabilities table for the exact SKU; AV1 hardware encode arrived with Arc/later Xe, not this iGPU.
RAM beyond 16GB and NIC count/speed are board/vendor decisions, not CPU specs — stated as decision criteria to verify per unit, never as fake product picks or prices.

Source-backed checks

HomeTechOps turns official docs and conservative safety rules into a shorter runbook. These links are the source trail for the page direction.

Intel ARK: Processor N100 specificationsUsed for the N100 spec baseline: 4 cores/4 threads, 6W TDP, 16GB validated max, single memory channel, no ECC.Intel: Comparison chart for N-series processorsUsed for the N100/N150/N200/N250 and 8-core i3-N305/N350/N355 core counts, clocks, and the 16GB single-channel ceiling.Intel: Media capabilities supported by Intel hardwareUsed to verify QuickSync per-SKU codec support — AV1 decode on Alder Lake-N but no AV1 hardware encode (encode is H.264/HEVC).

Diagnostic rule

1. Check the simplest signal first.
2. Change one thing at a time.
3. Confirm the layer (app, container, network) before deep config.
4. Prove the restore; don't assume it.

FAQ

N100 or N150 — is there a real difference?

Barely. The N150 is a modest clock-bump refresh of the N100 in the same envelope: 4 cores / 4 threads, ~6W TDP, 16GB validated max, single memory channel. Don't expect a tier jump — choose on price and availability, not capability. One caveat: a mini-PC that caps the N150 at 6–8W can sustain less than a higher-power-limit part, so the box's power limit matters as much as the SKU.

Will an N100/N150 handle Plex/Jellyfin transcoding?

For 1–2 streams, yes — that's the point of QuickSync, which makes hardware transcoding nearly free on CPU. The iGPU hardware-decodes AV1, H.264, HEVC and VP9, and encodes to H.264/HEVC. The catch: this generation does not hardware-encode AV1 (that came with Arc/later Xe), so verify against Intel's Media Capabilities table for your codecs. For many simultaneous transcodes, step up to the 8-core N305/N355. Background on the hardware path: Plex/Jellyfin transcoding hardware.

How much RAM and how many NICs should I get?

Intel validates these parts at 16GB on a single memory channel — fine for Pi-hole, Home Assistant, several containers and a transcode or two. Many boards physically take a 32GB SODIMM, but that's board-dependent and beyond Intel's stated max, so treat it as 'verify the specific unit', not a guarantee. NIC count and speed (single 1GbE vs dual 2.5GbE) are board decisions too — pick by whether you need a second interface (e.g. a router/VLAN or a separate management link), and confirm it on the board's spec.