Home Server Power Consumption: N100, Ryzen, OptiPlex and Synology Compared

Home Server Power Consumption: N100 vs Ryzen vs NAS
Quick answer
How much a home server costs you in electricity per year depends on the hardware. Here are real numbers:
  • Intel N100 mini PC: 6-15W at idle, roughly $7-19 per year at average US rates
  • Ryzen mini PC (5600H/7735HS): 9-25W at idle, roughly $11-32 per year
  • Dell OptiPlex SFF/Micro: 8-30W at idle, roughly $10-38 per year
  • Synology NAS (2-4 bay, HDD): 10-35W, roughly $13-44 per year
  • Each 3.5″ hard drive adds 5-8W, around $6-10 per year

Bottom line: N100 is the most power-efficient option for 24/7 operation.

Why electricity cost gets ignored until it’s too late

Home server power consumption is something people think about at the last possible moment. Usually after the purchase. Usually after the first bill arrives.

The pattern is predictable: you grab an old office PC, install Proxmox, feel great about it. Three months later you notice the electricity bill went up by a few dollars. Then you start wondering: what if I spin up a couple more containers? Add another drive? By the end of the year the server costs as much to run as another refrigerator in the house, and that’s the optimistic case.

The problem isn’t that the server is expensive. The problem is that 24/7 adds up fast. 20W continuous is 175 kWh a year. At a typical US residential rate of around $0.17/kWh that’s about $30 for just one device. Add a switch, a router, a UPS and a couple of drives, and the number gets more interesting.

This article breaks down real numbers for each popular platform, shows you how to calculate the cost yourself, and covers how to cut consumption without losing functionality.

How to calculate it: formula and typical rates

The formula is simple:


kWh/year = (Watts / 1000) * 24 * 365
cost/year = kWh/year * rate_per_kWh

For example, a server pulling 15W on average:


(15 / 1000) * 24 * 365 = 131.4 kWh/year
131.4 * 0.17 = $22.3/year (at a typical US rate of $0.17/kWh)

Residential electricity rates vary a lot by region. Here’s a rough picture:

Region Rate, $/kWh 15W/year, $ 50W/year, $
US average 0.17 22 74
California (high) 0.30 39 131
UK average 0.28 (GBP equiv.) 37 123
Germany 0.40 53 175
Low-cost region (e.g. hydro states) 0.10 13 44

This is where it gets interesting. The difference between an N100 at 10W and an old server at 80W is roughly $100/year at US average rates. Over five years, that’s nearly $500. That money buys you a new N100 mini PC with room to spare.

How power consumption actually works: physics in 3 minutes

Before looking at the numbers, it helps to understand where they come from. Every electronic device draws power, measured in watts. Power is voltage times current. A server plugged into a wall outlet (let’s say 120V in the US) draws some current, say 0.1A. Multiply: 120 * 0.1 = 12W.

But what matters isn’t peak power, it’s the average over time. That average is what gets converted into kWh, the units that actually show up on your electricity bill.

The key thing about servers at idle: modern CPUs dynamically scale voltage and frequency down to a minimum through C-states (processor sleep states). An N100 in C6 state effectively powers down most of its circuitry and draws just a few watts. In C0 (full load) it draws tens of watts. The transition between states happens in microseconds.

This explains why the same hardware can draw 8W with C-states enabled and 20W without them. The difference isn’t the hardware, it’s the configuration.

The power supply adds its own inefficiency on top. An 80 PLUS Bronze unit runs at 82-85% efficiency. That means to deliver 10W to the system, the PSU pulls 12-13W from the wall. The difference turns into heat. A quality 80 PLUS Gold unit (87-90% efficiency) pulls 11-12W for the same load. At small power levels the dollar difference is minor, but it explains why enterprise servers ship with quality PSUs.

Intel N100: the efficiency champion

The N100 is an Intel Alder Lake-N processor: 4 E-cores, 6W TDP, released in 2023. It’s what ships in most budget mini PCs right now: Beelink EQ12, GMKTec G3, various CWWK and Minisforum boxes.

Real measurements (watts at the wall):

Mode Consumption kWh/year Cost/year (US avg.)
Idle, headless 6-8W 53-70 $9-12
Idle, mixed workload 9-12W 79-105 $13-18
Moderate load (Docker, VMs) 12-18W 105-158 $18-27
Peak load 25-30W 219-263 $37-45

These numbers are real, measured by the community with Kill-A-Watt meters and Tasmota smart plugs.

Important detail: at peak the N100 hits 25-27W, but that’s short lived. In a typical homelab with standard workloads (Pi-hole, Home Assistant, Wireguard, a handful of containers), average draw stays around 10-14W.

Proxmox on N100 is a popular combo. Here’s what idle looks like without any VMs running:


# Check current CPU package power (Linux)
cat /sys/class/powercap/intel-rapl:0/energy_uj
# or via turbostat (requires root)
turbostat --quiet --show PkgWatt,CoreWatt,RAMWatt sleep 5

One small trap: C-states. If C-states are disabled in BIOS, or a USB device is passed through to a VM, idle consumption can jump from 9W to 20W. This isn’t a hardware bug, it’s how the processor architecture works. Check your BIOS settings: Speed Step and C-states should both be enabled.

Ryzen mini PC: more power, more draw

Ryzen servers in a homelab usually mean Beelink SER5/SER6 or similar boxes built on 5600H, 5700U, 6800H or 7735HS. These are 35-45W TDP processors, but real idle consumption is much lower.

Processor Idle, W Load, W Cost/year idle (US avg.)
Ryzen 5 5560U / 5600H 9-12W 35-45W $13-18
Ryzen 7 5700U 10-15W 35-50W $15-22
Ryzen 7 6800H 9-12W 55-74W $13-18
Ryzen 7 7735HS / 8845HS 18-25W 60-90W $27-37

Here’s a detail people often miss. A Ryzen 7 7735HS running Windows 10 LTSC draws 30-40W at idle, which is abnormal, caused by system interrupts. The same processor under Ubuntu 24.04 idles at 8-9W. A 4x difference just from the choice of OS.

Older Ryzen 5000-series chips (5600H, 5700U) behave close to the N100 at idle. The difference shows up under load: Ryzen wins by a wide margin. If you need virtualization with several heavy VMs or video transcoding, Ryzen pays for itself.

For typical homelab workloads, Docker, VPN, monitoring, DNS, the N100 does the job cheaper.

Dell OptiPlex: a used server with surprises

OptiPlex is a popular choice for homelabs on the used market. SFF or Micro chassis, i5/i7, 16-32GB RAM for next to nothing. The catch: consumption depends heavily on CPU generation, and the spread is wide.

Model Processor Idle, W Load, W Cost/year (US avg.)
OptiPlex 3070 Micro i5-9500T 8-12W 35-55W $12-18
OptiPlex 5060/7060 SFF i5-8500 / i7-8700 15-30W 65-100W $22-45
OptiPlex 7000 Micro i7-12700 (Alder Lake) 5.5-6.5W 85-180W $8-10
OptiPlex 3080 Tower G5900 7-14W 45-80W $10-21

Key takeaway: OptiPlex Micro units with Alder Lake processors (12th gen, 2022+) are roughly as efficient as the N100. Older SFF models on 8th and 9th gen chips can pull 20-30W at idle, which is noticeable.

T-series processors (i5-9500T instead of i5-9500) use 30-40% less power on average for the same performance in typical server workloads. When shopping for a used OptiPlex, look for the T in the CPU model name.


# Check power consumption in Linux without an external meter
# Via PowerTop
apt install powertop -y
powertop --auto-tune
powertop --time=30

# Via RAPL (Intel)
cat /sys/class/powercap/intel-rapl:0/constraint_0_power_limit_uw

Synology NAS: quiet, but not free

Synology isn’t a server in the classic sense. It’s a purpose-built NAS appliance where low power draw is a design priority. Still, there are nuances.

Model Processor Drives Idle, W Load, W Year (US avg.)
DS224+ Intel J4125 2x HDD 6-10W 16-20W $9-15
DS720+ Intel J4125 2x HDD 6-8W 16W $9-12
DS723+ AMD R1600 2x HDD 8-12W 21W $12-18
DS923+ AMD R1600 4x HDD 11-15W 35W $16-22
DS1522+ AMD R1600 5x HDD 20-28W 55W $30-42

Two things affect Synology power draw the most.

First, drive hibernation. Enable disk sleep through DSM (Control Panel → Hardware → HDD Hibernation), and idle consumption drops from 11W to 3-4W. The tradeoff is a 3-5 second delay on the first access after sleep.

Second, CPU generation. The DS923+ on AMD R1600 draws nearly twice as much at idle as the DS720+ on Intel J4125. This is a known issue Synology hasn’t fixed. If electricity cost matters to you, go with the Intel-based models.

Typical homelab architecture: power breakdown

Блок-схема: Wall outlet, UPS / Surge protector, Router / Switch - 5-15W, Main server, NAS / Storage, N100: 6-15W
Typical homelab architecture: power breakdown

Supporting hardware: what people forget to count

Counting only the main server is a mistake. A typical homelab runs plenty of other gear that never sleeps either.

Component Consumption Year (US avg., $0.17/kWh)
Wi-Fi router 5-15W $7-22
8-port managed switch 5-10W $7-15
3.5″ HDD (7200 RPM, active) 6-8W $9-12
3.5″ HDD (5400 RPM NAS, active) 4-6W $6-9
M.2 NVMe SSD 0.5-3W $0.7-4
600VA UPS on standby 8-15W $12-22
Raspberry Pi 4 3-6W $4-9

Take a real scenario: N100 + Synology DS923+ with 4 HDDs + router + switch.

Without optimization: 10 + 30 + 10 + 8 = 58W. A year at US average rates: about $86. With Synology drive hibernation and proper C-states on the N100: 10 + 4 + 10 + 8 = 32W. A year: about $48. That’s $38/year saved just from configuration.

HDD vs SSD: impact on total consumption

Common myth: SSDs are more power-efficient than HDDs. The truth is more nuanced.

An NVMe SSD under active load draws 2-5W. SATA SSD: 1-3W. A 3.5″ HDD under active load: 6-8W, in standby: 0.5-1W (if hibernation is enabled).

For server use, idle behavior matters most: a server spends most of its time not reading or writing data. With hibernation enabled, an HDD at idle is actually more efficient than an SSD, which always draws some baseline power.

Practical conclusion: use SSD or NVMe for the OS and active data. Use HDDs with hibernation enabled for archives and media libraries. That combo beats an all-NVMe setup on total consumption.

How to measure your own server’s power draw

Without measurement, you’re just guessing. You need a meter on the outlet.

Two options:

Budget: a smart plug with power monitoring. Think Gosund SP111, Athom v2 flashed with Tasmota, or Nous A1T. Costs $10-20. Shows current watts and accumulated kWh. Integrates with Home Assistant.

Precise: a dedicated power meter like a Kill-A-Watt or P3 P4400. Higher accuracy, but no remote monitoring.

What to measure:


# Scenarios for taking readings:
# 1. Idle - system up, no active tasks, 10 minutes
# 2. Load - run a stress test
stress-ng --cpu 4 --timeout 60s

# 3. Typical workload - run real services
# and take an average over an hour

# Via smart plug with MQTT into Home Assistant:
# sensor.smart_plug_power - current watts
# sensor.smart_plug_energy - accumulated kWh

Record three values: minimum at deep idle, average under typical load, peak under max load. Use the average for your yearly cost calculation.

Optimizing consumption: what actually works

Not every tweak is equally useful. Here’s what actually moves the needle.

C-states and Speed Step in BIOS: enable them, no question. This lets the processor drop voltage and frequency at idle. Without it an N100 holds 15-18W instead of 8W. Check your BIOS: CPU C States: Enabled, Intel SpeedStep: Enabled.

Headless mode: minus 2-5W. A server with no monitor attached draws less. If a display is only needed for setup, plug it in temporarily.

Drive hibernation: minus 5-20W. On Synology DSM: Control Panel → Hardware & Power → HDD Hibernation, enable it. On Linux:


# Set spindown delay (in seconds / 5)
# Value 120 = 10 minutes
hdparm -S 120 /dev/sda

# Check current drive state
hdparm -C /dev/sda
# "drive state is: standby" means the disk is asleep

Disable unused interfaces. Bluetooth, a Wi-Fi adapter on a mini PC that’s wired anyway: disable them in BIOS. 1-3W saved sounds trivial, but over a year that’s $2-5 for nothing in return.

PowerTop on Linux: auto-tune.


apt install powertop
# Analyze consumption
powertop --time=30
# Apply optimizations automatically
powertop --auto-tune

# To apply auto-tune on every boot:
cat > /etc/systemd/system/powertop.service << 'EOF'
[Unit]
Description=PowerTop auto-tune
After=multi-user.target

[Service]
Type=oneshot
ExecStart=/usr/sbin/powertop --auto-tune

[Install]
WantedBy=multi-user.target
EOF

systemctl enable powertop.service

OS choice matters. A base Proxmox install draws slightly more than a clean headless Debian, a difference of 2-5W. Windows Server with default settings on the same hardware can pull 3-4x more than headless Linux. If the goal is efficiency: headless Linux, no GUI.

Platform comparison: summary table

Platform Typical idle, W Year, US avg., $ Peak, W Best use case
N100 mini PC 8-12 $12-18 25-30 Docker, VPN, Home Assistant, DNS, monitoring
Ryzen 5 5600H mini PC 9-15 $13-22 40-55 Light virtualization, NAS + Docker
Ryzen 7 7735HS mini PC 18-25 $27-37 70-90 Heavy virtualization, LLM workloads, transcoding
OptiPlex Micro (Alder Lake) 6-10 $9-15 85-150 Proxmox with several VMs, moderate load
OptiPlex SFF (8-9th gen) 15-30 $22-45 65-120 Only if you already own one
Synology DS224+ (2 HDD) 6-10 $9-15 16-20 NAS, backups, file storage
Synology DS923+ (4 HDD) 11-15 $16-22 35 Redundant NAS, media server

A single 3.5″ hard drive running constantly adds $9-12/year to any of these numbers. Five drives add $45-60/year on top of the base server draw. At that point the logic starts to shift: maybe buy fewer, larger drives instead.

Drive selection: impact on multi-year costs

Drives are the second biggest factor after the processor, especially once you have several of them.

Typical scenario: someone buys a Synology DS923+ and loads it with 4x 8TB drives. Four 3.5″ HDDs running actively add 24-32W on top of the NAS’s base 11W. Total 35-43W. With hibernation: 11-15W. The difference, 20-28W, is $30-42/year.

Takeaway: drive count matters less than duty cycle. Most home NAS units sit idle most of the time, not reading or writing. Enable hibernation and get a nearly free win.

What to pick for NAS drives in 2025:

NAS-specific drives like WD Red Plus or Seagate IronWolf draw 4-6W in operation versus 8-10W for standard desktop HDDs. Over three years across 4 drives, that’s $15-20 saved. Modest, but real.

SSDs in a NAS make sense for the OS and active data, but not for archival storage. A 4TB SSD costs 5-8x more than an equivalent HDD, and the electricity savings from hibernation already covers most of the gap on the HDD side.

5-year calculation: N100 vs an old server

Let’s take two realistic scenarios and calculate total cost of ownership over 5 years.

Parameter N100 mini PC Dell PowerEdge R720
Purchase price $80-130 $60-170 (used)
Idle consumption 10-15W 130-170W
Electricity/year (US avg., 10W/130W) $15 $193
Electricity over 5 years $75 $965
Difference in electricity alone over 5 years -$890 in favor of N100
Total cost over 5 years $155-205 $1025-1135

That’s why a “cheap” $60 server off eBay turns out to be the most expensive option. The upfront price is less than 15% of the total 5-year cost. The rest is electricity, maintenance and noise in the room.

Though to be fair, the R720 delivers things an N100 physically can’t: dual Xeon processors, 16+ cores, 256GB RAM, PCIe slots for RAID controllers, hot-swap drives. If you genuinely need those capabilities, that’s one thing. If you don’t, it’s an expensive toy.

Configuring automatic power reduction

A few commands you can apply right now on any Linux server.


# 1. Enable powersave governor on all cores
for cpu in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do
    echo powersave > $cpu
done

# Verify
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor | sort -u

# 2. Enable USB autosuspend
echo 'auto' > /sys/bus/usb/devices/usb1/power/control
echo 'auto' > /sys/bus/usb/devices/usb2/power/control

# Or via powertop:
powertop --auto-tune

# 3. Set drive spindown delay (15 minutes = 180)
hdparm -S 180 /dev/sda
hdparm -S 180 /dev/sdb

# Make it persistent via /etc/hdparm.conf:
echo '/dev/sda {
    spindown_time = 180
}
/dev/sdb {
    spindown_time = 180
}' >> /etc/hdparm.conf

# 4. Disable HDMI output if not needed (saves 2-5W)
tvservice -o  # Raspberry Pi
# For x86:
echo 1 > /sys/class/backlight/*/bl_power  2>/dev/null || true

# Quick script to check current power-saving state
#!/bin/bash
echo "=== CPU Governor ==="
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

echo ""
echo "=== C-states usage ==="
grep -h "usage" /sys/devices/system/cpu/cpu0/cpuidle/state*/usage 2>/dev/null | head -6

echo ""
echo "=== USB autosuspend ==="
for dev in /sys/bus/usb/devices/*/power/control; do
    echo "$dev: $(cat $dev)"
done

echo ""
echo "=== HDD spindown status ==="
for disk in /dev/sd*; do
    if [ -b "$disk" ]; then
        state=$(hdparm -C "$disk" 2>/dev/null | grep "drive state")
        echo "$disk: $state"
    fi
done

Run this script after applying settings and after every reboot to confirm the settings stuck.

When upgrading hardware actually pays for itself

A simple payback calculation. Take the difference in power draw, multiply by rate and by a year, and you get the annual savings. Divide the cost of the new hardware by the annual savings, and you get the payback period.

Example: an old i7-4770 PC draws 45W at idle. An N100 mini PC draws 10W. Difference: 35W.


# Payback calculation
watt_diff=35
rate=0.17
new_hw_cost=130

annual_savings=$(python3 -c "print(round($watt_diff / 1000 * 24 * 365 * $rate, 2))")
payback=$(python3 -c "print(round($new_hw_cost / ($watt_diff / 1000 * 24 * 365 * $rate), 1))")

echo "Annual savings: \$${annual_savings}"
echo "Payback period: ${payback} years"

Result: annual savings of about $52, payback period of about 2.5 years at US average rates. An N100 costing $130 pays for itself in roughly 2-3 years just from the electricity difference. And that’s before counting that the N100 runs quieter, doesn’t heat up the room and doesn’t need a new fan every three years.

In high-rate regions like California or Germany, the payback happens faster. At low rates it stretches out further. Run the numbers for your own region.

How much a homelab costs per year: real scenarios

Now let’s put it all together. Four realistic scenarios, from minimal to a full homelab. All figures use a US average rate of $0.17/kWh.

Scenario 1: Minimal homelab
N100 mini PC (10W) + MikroTik hEX S router (8W) + TP-Link 8-port switch (5W).
Total: 23W.
Year: about $34.
What runs on it: Pi-hole, WireGuard VPN, Uptime Kuma, a handful of Docker services.

Scenario 2: Mid-size homelab with NAS
OptiPlex 3070 Micro (12W) + Synology DS723+ with 2 HDDs, hibernation enabled (9W) + MikroTik hEX S (8W) + switch (5W).
Total: 34W.
Year: about $51.
What runs on it: Proxmox with 4-6 LXC containers, Jellyfin, Nextcloud, NAS with backups.

Scenario 3: Full homelab
Ryzen 5 mini PC (15W) + Synology DS923+ with 4 HDDs, hibernation enabled (13W) + MikroTik CCR (12W) + 24-port PoE switch (18W) + UPS on standby (12W).
Total: 70W.
Year: about $104.
What runs on it: Proxmox with 10+ VMs, k3s, a full RAID-backed NAS, PoE-powered Wi-Fi access points.

Scenario 4: “Why do I have a data center in my house”
Dell PowerEdge R720 (150W) + Synology DS1522+ with 5 HDDs (25W) + switch (15W) + UPS (15W).
Total: 205W.
Year: about $305.
What runs on it: roughly the same as scenario 3, but with bragging rights in the chat.

The gap between scenario 1 and scenario 4 is about $270/year. Over ten years, that difference pays for two or three new N100 boxes with memory and storage included.

Though to be fair, the R720 delivers things an N100 physically can’t: dual Xeon processors, 16+ cores, 256GB RAM, PCIe slots for RAID controllers, hot-swap drives. If you genuinely need those capabilities, that’s one thing. If you don’t, it’s an expensive toy that also heats your apartment.

Proxmox: power consumption specifics under virtualization

Proxmox is a hypervisor, and it has its own quirks that affect power draw, quirks you won’t find documented anywhere.

A clean Proxmox install with no VMs on N100 draws 9-12W. Spin up one lightweight VM (Ubuntu server, 1 vCPU, 512MB RAM) and you add 1-2W. Run 10 LXC containers and total draw still stays around 14-16W. Virtualization itself barely affects consumption. What matters is what those VMs actually do.

Three things do genuinely raise consumption under Proxmox, though:

First: live migration and load balancing. If you’re running a cluster and a VM is being moved, you get a short spike in network and CPU load.

Second: ZFS with ARC enabled. ZFS tries to use all available RAM as cache. More RAM in active use means higher draw. On systems with 16GB+ this is noticeable. Limit ARC:


# Check current ARC size
cat /proc/spl/kstat/zfs/arcstats | grep -E "^size|^c_max"

# Limit ARC to 4GB
echo "options zfs zfs_arc_max=4294967296" > /etc/modprobe.d/zfs.conf
update-initramfs -u

# Verify after reboot
cat /proc/spl/kstat/zfs/arcstats | grep "^c_max"

Third: IOMMU and passthrough. If you’re passing a GPU or USB controller into a VM, IOMMU is active and keeps extra load on the PCIe bus. That’s 2-5W constantly, even if the VM with the GPU isn’t currently using it.


# Check if IOMMU is active
dmesg | grep -e DMAR -e IOMMU | head -5
# If you don't need passthrough, better to disable it in BIOS

# Check C-states usage in Proxmox
cpupower monitor -m "Idle_Stats"
# You should see C6/C7/C8 states being used

Network equipment: hidden costs

Most people only count the server. They forget about everything else sitting in the rack or closet.

Equipment Consumption Year (US avg.) Note
TP-Link TL-SG108E (8-port managed) 4-6W $6-9 Passively cooled, silent
MikroTik hEX S 7-9W $10-13 Full-featured router
MikroTik CSS326-24G-2S+ 12-18W $18-27 24-port, with SFP+
Ubiquiti UniFi USW-24-PoE 13-25W $19-37 Without PoE load
Raspberry Pi 4 (4GB) 3-6W $4-9 Under light load
APC 600VA UPS 10-18W $15-27 Losses from charging and inverter

A UPS is its own story. It always draws power, even when the grid is fine. Cheap UPS units run at 85-90% efficiency, meaning 10-15% of electricity just turns into heat. A good online UPS at 95%+ efficiency costs more, but pays off under heavier load.

For a minimal homelab without demanding workloads, a UPS is often overkill. Graceful shutdown is often better handled by a small Pi-UPS battery, or by enabling auto-power-on in BIOS (AC Recovery: Power On).

What actually affects your yearly bill

Let’s bring everything together, from most important to least.

Platform choice matters most. The gap between an N100 and an old rack server is 100-150W at idle. That’s 90% of all the potential savings. Everything else, tuning, optimizations, drive selection, accounts for 20-40% of what’s left.

C-states and Speed Step. Enabling C-states drops N100 idle draw from 18-20W to 8-10W. That 10W difference is roughly $15/year at US average rates. A free BIOS setting.

OS choice. Headless Linux versus Windows: a 10-30W difference on the same hardware. On a Ryzen 7735HS this is dramatic: Ubuntu at 8-9W, Windows 10 LTSC at 30-40W. For a server you don’t need to touch with a monitor and keyboard, go Linux only.

Drive hibernation. Each active 3.5″ HDD draws 6-8W. Asleep: 0.5-1W. With 4 drives that’s a 20-28W difference, especially critical for NAS boxes where drives sit idle most of the time.

Disabling monitors and displays. Headless mode saves 2-5W. If a monitor is physically connected but unused, it’s still initialized and keeps draw higher.

Powertop auto-tune. On a typical Linux system this saves 2-8W by optimizing USB, PCI and other subsystems. Free, one command to apply.

RAM choice. DDR5 draws about 20% less than DDR4 at the same capacity. But the absolute difference is 1-2W per module. Not a reason to pick a platform on its own, but a nice bonus on newer systems.

Common mistakes when estimating consumption

After years of running homelabs, you start seeing the same mistakes over and over when people estimate electricity cost.

Calculating from CPU TDP. TDP is the thermal envelope under maximum load. An N100 rated at 6W TDP actually draws 6-15W at idle once you count the whole system, including PSU and memory. There’s no direct relationship.

Ignoring the power supply. PSU efficiency runs 80-90% for good units, 70-75% for cheap ones. That means if a system draws 50W internally, it pulls 55-65W from the wall. Always measure at the outlet, not from software monitoring readouts.

Only measuring idle. A server isn’t always idle. If backups run hourly, antivirus scans run daily, and Jellyfin is streaming continuously, average consumption is higher than the minimum. For an accurate estimate you need a full-day average.

Forgetting boot-up draw. Power spikes during startup. POST, OS boot, service initialization, all of that is a short-term peak. If a server reboots frequently due to instability, that’s extra spikes every few hours.

Using outdated rates. Electricity rates change over time and vary by utility and season. Check your actual current rate before running any calculation.


# Quick cost calculator: run with your own parameters
#!/bin/bash
AVG_WATTS=${1:-15}
RATE=${2:-0.17}

python3 << EOF
avg_w = $AVG_WATTS
rate = $RATE
daily_kwh = avg_w / 1000 * 24
monthly_kwh = daily_kwh * 30
yearly_kwh = daily_kwh * 365

print(f"Average draw: {avg_w}W, rate: \${rate}/kWh")
print(f"Per day:   {daily_kwh:.3f} kWh = \${daily_kwh * rate:.2f}")
print(f"Per month: {monthly_kwh:.1f} kWh = \${monthly_kwh * rate:.2f}")
print(f"Per year:  {yearly_kwh:.0f} kWh = \${yearly_kwh * rate:.2f}")
print(f"Over 5 years: {yearly_kwh * 5:.0f} kWh = \${yearly_kwh * 5 * rate:.2f}")
EOF

Monitoring: setting up power tracking in 30 minutes

Doing the math by hand every month isn’t sustainable. You need automation.

If you already run Home Assistant, a power-monitoring smart plug connects in about 10 minutes. Here’s how to add a simple cost sensor:


# configuration.yaml in Home Assistant
sensor:
  - platform: template
    sensors:
      server_daily_cost:
        friendly_name: "Server - daily cost"
        unit_of_measurement: "$"
        value_template: >
          {{ (states('sensor.smart_plug_power') | float * 24 / 1000 * 0.17) | round(2) }}
      server_monthly_cost:
        friendly_name: "Server - monthly cost"
        unit_of_measurement: "$"
        value_template: >
          {{ (states('sensor.smart_plug_power') | float * 24 * 30 / 1000 * 0.17) | round(2) }}

Replace 0.17 with your actual rate. Replace sensor.smart_plug_power with your sensor’s name.

If you don’t have Home Assistant yet, you can use Grafana + Prometheus + node_exporter instead. More setup work, but more flexible for monitoring an entire homelab.

Monitoring without Home Assistant: Netdata

If Home Assistant isn’t in the picture, install Netdata. It’s a monitoring agent that sets up in 5 minutes and gives you a solid web dashboard with CPU, memory, network and temperature metrics.


# Install Netdata (works on Debian/Ubuntu/Proxmox)
wget -O /tmp/netdata-kickstart.sh https://get.netdata.cloud/kickstart.sh
sh /tmp/netdata-kickstart.sh --dont-wait

# Access: http://IP:19999
# CPU power data can be viewed via the RAPL plugin
# (requires the intel-rapl driver)
modprobe intel_rapl_common

Netdata doesn’t measure the outlet, it shows internal system metrics. For real power consumption you still need a smart plug. But Netdata is good at showing correlation: when a backup started, draw went up, when a cron job finished, it dropped back down.

Automatic shutdown overnight

If your homelab doesn’t need to run 24/7, you can schedule automatic shutdowns overnight. On Synology this is built in: Control Panel → Hardware → Power Schedule. Set the active window, and the device turns itself on and off on that schedule.

On a Linux server:


# Shut down at 2:00, wake at 8:00
# Step 1: configure RTC wakeup (BIOS RTC alarm)
# Check support
cat /sys/class/rtc/rtc0/wakealarm

# Set wake time (unix timestamp)
WAKE_TIME=$(date -d "tomorrow 08:00" +%s)
echo 0 > /sys/class/rtc/rtc0/wakealarm
echo $WAKE_TIME > /sys/class/rtc/rtc0/wakealarm

# Confirm
cat /sys/class/rtc/rtc0/wakealarm

# Step 2: add a shutdown cron job
# Shut down every night at 2:00
echo "0 2 * * * root echo 0 > /sys/class/rtc/rtc0/wakealarm && echo \$(date -d 'tomorrow 08:00' +%s) > /sys/class/rtc/rtc0/wakealarm && /sbin/shutdown -h now" > /etc/cron.d/server-schedule

The server shuts down at 2:00, the BIOS RTC alarm wakes it back up at 8:00 automatically. If “Power On By RTC Alarm” is enabled in BIOS, this works without any extra scripting.

Running 16 hours instead of 24 saves 33% of electricity. An N100 in this mode would cost about $8-12/year instead of $12-18. Small amount, but worth it for some use cases: a CI/CD runner, a media server nobody watches overnight.

How ambient temperature affects power draw

A factor that rarely gets mentioned: room temperature affects consumption, specifically through fan behavior and thermal throttling.

The N100 is passively cooled or has a small fan. At 20°C room temperature it runs quiet with minimal draw. At 30°C in summer the fan spins up, the CPU may start throttling, and consumption can rise 10-15%.

Synology with HDDs is especially sensitive. At higher temperatures DSM increases drive cooling fan speed. The fan motor itself adds 3-5W. Drives also draw slightly more when running hot.

Practical consequence: a server sitting in a poorly ventilated cabinet draws more than the same server on an open shelf. Sometimes an open layout, or just a small gap in the cabinet door, delivers real fan-related savings.

Troubleshooting: why consumption is higher than expected

What to do if your server draws more than expected

You got a power meter, measured consumption, and the number is unpleasantly high. What now?

First step: isolate the problem. Measure with the drives pulled out. If draw drops by 20-30W, the drives are spinning constantly and hibernation isn’t working. Check the settings.

Measure power in BIOS before the OS loads. If it’s already at 15W when the spec sheet says 6-8W, the problem is in the hardware or BIOS settings. Check C-states, Speed Step, and whether power-saving options are enabled.

If consumption is normal in BIOS but high once the OS loads, the problem is software. On Linux, run powertop and look at “Wakeups/sec”. If the value is above 100-200, something is waking the CPU every second and preventing it from reaching deep sleep states.


# Find processes with high wakeup counts
powertop --time=30 2>/dev/null | head -50

# Or via perf
perf stat -e power:cpu_idle sleep 10

# Find processes blocking C-states
grep -r "pm_qos" /proc/*/net/ 2>/dev/null | head
cat /proc/sys/kernel/timer_migration

# List processes by CPU usage (even small % keeps the CPU in C0)
top -b -n 1 -d 1 | head -30

Typical culprits: systemd-journald writing too frequently, ntpd without a proper polling interval, a Docker container with constant network polling, a metrics collector scraping every second.

Each has its own fix: increase the poll interval, buffer logs, switch metrics collection to push instead of pull. The details show up in logs and in powertop’s output.

Symptom Cause Fix
N100 pulls 18-20W at idle C-states disabled in BIOS BIOS: CPU C States: Enabled; Speed Step: Enabled
Ryzen under Windows pulls 30-40W at idle System interrupts from ntoskrnl.exe Switch to Linux, or track down the problem driver with LatencyMon
Proxmox pulls 20+W instead of 10W USB passthrough to a VM is blocking C-states Check: cat /sys/module/usbcore/parameters/autosuspend, should be 2
Synology pulls 35W at idle Drive hibernation disabled DSM: Control Panel → Hardware → HDD Hibernation: enable it
OptiPlex won’t enter power-saving mode Power profile not configured in Windows/Linux Linux: cpupower frequency-set -g powersave
Consumption spikes unpredictably Background processes (backups, indexing) Move heavy jobs to a night window, check cron

# Diagnosing C-states on Linux
grep . /sys/devices/system/cpu/cpu0/cpuidle/state*/name
grep . /sys/devices/system/cpu/cpu0/cpuidle/state*/usage

# Check the current CPU governor
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

# Switch to powersave
cpupower frequency-set -g powersave

# Check USB autosuspend
cat /sys/module/usbcore/parameters/autosuspend
# Should be 2, if it's 0, C-states may not work properly
echo 2 > /sys/module/usbcore/parameters/autosuspend

FAQ

How much power does an N100 mini PC use per day?

In a typical homelab running Docker and a few services: 0.15-0.25 kWh per day. At idle with no workload: 0.14-0.18 kWh. At US average rates ($0.17/kWh) that’s $0.025-0.04 per day, or around $9-16 per year. One of the cheapest options for round-the-clock operation.

How do I find out how much electricity my server uses?

You need a measurement device, a power meter or a smart plug with power monitoring. Without direct measurement, the numbers are just estimates. A Gosund or Athom smart plug flashed with Tasmota costs $10-20 and shows current watts plus accumulated kWh. It’s the best investment for understanding actual consumption.

Why is N100 more power-efficient than Ryzen for a home server?

The N100 draws 6-15W at idle, a Ryzen 5600H draws 9-15W, a Ryzen 7735HS draws 18-25W. For typical server tasks (DNS, VPN, Docker, monitoring) the N100 has more than enough performance. Ryzen pays off if you need heavy virtualization or video transcoding without hardware acceleration.

Is it worth running an old Dell PowerEdge for a homelab?

Not from a power standpoint. An old rack server (R710, R720) draws 120-170W at idle, that’s $180-255/year in electricity alone at US average rates. An N100 mini PC costing $80-130 pays for itself just on the electricity difference within 1-2 years. After that it’s pure savings. Rack servers make sense only if you genuinely need specific hardware: many PCIe slots, large amounts of ECC RAM, redundant PSUs.

Does RAM capacity affect server power consumption?

Yes, but modestly. Each DDR4 module in use draws about 2-3W. The difference between 8GB (1 module) and 32GB (2 modules) is around 3-6W. Over a year that’s $4-9. Not critical, but it counts.

Which is more efficient: a Synology NAS or a mini PC used as a NAS?

Roughly equal, with proper configuration. A Synology DS224+ with 2 HDDs at idle: 6-10W. An N100 with a SATA card and 2 HDDs: 10-16W. Synology saves 3-6W, which is $4-9/year. The choice should come down to functionality (DSM versus your own stack), not electricity.

Can you reduce Proxmox power draw without new hardware?

Yes, and quite a bit. Enable C-states and Speed Step in BIOS, that alone saves 5-10W on an N100. Run powertop –auto-tune for another 2-5W. Switch the CPU governor to powersave with cpupower frequency-set -g powersave. Disable monitors in headless mode. Combined, that’s 10-20W off the baseline, for free.

How much does a home server cost per month?

Depends on the platform and your rate. An N100 mini PC averaging 12W draws about $1.50/month at US average rates. A Synology DS923+ with 4 HDDs and hibernation at 13W: about $1.65/month. Together with a router and switch, a full setup runs around $3-6/month. An old rack server at 150W: $18-22/month on its own.

How do I enable drive hibernation on Synology DSM?

Open DSM, go to Control Panel → Hardware and Power → HDD Hibernation. Check “Enable HDD hibernation” and set the idle timeout, usually 10-20 minutes. In Advanced Power Settings you can also enable “Enable WOL on LAN”. Once active, drives sleep after the specified period of no access. The first request after sleep adds a 2-5 second delay, which is normal.

Is it worth waiting for more efficient hardware

Looking at the next couple of years: Intel N200 and N305 are already available and offer more performance at similar or slightly higher power draw. Intel’s Arrow Lake-N (the next generation after Alder Lake-N) is expected to bring better per-core efficiency.

AMD continues its Ryzen Embedded line in the low-power segment. New generations promise lower idle draw through improved power management, especially under Linux.

If you already have an N100, there’s nothing to upgrade to yet. The gain from the next generation will be 2-4W at idle. Over 5 years that’s $8-16 saved. Not worth it.

If you’re buying now, the N100 remains the best balance of price, performance and power draw for a standard homelab in 2025.

Bottom line: what to pick and what to budget

If you’re just starting out with a homelab: get an N100 mini PC. Pay $80-130 upfront, spend $12-18/year on electricity, get a platform that handles everything a typical homelab needs.

If you need virtualization with 5+ VMs: an OptiPlex Micro on Alder Lake (12th gen) or a Ryzen 5 mini PC. Idle draw similar to the N100, under load noticeably higher, but performance is in a different league.

If you need terabytes of storage: Synology is a purpose-built tool with solid DSM software, good support, and minimal consumption when hibernation is configured properly. Don’t replace it with a self-built NAS just to save a few dollars.

If you’re already running an old office tower or rack server: run the numbers for real. Measure it with a power meter. Compare against the price of a decent N100. Odds are the replacement pays for itself within a year or so.

Electricity is an operating cost. It’s invisible at purchase time, but it compounds year after year. The difference between the right and wrong hardware choice is $50-100/year, every single year. Do the math once, save many times over.

Electricity is an operating cost. It’s invisible at purchase time, but it keeps adding up year after year without stopping. The difference between the right and wrong hardware choice is $50-100 annually, year after year.

Do the math once, save many times over. Grab a power meter, measure your current server’s actual draw. Compare it against the tables above. If the difference is more than 30-40W at idle, it’s already worth thinking about optimization or replacement.

Three things you can do today: enable drive hibernation in Synology DSM, check C-states in your server’s BIOS, run powertop –auto-tune on your Linux host. All free, all give a measurable result.

If you measured your own hardware and got surprising numbers, drop them in the comments. Let’s figure it out.

Summary: power consumption
The most efficient options for a 24/7 homelab server:
1. N100 mini PC: 6-15W, $9-16/year at US average rates
2. OptiPlex Micro on Alder Lake: 6-10W, $9-15/year
3. Synology DS224+ with hibernation: 4-8W, $6-12/year

What will reliably inflate your bill: an old rack server, HDDs without hibernation, disabled C-states in BIOS, Windows on a home server.

Tool for measuring: a smart plug with power monitoring, $10-20, pays for itself within a month of actually paying attention to your usage.

Оставайтесь на связи

Рецепты от IT-боли. Без воды, без рекламы, без маркетинговой шелухи.

Подписаться на IT-Аптеку →
Поделитесь:

4 thoughts on “Home Server Power Consumption: N100, Ryzen, OptiPlex and Synology Compared”

  1. Павел Т.

    У меня три ноды в кластере Proxmox: N100 для лёгких LXC, Ryzen 7735HS под VMs и старый OptiPlex как бэкап. Никто не пишет, что при объединении в кластер с Ceph или общим хранилищем сетевой трафик и синхронизация выжирают больше, чем сами железки. У меня 2.5G свитч начал греться и жрать 15W вместо 8W после включения репликации. А как у тебя меняется профиль энергопотребления, когда ноды простаивают, но должны поддерживать кворум и репликацию? Пробовал ли замерять разницу между локальным ZFS и сетевым Ceph на этих платформах?

    1. Кластер с кворумом и репликацией это отдельная история по питанию. У меня два N100 и один OptiPlex Micro, все на локальных ZFS, без Ceph. Причина именно в том, что ты описал: синхронизация и сетевой стек держат CPU в C1-C2 вместо C6, и idle на N100 уходит с 8W до 14-16W на каждой ноде. Плюс 2.5G свитч под репликацией греется сильнее, я замерял Tasmota: 12-14W против 7W в простое. Кворум сам по себе не жрёт, но corosync heartbeat каждую секунду не даёт процу уйти в глубокий сон. Проверить можно так: powertop –auto-tune и смотреть, сколько времени CPU проводит в C6. У меня после тюнинга N100 держит C6 больше 80% времени, и idle падает до 9W даже с включённым кворумом. По Ceph vs ZFS: на 2.5G разница будет в пользу локального ZFS по ваттам, потому что каждый write идёт по сети и будит все ноды. На 10G картина другая, но там уже сетевое оборудование съест больше, чем сэкономишь. Если нужна отказоустойчивость на трёх нодах, я бы смотрел в сторону Proxmox с ZFS replication раз в 15 минут, а не Ceph. Репликация запускается по расписанию, и ноды успевают поспать между циклами.

  2. Дмитрий К.

    У меня два диска в Synology не засыпают: один 2.5″ SSD под систему, второй 3.5″ HDD под файлы. DSM показывает, что спит только HDD, а SSD крутится всегда, и вместе с ним просыпается всё. Живу в Германии, тариф 0.38 евро, так что лишние 4-5 Вт в год это уже ощутимо. Пробовал вытаскивать SSD вообще, но система без него не грузится. Как заставить систему спать, не убирая загрузочный диск?

    1. SSD под DSM действительно не уснёт: система пишет логи каждые пару минут, и это выдёргивает HDD из сна. В большинстве случаев виноваты не настройки питания, а демоны вроде systemd-journald и сбора метрик SMART. Проверить, кто будит диски, можно так: включи по SSH, дальше смотри /var/log/messages и /var/log/synolog/synosys.log, там видно, какие процессы обращаются к диску в момент выхода из сна. Ещё помогает отключить периодический сбор SMART в Центре хранения, снизить частоту логирования до warning и убрать индекс поиска из планировщика. На DSM 7.2 общесистемный сон дисков часто срабатывает только когда оба диска не используются, а с SSD под систему это редкость. Если SSD маленький, выгоднее держать систему на M.2, если модель поддерживает кэш, тогда 3.5″ остаётся единственным активным и засыпает предсказуемо. Ещё вариант: держать базу и логи на SSD, а HDD использовать как архив с планировщиком hyper backup раз в сутки, тогда его сон не зависит от остальной системы.

Leave a Comment

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

Scroll to Top