Measuring Server Wattage Before You Rent a Boise Rack Unit in 2026

November 20, 2026 · 10 MIN READ

Before renting a single rack unit at IDACORE Boise, measure your server's sustained wall power, not its PSU nameplate rating. Pull readings from IPMI or Redfish, confirm with an inline meter, and run a realistic load. IDACORE bills $300/kW/month on power draw, so a server averaging 200W costs $60/month no matter which U it occupies.

Why Does Wattage, Not Rack Space, Set Your Boise Colocation Bill?

Most colocation pricing you've seen is built around space. You rent a U, a half cabinet, or a full cabinet, and you pay a flat monthly fee whether the box inside pulls 90W or 900W. The provider bakes an assumed power allotment into that fee and pads it, because it has to cover the worst case.

We don't do that at IDACORE Boise. Per-U is how we rent space. Power is how we bill. You can rent one rack unit, and as far as we know we're the only provider in Boise that will, but the monthly number comes from $300/kW/month applied to what your server draws. The U itself doesn't carry a fee.

This changes what matters before you sign. Under a flat per-U model you mostly care whether your gear fits. Under power billing, your wattage is your invoice. A server drawing 200W costs $60/month (0.2 kW × $300). If you guess wrong in either direction, your budget is wrong.

The most common mistake I see is reading the power supply label and treating it as consumption. A 1U box with dual 750W PSUs isn't a 1,500W server. Those supplies are redundant, each sized to carry the whole system alone with plenty of headroom, and in normal operation they split a load that's often a fifth of a single PSU's rating. Budget from the nameplate and you'll overestimate by something like 5–8x. That isn't conservative planning. It's just a wrong number.

How Do You Read Server Power Consumption Through IPMI?

If your server has a BMC (iDRAC, iLO, or the generic BMC on most Supermicro and white-box boards), you already have a power meter. DCMI is the cleanest way to read it:

ipmitool -I lanplus -H 10.0.0.50 -U admin -P 'yourpass' dcmi power reading

On a typical modern 1U you'll get something like:

    Instantaneous power reading:                   187 Watts
    Minimum during sampling period:                112 Watts
    Maximum during sampling period:                291 Watts
    Average power reading over sample period:      196 Watts
    Sampling period:                          00000005 Seconds.
    Power reading state is:                   activated

That's system input power, the AC side, which is what a PDU sees and what you'll be billed on. Don't read too much into the min/max/average fields, though. The sampling window is often just a few seconds, so one reading is a snapshot and not a profile.

To get a profile, poll it. This loop logs one reading per minute to a CSV:

while true; do
  echo "$(date +%s),$(ipmitool dcmi power reading | awk '/Instantaneous/ {print $4}')"
  sleep 60
done >> power.csv

Run it for a week. Seven days captures overnight idle, business-hours load, backup windows, and whatever cron job spikes the CPU at 2 a.m. on Sundays.

What If DCMI Doesn't Work on Your Hardware?

Older boards and some budget BMCs don't implement DCMI power readings. Try the sensor list instead:

ipmitool sdr elist | grep -iE 'pwr|watt|power'

You may get per-PSU input readings. If so, add them together. With redundant supplies sharing load, each one reports only its half.

Newer platforms also expose power through Redfish:

curl -sk -u admin:yourpass \
  https://10.0.0.50/redfish/v1/Chassis/1/Power \
  | jq '.PowerControl[0].PowerConsumedWatts'

The chassis ID varies by vendor. Dell uses System.Embedded.1, and many others use 1. Hit /redfish/v1/Chassis first to list what's there.

Which Measurement Methods Can You Trust?

Not every tool that reports watts is measuring the thing you're billed on. Here's how the usual options compare:

Method What it measures Best use Watch out for
PSU nameplate Maximum rated output capacity Confirming circuit and PDU compatibility Not consumption. Never budget from it
IPMI DCMI / Redfish System AC input power Long-term logging, remote measurement Some older BMCs estimate or omit readings
Per-PSU IPMI sensors Input per supply Boards without DCMI Must sum across redundant PSUs
Inline plug meter Wall power at the outlet Verifying the BMC on a test bench Short-term only, one PSU cord at a time
Metered/switched PDU outlet Wall power per outlet Gold standard once racked You need access to one before you move
turbostat / RAPL CPU package power only Tuning CPU power profiles Misses fans, drives, NICs, PSU losses
nvidia-smi GPU board power only Sizing GPU-heavy 1U/2U boxes Add it to system power, don't substitute it

My recommendation: use the BMC for the week-long log, then spot-check it once against an inline meter on the bench. If the two agree within a few percent, trust the BMC. If they don't, trust the meter, and plan on the PDU reading once you're racked.

On dual-PSU servers, an inline meter on one cord gives you only part of the picture. Either meter both cords and add them, or temporarily run on a single PSU for the bench test.

What Load Should You Measure Under?

Idle numbers are flattering and useless. A server sitting at a login prompt might pull 110W. The same box running your production database at 2 p.m. on a Tuesday could pull twice that.

You want three numbers:

  1. Idle floor: overnight, minimal activity
  2. Sustained average: across a full week of real work
  3. Peak: the worst it gets under full load

If you can't measure against production traffic, simulate it. stress-ng gets you a realistic upper bound:

stress-ng --cpu 0 --vm 2 --vm-bytes 75% --io 2 --timeout 15m --metrics-brief

Log power the whole time. The sustained average tells you roughly what your monthly bill will look like. The peak tells you about headroom and circuit sizing. You need both before you sign.

If there's a GPU in the box, measure it under your real inference or render workload. nvidia-smi --query-gpu=power.draw --format=csv -l 1 shows the card's share, but the BMC reading is still the number that counts.

What Does a Real Pre-Move Measurement Look Like?

Here's a scenario we see constantly. A Treasure Valley SaaS team has a 1U box running in a converted office closet: Postgres replica, a couple of internal services, nightly backups. They want it in a real facility with N+1 UPS and cooling, and they want to know the bill before they commit.

The label says dual 750W PSUs. That's 1,500W of nameplate, which would scare anyone pricing per-kW. So they ran the IPMI logging loop for seven days and a 15-minute stress-ng pass:

Reading Watts
PSU nameplate (2 × 750W) 1,500
Overnight idle floor 135
Business-hours average 230
stress-ng peak 310
7-day sustained average 200

Their sustained average landed at 200W, so their power bill at IDACORE Boise is $60/month: 0.2 kW × $300. The nameplate figure was 7.5x higher than reality.

Two side effects they hadn't counted on. First, closets run warm, so their fans had been spinning harder than they will at controlled data center inlet temperatures. Fan power is a meaningful slice of a 1U's draw, and a cooler environment often trims it. Second, once they saw the numbers, they switched the BIOS from "maximum performance" to a balanced power profile and pulled two spinning drives that weren't doing anything. Under power billing, that cleanup pays for itself every month. Every 100W you remove is a tenth of a kW off the bill.

Try that under a flat per-U contract. You'll save nothing.

What Else Goes Into the Budget?

Power covers the server's footprint. Connectivity is the other line item, and at IDACORE Boise that means IP transit from our own ASN. We run our own BGP and make our own routing decisions. The people who built it founded and operated an ISP and peered at the Seattle Internet Exchange. Transit is billed on committed bandwidth at the 95th percentile, starting at 100M for $170/month, with 1G at $600/month.

That pricing is flat and predictable, and it doesn't punish you for traffic growth the way per-GB hyperscaler egress does. Seven carriers are on-net in the building: Zayo, Lumen/Level 3, Cogent, CenturyLink, Syringa, Cable One, and Hurricane Electric. Latency to Treasure Valley businesses is under 5ms, and Boise sits 23ms from Seattle, 22ms from Portland, and 14ms from Salt Lake City.

IDACORE Boise is housed in a SOC 2 Type II certified facility, a 34,000 SF, 1.4MW purpose-built building, and it's compliance-ready for HIPAA, financial, and government workloads. Standard terms run 12 months rather than the 36 that enterprise providers typically want.

Frequently Asked Questions

How much does it cost to colocate one 1U server in Boise?
At IDACORE Boise, a single 1U server is billed on the power it draws at $300/kW/month, not on the rack unit it occupies. A server drawing 200W costs $60/month (0.2 kW x $300). There's no power commitment minimum, and the smallest space you can rent is one U. Standard terms are 12 months, compared with the 36 months enterprise providers typically require. IP transit is priced separately by commit tier.

How do I check my server's power consumption using IPMI?
Run ipmitool dcmi power reading against the BMC, either locally or over the network with the -I lanplus interface. It returns instantaneous, minimum, maximum, and average input power in watts. On boards without DCMI support, check ipmitool sdr elist for power sensors, or query the Redfish Chassis Power endpoint for PowerConsumedWatts. Poll every minute for at least a week so you capture real workload patterns, not a single snapshot.

Should I use my power supply's wattage rating to estimate colocation costs?
No. The PSU nameplate is the maximum the supply can deliver, and redundant PSUs double that number without doubling consumption. A 1U server with dual 750W supplies commonly runs somewhere between 130W and 300W. Because IDACORE Boise bills $300/kW/month on power draw, pricing from the nameplate would overstate your bill several times over. Measure wall power under a realistic load instead.

Does IDACORE Boise charge a monthly fee per rack unit?
No. IDACORE Boise rents space by the U, starting at a single rack unit, but bills only on power at $300/kW/month. There's no per-U, per-cabinet, or per-square-foot charge. Typical market colocation pricing often uses a flat monthly fee per U or per cabinet, whatever the hardware draws. With power-based billing, a low-draw server costs less, and every watt you cut comes straight off your bill.

What network options come with a single server colocated in Boise?
IDACORE Boise sells IP transit from its own ASN, billed at the 95th percentile on committed bandwidth. The entry tier is 100M at $170/month ($1.70/Mbps). 1G is $600/month, 10G is $2,000/month, and 100G is $11,050/month. Seven carriers are on-net, including Zayo, Lumen/Level 3, Cogent, and Hurricane Electric, and latency to Treasure Valley businesses is under 5ms.

Once you've got a week of IPMI logs and a stress-test peak, you know exactly what a Boise rack unit will cost you. There's no padded power allotment and no flat per-U fee, just your measured draw at $300/kW/month, supported by humans who'll know your rack by name. Send us your measured wattage and get a Boise per-U colocation quote.

Ready to Implement These Strategies?

Our team of experts can help you apply these per-u colocation techniques to your infrastructure. Contact us for personalized guidance and support.

Get Expert Help