Skip to content
ServerCalc ServerCalc
Power & Energy

Supermicro Server Power Consumption Calculator

Power draw, heat and running cost across 22 SuperServers, from a 2019 X11 1U to an 8-way Blackwell node.

Inputs

Server

Loads that model's chassis, socket, memory and PSU defaults.

Totals scale across the rack or fleet.

2U chassis · 2 socket(s) · up to 12 drive bays

Processors & memory

TDP drives most of the dynamic power.

W

Empty slots draw nothing.

512 GB installed, about 50 W

Storage & accelerators
W

NICs, HBAs, DPUs and RAID controllers not already counted.

Power supply & load
W

Rating of the supplies actually carrying load: one unit in a 1+1 hot-spare pair, the whole active bank on a GPU node. Nameplate, not draw.

Sets utilization. Pick Custom to drag the slider.

Three-phase carries the same watts at 1/√3 of the line current.

Real power over apparent power. Sets the watts-to-VA gap.

40%
W

Read it off iDRAC, iLO, XCC, IPMI or a metered PDU. Overrides the estimate and recalibrates the whole curve. Leave at 0 to estimate.

Operating profile & cost

24 for always-on production.

$/kWh

Your blended rate per kWh, taxes included.

Facility overhead multiplier. 1.0 counts IT load only.

Grams of CO2e per kWh on your grid.

Rack & facility

Defaults to a full 42U rack minus 2U for switching.

The breaker feeding the rack PDU.

sq ft

Optional. Gives watts per square foot for the room.

kg

kgCO2e to build one server, from the vendor's Product Carbon Footprint report.

Years the manufacturing carbon is amortized over.

Supermicro boards ship with fan control on Optimal. The fan curve here matches that, not the louder Full Speed default some SKUs use.

  • A full rack peaks at 17190 W, more than the 4892 W a single circuit delivers. Budget 4 circuits per rack.
Results update live as you type.

Results

Estimated power draw
487 W
SYS-221H-TNR (2U Hyper) at 40% CPU utilization
Idle draw
261 W
Powered on, no workload.
Max draw (100% load)
860 W
Sustained worst case, not a boost spike.
PSU efficiency here
94.6%
29% of the 1600 W nameplate
Annual energy (IT)
4,268 kWh
24 h/day × 365 days/year
Energy cost per month
$53.35
Energy cost per year
$640
Facility power (PUE 1.50)
731 W
Facility cost per year
$960
IT load plus cooling and distribution overhead.
Heat output
1,663 BTU/hr
Cooling required
0.14 tons
One ton of cooling is 12,000 BTU/hr.
Current draw
2.4 A
At 208 V, power factor 0.98.
Apparent power
497 VA
Minimum UPS size
622 VA
Apparent power plus 25% headroom.
Operational carbon
2,459 kg CO2e/yr
384 g CO2e/kWh applied to facility energy.
Manufacturing carbon
256 kg CO2e/yr
1,280 kg per server over 5 years.
Total carbon
2,715 kg CO2e/yr
Operational plus manufacturing. 9% of it is manufacturing.
Servers per circuit
5
30 A at 208 V Single-phase, 80% derate, 4,892 W usable.

Per rack

Rack draw at this load
9.75 kW
20 servers per rack.
Rack draw at 100%
17.2 kW
The figure the PDU and breaker have to survive.
Circuits per rack
4
Rack space used
40U
Of 42U available.

Where the watts go

487 W from the wall

At 40% CPU utilization, per server.

CPUs
268 W · 55%
Memory
50 W · 10%
Storage
55 W · 11%
Board, BMC & I/O
56 W · 11%
Fans
32 W · 7%
PSU conversion loss
26 W · 5%

Power vs CPU utilization

Idle is 30% of max draw

Server power is not proportional to work done. A box at 0% utilization still burns its idle floor, which is why consolidation beats tuning: switching a server off saves the whole curve, tuning it saves only the sloped part.

02505007501.0k0%25%50%75%100%Your setting: 487 WCPU utilizationWall power
Wall powerIdle draw: 261 WMax draw (100% load): 860 W

About this calculator

Supermicro is the awkward one to size. Dell and HPE sell a few dozen configurations of a handful of chassis and publish a planning tool for each. Supermicro sells hundreds of SKUs, mixes and matches boards and chassis freely, and expects you to read the datasheet. That flexibility is exactly why people buy it, and exactly why a generic wattage table is useless for it.

So this calculator treats Supermicro the way you actually buy it: pick the closest chassis class, then describe your real build. Twenty-two systems are preloaded across four generations, and you change the CPU, DIMM count, drive counts and accelerators to match your bill of materials.

  • Current mainstream: the 1U SYS-121H-TNR and 2U SYS-221H-TNR Hyper systems on Intel Xeon, the AS-1115HS-TNR and AS-2115HS-TNR on AMD EPYC, and the SYS-621C-TN12R CloudDC.
  • Hopper-era AI: the 4U SYS-421GU-TNXR with four HGX H100 modules, the 4U SYS-421GE-TNRT with eight PCIe cards, and the 8U SYS-821GE-TNHR and AS-8125GS-TNHR with eight.
  • Blackwell-era AI: the 4U liquid-cooled SYS-422GS-NBRT-LCC, the 8U air-cooled SYS-822GS-NBRT, the 10U AS-A126GS-TNBR on EPYC, and the extraordinary 2U SYS-222GS-NB3OT-ALC that puts eight B300s in two rack units.
  • Still in the field: the X12 generation from 2021 (SYS-120U-TNR and SYS-620U-TNR Ultra, SYS-620C-TN12R CloudDC, SYS-420GP-TNR and SYS-420GU-TNXR GPU systems on Ice Lake) and the X11 generation from 2019 (SYS-1029P-WTR, SYS-6029P-TR, and the SYS-4029GP-TRT that ran a lot of the first deep learning wave). Refresh cases need the old number as much as the new one.
  • Dense storage: the 60-bay SSG-640SP-E1CR60, where the drives, not the CPUs, are the load.

Two things matter more on Supermicro than on the tier-one vendors. Fan mode is one: boards ship with fan control on Optimal, but plenty of SKUs and plenty of admins run Full Speed, which costs real watts for no compute. Titanium supplies are the other: Supermicro was early and aggressive on high-efficiency PSUs, and on a dense build the difference against a Platinum unit is worth having.

If the hardware is racked, read the truth off the BMC instead. IPMI reports it: ipmitool sensor list | grep -i pwr, or the power page in the IPMI web UI. Put that figure in the Measured draw field and the whole curve recalibrates to your machine.

The formula

Wall power = (CPUs + memory + drives + GPUs + board + fans) ÷ PSU efficiency at that load

Each component contributes an idle floor plus a share that scales with CPU utilization; PSU efficiency is taken off the 80 PLUS curve at the real load ratio, not the badge number.

Supermicro-specific things that move the number:

  • Fan mode. Standard and Optimal keep fans on a thermal curve. Full Speed pins them, which on a 2U with eight counter-rotating fans can be 80 to 120 W of pure overhead, plus the noise. Check with ipmitool raw 0x30 0x45 0x00. The fan figures here assume Optimal in a normal cold aisle; if you run Full Speed, add the difference under Other devices.
  • Titanium supplies. Most current Hyper and A+ systems ship 1200 W to 2000 W Titanium units at 96% peak efficiency. That is roughly 4 points better than Platinum at half load, which on a 24/7 node is a few hundred kWh a year.
  • Redundant vs single PSU. Many Supermicro SKUs ship single-PSU by default. A single supply carries the full load, so it sits higher on the efficiency curve, which is good for efficiency and bad for availability. The model here assumes one active supply either way.
  • Chassis mixing. Because boards and chassis are interchangeable, the platform and fan figures here are class averages for the form factor. If your build is unusual, the Measured draw field will beat any estimate.
  • Power supply banks. A 1U Hyper runs one active supply. The SYS-821GE-TNHR runs six 3,000 W Titanium units in 4+2, with an 8 x 3,000 W 4+4 option if you swap the centre fans. The PSU capacity field holds the capacity of the units actually carrying load, which is why an 821GE reads 12,000 W and not 3,000 W.
  • Liquid cooling changes density, not heat. Compare the 8U air-cooled SYS-822GS-NBRT against the 4U liquid-cooled SYS-422GS-NBRT-LCC: same eight B200s, but the liquid system spends about 480 W less on fans and fits in half the rack units. The 10 kW of accelerator heat is identical, it just leaves through a coolant loop instead of the back of the chassis.

Facility figures follow: BTU/hr = watts × 3.412, amps = watts ÷ (volts × PF) single-phase or ÷ (√3 × volts × PF) three-phase, kWh/year = watts × hours × days ÷ 1000.

Common use cases

  • Sizing PDUs and circuits for a Supermicro build where no vendor planning tool exists
  • Costing a GPU row of SYS-821GE-TNHR or SYS-422GS-NBRT-LCC nodes before committing to the power feed
  • Comparing an air-cooled 8U B200 node against the 4U liquid-cooled version on fan power and density
  • Building a refresh case: what an X11 SYS-6029P-TR costs to run against a current Hyper system
  • Deciding whether Titanium supplies pay back against Platinum at your tariff and load
  • Quantifying what Full Speed fan mode is costing across a fleet
  • Comparing a Supermicro build against an equivalent PowerEdge or ProLiant on running cost
  • Getting BTU/hr and cooling tons for a room of SuperServers

Frequently Asked Questions

How do I check actual power draw on a Supermicro server?
Through IPMI. On the command line, `ipmitool sensor list | grep -i pwr` or `ipmitool dcmi power reading` will give present, minimum, maximum and average draw on boards that support DCMI. In the IPMI web interface it is under Server Health, then Power Consumption, on X11 and later boards. Redfish also works on X12/X13 and H12/H13 boards via /redfish/v1/Chassis/1/Power. Whatever figure you get beats an estimate; enter it in the Measured draw field here and the calculator scales its whole curve to match.
Why is my Supermicro server drawing more than expected at idle?
Fan mode is the usual culprit. Many SKUs ship with, or get switched to, Full Speed, which pins every fan regardless of temperature. On a 2U with eight counter-rotating fans that is 80 to 120 W of pure overhead. Check with `ipmitool raw 0x30 0x45 0x00`: 0 is Standard, 1 is Full, 2 is Optimal, 4 is Heavy IO. Set Optimal with `ipmitool raw 0x30 0x45 0x01 0x02`. After that, check the BIOS CPU power management settings, since some Supermicro boards default to disabling C-states on performance-oriented SKUs.
How much power does a Supermicro 1U or 2U server use?
A 1U SYS-121H-TNR with two Xeon Gold 6430, 16 × 32 GB DDR5 and four NVMe drives idles near 230 W, sits around 450 W in production and peaks near 820 W. The 2U SYS-221H-TNR runs 20 to 40 W higher because of the larger fan wall and more drive bays. A single-socket AS-1115HS-TNR on an EPYC 9354 idles near 180 W and peaks near 520 W. GPU systems are a different scale entirely: an 8-way SYS-821GE-TNHR with H100 SXM modules peaks past 8 kW.
How much power does an 8-GPU Supermicro server need?
Around 8 to 10 kW at full load for an H100 or H200 SXM configuration, and more for the newest parts. The accelerators dominate: eight H100 SXM modules at 700 W each is 5.6 kW before you count two CPUs, 2 TB of memory, NVMe, the networking, the fans and PSU losses. That is a whole traditional rack's worth of power in 8U. Plan for three-phase distribution, multiple circuits per rack, and either rear-door heat exchangers or direct liquid cooling. Select the model here, set the phase and circuit rating, and the calculator tells you how many circuits per rack you need.
Are Supermicro Titanium power supplies worth the extra cost?
On anything running 24/7, usually yes. Titanium is about 96% efficient at half load against roughly 92% for Platinum. On a server drawing 500 W at the wall, those four points are about 175 kWh a year, plus the cooling energy no longer needed to remove that waste heat. Across a rack it compounds. The exception is a lightly loaded server: efficiency falls off below 20% load for every grade, so a right-sized Platinum unit can beat an oversized Titanium one. Compare by setting the PSU capacity and grade fields and watching the PSU efficiency here row.
How much power does a Supermicro Blackwell B200 or B300 server need?
Around 10 to 12 kW per node at full load. The 4U liquid-cooled SYS-422GS-NBRT-LCC and the 8U air-cooled SYS-822GS-NBRT both carry eight HGX B200 modules at 1,000 W each, which is 8 kW of accelerator before the CPUs, up to 3 TB of DDR5, NVMe, networking and conversion losses. The 2U SYS-222GS-NB3OT-ALC with eight B300s at 1,200 W each peaks near 12 kW in two rack units, which is only possible because direct liquid cooling takes essentially all the GPU heat out through coolant. Select the model, set your phase and circuit rating, and the calculator reports circuits per rack.
Does liquid cooling reduce a Supermicro AI server power consumption?
A little, and not where people expect. Compare the two eight-B200 systems here: the air-cooled 8U SYS-822GS-NBRT spends around 700 W on fans at full load, the 4U liquid-cooled SYS-422GS-NBRT-LCC about 220 W. That is a real saving of roughly 480 W per node, about 5% of system power, plus whatever your facility saves on chilled air. What liquid cooling does not do is reduce the accelerator heat: eight B200s produce 8 kW either way. The win is density and facility efficiency, letting you put the same GPUs in half the rack units and reject the heat to water instead of air.
How many Supermicro GPU servers fit in a rack?
Power decides, not rack units. Twenty 2U SYS-222GS-NB3OT-ALC nodes physically fit a 42U rack and would draw about 240 kW, which no conventional rack can deliver or cool. This calculator defaults servers-per-rack to whichever constraint binds first, height or a 60 kW rack power budget, which is why a dense GPU node defaults to five per rack rather than twenty. Change the figure in the Rack and facility section to match your actual power feed, and the calculator will tell you the rack draw, the circuits needed, and warn you when a full rack exceeds one circuit.
Is it worth replacing an old X11 or X12 Supermicro server?
Run the numbers rather than assuming. An X11 SYS-6029P-TR with dual Xeon Gold 6248R idles near 210 W and peaks near 660 W. A current SYS-221H-TNR idles near 260 W and peaks near 860 W, so the new machine draws more, while doing several times the work. Electricity alone rarely justifies a refresh; performance per watt does. The other half of the calculation is embodied carbon: building a new server costs roughly 1,000 to 1,600 kgCO2e before it is switched on, so on a clean grid keeping the old one longer can be the lower-carbon choice. Both generations are in the dropdown, and the Total carbon row shows operational and manufacturing side by side.
My exact Supermicro SKU is not in the list. What do I do?
Pick the closest chassis by form factor and socket count, then correct the fields to match your build. Supermicro's part numbers encode this: in SYS-221H-TNR, the 2 is 2U, the 21 is the generation and board family, H is Hyper, and TNR indicates the CPU and networking configuration. What actually drives the power number is the CPU TDP, socket count, DIMM count, drive counts and accelerators, all of which you can set directly. The chassis choice only affects the platform and fan baseline, which is a small share of the total on a loaded machine.
Does this cover BigTwin, FatTwin, SuperBlade and GB200 NVL72 racks?
Not the multi-node or rack-scale systems, no, and deliberately. Twin and Blade chassis share power supplies and fans across two, four or more nodes, so per-node draw depends on how many bays are populated and the shared overhead does not split evenly. The SRS-GB200-NVL72 and SRS-GB300-NVL72 are 48U rack-scale systems with 72 Blackwell GPUs and 36 Grace CPUs on shared liquid cooling and power distribution, rated at 200 to 250 kW per rack; they are the rack, not a server you put in one, so plan those from the published per-rack figures. Every single-node SuperServer is covered here. If you want proper multi-node support, say so through the feedback form.

Spot an error? Have feedback?

Tell us what is wrong with the math, what is missing, or which server model you would like added. We read everything.

Related Calculators

Power & Energy

NVIDIA Server Rack Power Consumption Calculator

NVIDIA rack power calculator for GB200 and GB300 NVL72, DGX B200, H200, H100 and HGX nodes. Get kW per rack, amps, circuits, liquid cooling load, weight and annual cost.

Open calculator
Power & Energy

42U Server Rack Power Consumption Calculator

Build a 42U rack from real servers, switches and storage and get its power draw in kW, amps, circuits, heat in BTU/hr, weight and running cost. Space, power and weight budgets in one view.

Open calculator
Cooling & Thermal

Server Room BTU Calculator

Server room BTU calculator for air conditioning sizing. Work out BTU/hr heat load from your servers, UPS, lighting and walls, then the AC capacity in BTU and tons to buy.

Open calculator
Virtualization

Terminal Server Calculator

Terminal server calculator for RDS session hosts: users per server, hosts needed, vCPU and RAM, profile storage, bandwidth and running cost. Built on Microsoft's sizing guidance.

Open calculator
Power & Energy

Server Power Consumption Calculator

Server power consumption calculator for Dell PowerEdge, HPE ProLiant, Lenovo ThinkSystem, Supermicro and Cisco UCS. Get watts, kWh, running cost, BTU/hr and amps.

Open calculator
Power & Energy

Dell Server Power Consumption Calculator

Dell PowerEdge power consumption calculator covering R770, R670, R7725, R760, R750, R740 and the AI line: XE9680, XE9680L, XE9685L, XE8640, XE9640, XE7745, R760xa.

Open calculator
Power & Energy

HP Server Power Consumption Calculator

HP / HPE ProLiant power consumption calculator for DL360 and DL380 Gen10 and Gen11, DL325, DL345, ML350 and DL20. Watts, kWh, cost, BTU/hr and amps.

Open calculator