The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Measure two different outcomes instead of forcing them into one number. Use the ISO/IEC 30134-7:2023 Cooling Efficiency Ratio (CER) method for cooling energy performance, and report Power Usage Effectiveness (PUE) separately for whole-facility overhead. For AI productivity, count completed tokens for a named model and workload, divide by energy measured at a stated boundary, and publish the test conditions so another team can reproduce it.
There is no universally standardized tokens-per-watt definition in the standards and guidance covered here. A credible result therefore identifies the token counter, model, quality requirement, concurrency, interval, and meter boundary. Cooling can change both overhead energy and chip throughput, so a fair comparison reports cooling energy, thermal conditions, and useful output together.
Decide what you are trying to measure
“Efficiency” can mean several different things in an AI facility. Write the question before choosing a meter or formula:
- Cooling-plant efficiency: How much electricity the cooling system uses to move or reject heat at a defined boundary. CER is the purpose-built KPI.
- Facility overhead: How much total data-center energy is required for each unit of IT-equipment energy. PUE answers this broader question.
- AI hardware efficiency: How much energy the accelerators, servers, or IT room use while running a workload.
- Service productivity: How many acceptable output tokens, completed requests, or completed tasks are delivered per unit of energy.
These measures are related but not interchangeable. A site can improve CER while its PUE changes little, or produce more tokens per joule while consuming more cooling electricity. State which outcome is being optimized and publish the other relevant measures alongside it.
#1 Best Overall
- 3-Phase Power Meter With Real-Time Data: see energy usage and solar generation vs. consumption with instant readings in kWh, Watts, Amps, Volts, Power Factor and cost ($).
- Connection Options- by WiFi (2.4 GHz) or wired Ethernet. NEED HELP? Reach out to Eyedro's Technical Support Team for assistance with the setup of your Eyedro device.
- High Resolution Watt Meter Data: once connected, Eyedro maintains full 24/7 usage history. Supporting fixed, tiered and time-of-use rate structures, MyEyedro features responsive real-time graphs, providing you with hourly, daily, weekly and monthly consumption and costs, as well as estimates of what your current electrical bill will cost.
- Multiple Voltage Options: 600V Delta, 480V/277V Wye, 208V/120V, 3-Phase High-Leg Delta and more.
- Directional Solar & Net Metering Ready: differentiates power consumed vs. generated, perfect for solar, grid-tie and net metering scenarios. To break out grid consumption from solar generation and use the MyEyedro Net Meter plugin 2 x Eyedro monitors are required.
Set boundaries and reporting periods before collecting data
Facility boundary for PUE
ISO/IEC 30134-2:2026 defines PUE as total data-center energy divided by IT-equipment energy over the same continuous 12-month period. The numerator and denominator must cover matching dates and the same declared site boundary. Include power-conversion and distribution losses consistently, and document imported energy, on-site generation, and excluded loads.
The standard also distinguishes measurement categories and derivatives such as interim PUE and partial PUE. A result from a month, quarter, commissioning test, or partial building is not an annual PUE; label it as an interim or partial measure and give the interval and boundary.
Cooling boundary for CER
Define whether the cooling boundary contains chillers, pumps, condenser or chilled-water pumps, cooling towers, dry coolers, heat-rejection equipment, air handlers, controls, and room or rack fans. State whether embedded server fans are inside the cooling account or inside IT energy. Keep that choice consistent when comparing sites or operating modes.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchISO/IEC 30134-7:2023 is the standard KPI for cooling energy performance. Its public abstract covers measurement, calculation, reporting, and interpretation, but the detailed formula and category instructions require the full standard. Do not substitute an improvised ratio and call it CER. The abstract also excludes cooling systems that are not powered by electricity, such as heat-driven absorption chillers; disclose that limitation if such equipment is in scope.
IT and workload boundaries
For an AI result, identify the exact point at which energy is measured: accelerator board, server, rack, IT distribution, or the entire facility. A board-level number cannot be compared directly with a facility-level number. If the denominator includes cooling, lighting, storage, networking, or power-distribution losses, say so.
Record the measurement interval and operating state. A warm-up period, a burst run, and a continuous production day answer different questions. For annual infrastructure KPIs, use the required continuous year. For a workload benchmark, use a window long enough to include steady-state behavior and report its duration.
Rank #2
- 3-Phase Power Meter with Real-Time Data: see energy usage and solar generation vs. consumption with instant readings in kWh, Watts, Amps, Volts, Power Factor and cost ($).
- Connection Options- by WiFi (2.4 GHz) or wired Ethernet. NEED HELP? Reach out to Eyedro's Technical Support Team for assistance with the setup of your Eyedro device.
- Multiple Voltage Options: 600V Delta, 480V/277V Wye, 208V/120V, 3-Phase High-Leg Delta and more.
- High Resolution Watt Meter Data: once connected, Eyedro maintains full 24/7 usage history. Supporting fixed, tiered and time-of-use rate structures, MyEyedro features responsive real-time graphs, providing you with hourly, daily, weekly and monthly consumption and costs, as well as estimates of what your current electrical bill will cost.
- This Electric Meter Features Non-invasive Installation: uses split-core current sensors (200A included, 5A–6000A compatible) with no wiring changes, no special tools.
How to measure cooling efficiency with CER
1. Inventory the cooling system
Map the heat path from the IT load to the outdoor environment or a heat-recovery sink. List each electrically powered device and the meter that covers it. The inventory should show which pumps, fans, compressors, towers, valves, controls, and heat exchangers are included. If a device serves both the data center and another building load, allocate it with a documented method rather than silently assigning all of its energy to the data center.
2. Install or validate meters
Use revenue-grade or otherwise documented electrical meters appropriate to the voltage and phase of the equipment. Verify transformer ratios, time stamps, sampling intervals, and calibration status. A three-phase power analyzer may be suitable for a feeder or plant, but a consumer plug-in meter is not a valid substitute for facility feeders.
Where the selected cooling method requires thermal-transfer data, collect the physical quantities used by the hydronic or air-side calculation. For a fluid loop, that commonly means supply and return temperatures, flow rate, and the fluid’s volumetric heat capacity. These quantities help quantify heat moved across a boundary; they do not, by themselves, reproduce every CER category or reporting rule.
3. Apply the ISO method and state exclusions
Calculate CER using the category and boundary rules in ISO/IEC 30134-7:2023. Report the standard edition, the covered cooling equipment, the measurement interval, and any equipment or energy source excluded by the method. If heat-driven equipment is present, explain that the public CER scope does not cover it rather than presenting an electric-only result as the performance of the entire thermal system.
4. Capture conditions that change the result
Log outdoor temperature and humidity, supply-air or supply-liquid set points, rack-inlet temperatures, return temperatures, cooling mode, economizer status, and IT load. A lower temperature set point generally raises cooling demand; overly narrow humidity control can do the same. Hot-aisle and cold-aisle containment, leakage, and mixing should be noted because recirculated hot exhaust makes the cooling plant work harder and can trigger higher fan speeds.
5. Report the result with its time basis
Give the measured cooling energy and the period over which it was collected, not only a rounded ratio. Separate steady-state, seasonal, and commissioning values. A short, favorable weather window is an interim observation, not an annual result and not directly interchangeable with a 12-month standard metric.
Rank #3
- WITH TRANSPARENT SHELL-- This power meter adopts transparent shell, which is convenient to observe the operation status of electronic components inside the meter.
- LCD DISPLAY WITH BACKLIGHT-- This energy meter is equipped with LCD backlight display is used to read data clearly, convenient to operate.
- LOW POWER CONSUMPTION-- The power consumption of the product is very low, saving money. The theoretical power consumption is less than or equal to 2 W.
- HIGH STABILITY-- The kilowatt hour meter has stable performance and can run steadily in the range of -20-55 degrees Celsius.
- APPLICATION-- 3 phase electric power meter can be widely used in power metering systems, standard: IEC62052-11 IEC62053-21.
How to measure PUE without mistaking it for cooling efficiency
Use the basic relationship:
PUE = total data-center energy ÷ IT-equipment energy
The numerator includes the energy used inside the declared facility boundary, while the denominator covers the IT equipment operating within that boundary. Cooling is only one part of the numerator; electrical distribution, lighting, security, controls, and other support systems may also be included.
| Metric | What the numerator represents | What the denominator represents | What it can answer |
|---|---|---|---|
| Annual PUE | Total data-center energy over a continuous 12 months | IT-equipment energy over the same 12 months | Facility overhead per unit of IT energy |
| Interim PUE | Measured facility energy over a shorter declared period | IT energy over that same period | Provisional operating performance; not an annual standard result |
| Partial PUE | A declared portion of facility energy | IT energy serving the declared portion | Performance of a subsystem or partial boundary |
| CER | Cooling-system energy or performance quantities inside the CER boundary | Defined by ISO/IEC 30134-7 categories | Cooling-specific energy performance |
ISO/IEC 30134-2:2026 explicitly says its KPI series does not set limits or targets and does not imply an aggregate “overall” score. Therefore, do not label a facility efficient solely because its PUE is low, and do not combine PUE and CER into an unlabelled composite.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsHow to calculate tokens per watt reproducibly
First choose whether the denominator is energy or power. Energy-based reporting is usually clearer: count defined output tokens over a window and divide by joules or kilowatt-hours measured over that same window. If you report tokens per joule, conversion to tokens per watt-hour requires the time basis because one watt-hour is 3,600 joules.
A ratio of token throughput (tokens per second) to instantaneous power (watts) can be useful as a rate, but call it a rate and state the interval. Do not present it as an unqualified “tokens per watt” figure when the time basis is omitted.
1. Freeze the model and software configuration
Record the model name and version or checkpoint, tokenizer, precision, quantization, accelerator model and count, host CPU and memory, driver, framework, serving runtime, operating-system version, and relevant power-management settings. A model update or runtime change can alter both tokenization and energy use.
Rank #4
- LCD Display Screen: The electric meter uses digital backlight LCD display which shows three-phase voltage, current, power, frequency and other electrical parameters display, active and reactive energy display.
- RS485 Communication: With a RS485 communication interface to host computer for data exchange, the digital power meter comply with the Modbus-RTU protocols, the terminals are marked A and B.
- Real Time Alarm: The three-phase power meter can do real-time alarm of loss of press, over-voltage, under-voltage, phase failure, loss of flow, over-current, current cutoff, overload and so on.
- Harmonic Measurement: Intelligent Power Meter can measure 2-31 harmonics of each phase voltage /current(graded harmonic content, total distortion rate, odd total distortion rate, even total distortion rate).
- Wide Applications: This meter is applicable to campus dormitories and large-scale of electrical energy metering, can also be used in enterprises and institutions for power management assessment.
2. Define exactly what counts as output
Specify whether the numerator includes only generated completion tokens or also input and prompt-processing tokens. State how special tokens, tool calls, rejected outputs, retries, streamed partial responses, and end-of-sequence tokens are handled. Use the same tokenizer and accounting rule for every run.
Free tools Windows power users keep installed
One-click scans. No signup required.
3. Fix the task and quality threshold
Publish the prompt set or a sufficiently precise task description, expected output format, maximum output length, sampling settings, and stopping rule. A shorter answer can produce a higher token rate without delivering the same service. Add a quality or task-completion criterion, such as an accuracy threshold or human-validated acceptance rule, and discard or count failed requests according to a rule declared in advance.
4. Control batch size and concurrency
Record batch size, number of simultaneous requests, queueing policy, context length, and input/output mix. Batching often improves hardware utilization and changes energy per token. A high-concurrency server result should not be compared with a single-user latency run as if they were the same workload.
5. Warm up, then measure a defined window
Warm the model and runtime until compilation, memory allocation, cache creation, and frequency behavior are representative. Mark the start and end time of the measurement window. Capture total energy in joules or kilowatt-hours if the meter supports it; otherwise integrate power samples over time and document the sampling interval. Report tokens per second and average power as separate fields so readers can see whether a ratio changed because of throughput, power, or both.
6. Align numerator and denominator
Count only tokens generated during the energy window, or explicitly account for work that began before or ended after it. If prompt processing is included in the energy but excluded from the token count, the result is biased. The same rule applies to idle power, cooling, storage, and network energy: either include them in both comparable runs or exclude them from both.
7. Publish the result in a complete sentence
A useful statement identifies the workload, output count, duration, energy, boundary, and quality requirement:
Best Value
- 3-Phase Power Meter With Real-Time Data: see energy usage and solar generation vs. consumption with instant readings in kWh, Watts, Amps, Volts, Power Factor and cost ($).
- Connection Options- by WiFi (2.4 GHz) or wired Ethernet. NEED HELP? Reach out to Eyedro's Technical Support Team for assistance with the setup of your Eyedro device.
- High Resolution Watt Meter Data: once connected, Eyedro maintains full 24/7 usage history. Supporting fixed, tiered and time-of-use rate structures, MyEyedro features responsive real-time graphs, providing you with hourly, daily, weekly and monthly consumption and costs, as well as estimates of what your current electrical bill will cost.
- Multiple Voltage Options: 600V Delta, 480V/277V Wye, 208V/120V, 3-Phase High-Leg Delta and more.
- Directional Solar & Net Metering Ready: differentiates power consumed vs. generated, perfect for solar, grid-tie and net metering scenarios. To break out grid consumption from solar generation and use the MyEyedro Net Meter plugin 2 x Eyedro monitors are required.
For {{model and version}} on {{hardware and software}}, {{defined output tokens}} were generated over {{measurement interval}}, using {{energy}} measured at the {{power boundary}}, while meeting {{quality or task-completion criterion}}.
For example, the arithmetic of 2,000 tokens over 100 joules is 20 tokens per joule; that number has no meaning outside the model, task, quality rule, and meter boundary that produced it.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Keep the companion KPIs separate
| KPI | Definition or scope | Use it for | Important qualification |
|---|---|---|---|
| PUE | Total data-center energy divided by IT-equipment energy | Facility overhead | Use matching continuous 12-month periods for the ISO annual metric; it does not measure useful AI output. |
| CER | Cooling-efficiency KPI in ISO/IEC 30134-7:2023 | Cooling-system energy performance | Follow the standard’s detailed formula and categories; its public abstract excludes non-electric cooling systems such as heat-driven absorption chillers. |
| WUE | ISO/IEC 30134-9:2022 water-use KPI; a DOE/FEMP site form is annual site water use in liters divided by annual IT energy in kWh | Water impact of operating the facility | Water and energy can move in opposite directions. |
| CUE | ISO/IEC 30134-8:2022 operational carbon-emissions-intensity KPI | Use-phase carbon impact | Carbon factors and accounting boundaries must be disclosed. |
| Tokens per energy | Defined output tokens divided by measured energy | AI service productivity | Fix model, task, quality, concurrency, token accounting, and meter boundary. |
| Embedded fan or TUE accounting | IT-side fan and thermal energy treatment | Boundary clarity | LBNL notes that total PUE can differ from measured PUE when embedded server fan modules are treated differently. |
Account for cooling’s effect on AI throughput
Cooling is not merely a facility overhead line item. Thermal status can change accelerator frequency, throttling, completion time, and therefore tokens delivered during a fixed interval. Lawrence Berkeley National Laboratory reports that the same job can perform differently under different cooling conditions and that direct-to-chip liquid cooling can keep chips below thermal-throttling thresholds. Those findings do not guarantee a particular saving at a specific site; they show why a cooling comparison must report both energy and workload output.
For every cooling mode, log rack-inlet and chip temperatures, throttling events, accelerator clocks, utilization, fan speed, queue depth, batch or concurrency, and completed tokens. If one system finishes a fixed task faster, include the energy used until completion; if you compare a fixed time window, include the different amount of accepted work. These are different experiments and should not be mixed.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Make comparisons fair across seasons and facilities
- Match the workload: Use the same model version, prompts, token limits, quality test, batch size, concurrency, and software settings.
- Match the hardware boundary: Keep meter locations and included fans, power conversion, networking, and storage identical.
- Repeat under representative conditions: Run across meaningful utilization and outdoor-temperature ranges rather than relying on one cool night or one peak burst.
- Record cooling mode: Identify mechanical cooling, economizer operation, liquid-loop state, set points, and heat-rejection equipment in each run.
- Measure uncertainty: Report meter accuracy, calibration date, sampling interval, and any allocation or estimation method.
- Compare multiple axes: Examine cooling electricity and thermal conditions, IT energy, useful throughput and quality, water, carbon, and the measurement interval together.
Economizer value depends on climate, set points, and the number of hours when outdoor conditions are suitable. Hot- and cold-aisle management should prevent cool intake air from mixing with hot exhaust. DOE/FEMP operating guidance also emphasizes IT environmental conditions, air management, cooling and electrical systems, and heat recovery when assessing efficiency.
What published PUE figures do—and do not—tell you
Published figures are context, not universal targets. The source, date, population, and measurement method matter.
| Figure | Source and date | How to interpret it |
|---|---|---|
| PUE 1.6 | U.S. Department of Energy/Federal Energy Management Program, 2024 Best Practices Guide | Guide-era average-data-center context, not a current universal baseline. |
| Below 1.1 | Same 2024 guide, describing some recent super-efficient centers | An example achieved by some facilities, not a design guarantee or target for every climate and workload. |
| Annual PUE 1.55 in 2022 | Uptime Institute 2022 Global Data Center Survey, cited by the 2024 DOE/FEMP guide | A survey average for large data centers in that year; do not treat it as an AI-specific value. |
| PUE 2.0 and approach to 1.0 | DOE/FEMP Cooling Water Efficiency Opportunities for Federal Data Centers, January 9, 2019 | Older page context with different wording and population from the 2024 guide; the theoretical lower bound is not a practical promise. |
None of these sources publishes a current, AI-specific tokens-per-watt benchmark or a universal CER benchmark. The 2024 guide gives “Gflops or Transactions per Watt” as examples of utilization metrics that can accompany PUE-type measures; that is not a token benchmark.
Common measurement failures and their fixes
- Calling PUE a cooling score: Re-label it as facility overhead and add CER for cooling-specific performance.
- Reporting a one-minute power sample as energy efficiency: Integrate power over the workload window and give the interval.
- Leaving the meter boundary unstated: Name the exact point and list included fans, conversion losses, networking, storage, and cooling loads.
- Counting tokens without a quality rule: Require an acceptance or task-completion criterion and disclose failed or retried requests.
- Comparing different batching or output lengths: Match concurrency, input/output mix, context, and stopping rules.
- Ignoring thermal throttling: Log temperatures, clocks, throttling, and completion time; cooling can change useful throughput.
- Mixing annual and short-period results: Label interim or partial values and never present them as ISO annual KPIs.
- Combining water, carbon, and energy into one score: Report WUE and CUE separately unless a documented aggregation method is supplied.
- Using a consumer meter on a facility feeder: Select instrumentation rated for the electrical installation and have qualified personnel install it.
A practical publication checklist
Before releasing a result, confirm that the report contains:
Recommended Free Tools
- Metric name, standard edition, and declared purpose.
- Facility, cooling, IT, and workload boundaries, including treatment of embedded server fans.
- Meter models or classes, calibration status, sampling interval, and time synchronization.
- Measurement dates, duration, season or outdoor conditions, and whether the result is annual, interim, or partial.
- Cooling equipment and operating mode, set points, inlet temperatures, and thermal-transfer measurements where applicable.
- Model version, tokenizer, hardware, runtime, precision, prompt and response mix, concurrency, warm-up, and quality criterion.
- Token-counting rule, energy units, power units, and a statement connecting numerator and denominator to the same interval.
- PUE, CER, WUE, CUE, and workload-throughput values kept as separate KPIs.
- Uncertainty, exclusions, allocation methods, throttling events, and any failed or retried work.
This structure lets an operator answer the central question without overclaiming: how much cooling energy was used, how much facility overhead surrounded the IT load, and how much acceptable AI work the measured system delivered for that energy.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
