What matters
- Engine, hub, and roller dynamometers observe power at different boundaries with different procedures.
- Drivetrain and tire losses are not one fixed percentage across speed, torque, temperature, gear, setup, and control state.
- Back-calculated crank horsepower is a model output and must never be relabeled as measured engine power.
Why this question matters
A dyno is a measurement system with a correction mode, setup, procedure, and repeatability envelope. The most useful result is a documented comparison, not the largest isolated number.
What the evidence supports
SAE J1349 defines a repeatable installed-net engine power test intended to reflect in-service performance, while SAE 2002-01-0887 explicitly treats inertia-dyno wheel output, SAE net power, and power delivered to the ground as different reference frames with multiple loss sources. Dynojet likewise distinguishes engine-dyno crank power from chassis-dyno wheel power. Atmospheric correction changes a result to a defined reference condition; it does not move the measurement point upstream through the drivetrain.
Name the measurement boundary before comparing numbers
Installed net engine power under SAE J1349 belongs to an engine test procedure and configuration. A hub dynamometer measures at the driven hubs after much of the powertrain but without the wheel-and-tire interface. A roller chassis dynamometer measures the vehicle through driven tires contacting drums or rollers. Those are valid but different system boundaries; “horsepower” alone does not identify which one was observed.
The dyno type also matters inside each category. Inertia systems infer power from accelerating a known inertia, while load-bearing systems can control or sweep operating conditions and measure reaction torque according to their design. Test gear, ramp rate, tie-down, tire pressure and temperature, roller contact, hub adapters, warm-up, control modes, RPM pickup, filtering, smoothing, and atmospheric correction all belong in a fair comparison.
Why a universal loss percentage cannot describe the path
Power is consumed by gear and bearing friction, lubricant churning, seals, pumps, shafts and joints, differential action, tire deformation, and other rotating components. Some contributions vary more with speed, some with transmitted load, and many with lubricant, tire, and component temperature. Automatic transmissions, torque converters, clutches, AWD couplings, and active control strategies can change state during a test.
A single percentage assumes every combined loss scales in constant proportion to engine power. That may be a convenient scenario input, but it is not a law of the drivetrain. Two vehicles producing the same wheel power can have different upstream engine power, and the same vehicle can show a different apparent ratio in another gear, temperature state, acceleration rate, tire setup, or dyno method.
Use the estimate without corrupting the measurement
Preserve the measured wheel or hub result first. If a planning exercise requires an engine estimate, label the assumed fraction, show the equation, and report a range rather than false decimal precision. Never add the percentage directly to wheel power: a model assuming 15% loss divides by 0.85 because the measured output is modeled as 85% of the input.
For evaluating a modification, repeated before-and-after runs on the same dyno with the same configuration are usually more defensible than converting both runs to advertised crank numbers. Report the complete run set, correction standard, smoothing, gear and test procedure, not only the best peak. A change in wheel output can demonstrate a measured difference under those conditions without claiming that the dyno directly measured crankshaft gain.
At 400 whp, a 12% assumption gives 454.5 hp and an 18% assumption gives 487.8 hp—about a 33 hp spread created only by the assumption. These percentages are hypothetical sensitivity inputs, not recommended losses or a confidence interval for the actual engine. Neither estimate is a measured engine output.
Illustrative example, not a measured vehicle test.
A repeatable workflow
- Identify engine, hub, or roller measurement and record the dyno model and method.
- Preserve measured power, correction mode, smoothing, gear, ramp or load, temperatures, tire and tie-down setup, and every run.
- Use same-procedure repeated wheel or hub output for before-and-after comparisons.
- If an engine estimate is unavoidable, show the assumed model and sensitivity range beside—not instead of—the measured result.
Where the shortcut breaks
Coast-down procedures, software models, or separately instrumented powertrain tests can estimate losses only within their stated methods. They do not turn a routine chassis-dyno pull into a direct J1349 engine-power measurement.
Sources and update method
TunerBench prefers government, standards-body, and component-manufacturer documentation. This guide is reviewed against the sources below and should be revisited when regulations, product data, or vehicle technology changes.