CPU Performance & Performance-per-Watt Estimator
Calculate a processor's theoretical peak floating-point throughput (Rpeak) from cores, sustained vector clock, SIMD width and FMA units, and relate it — or your own measured benchmark score — to power.
Results
- FLOPs per cycle per core
- 32calculated
- Theoretical peak (Rpeak)
- 9,830 GFLOPStheoretical
- Peak per watt
- 13.65 GFLOPS/Wtheoretical
Rpeak is never reached by real applications; HPL typically achieves a large fraction on CPUs, while memory-bound codes achieve far less.
What it calculates
- FLOPs per cycle per core
- Theoretical peak GFLOPS
- GFLOPS per watt
- Measured score per watt
Worked example
Using the default inputs below; change any input above to recalculate.
| CPUs | 2 |
|---|---|
| Cores per CPU | 64 |
| Sustained all-core vector clock | 2.4 GHz |
| SIMD width | 512-bit (AVX-512) |
| FMA units per core | 2 |
| Precision | FP64 (double) |
| Power per CPU (TDP) | 360 W |
| Your measured benchmark score (optional) | 0 |
| FLOPs per cycle per core | 32 | calculated |
|---|---|---|
| Theoretical peak (Rpeak) | 9,830 GFLOPS | theoretical |
| Peak per watt | 13.65 GFLOPS/W | theoretical |
Methodology
Each FMA instruction performs two floating-point operations on every SIMD lane. FLOPs per cycle per core = (SIMD width ÷ precision bits) × 2 × FMA units. Peak = CPUs × cores × sustained clock × FLOPs per cycle.
Formulas
- Rpeak
- R = sockets × cores × f × (W/p) × 2 × FMA_units
Assumptions
- All cores sustain the given clock under vector load.
- TDP approximates package power under sustained load.
Limitations
- Real CPU power can exceed TDP during boost windows.
- Peak FLOPS do not predict memory-bound performance — see the memory bandwidth calculator.