MaxxPI³ is a CPU benchmark for Windows x64
 
It measures how fast your processor can compute the number π to millions of decimal places. In addition, it detects your
hardware configuration via SIV (CPU, mainboard, memory modules, timings) and lets you submit your score to the online database.
 
Computing π at this scale is one of the most demanding, provably correct workloads a CPU can run: it hammers the integer execution
units, the caches and – at large digit counts – the memory subsystem, and any computational error, anywhere, changes the result.
 
That makes MaxxPI³ two things at once:
  • A performance benchmark: How much sustained arithmetic throughput does your system deliver?
  • A stability test: An overclocked system that produces even a single wrong bit will fail the built-in verification and MaxxPI³ can tell you which core was at fault.
How does the calculation work?
 
MaxxPI³ uses the Chudnovsky algorithm, the same formula used for world-record π computations with binary splitting, the term range is
recursively split, partial products are computed as exact big integers, and the parts are combined pairwise. This turns the problem intolarge
integer multiplications which is exactly what makes it a good CPU test. The big-number arithmetic is performed by the GMP & MPFR libraries
 
What do Single and Multi mode measure?
  • Single: The entire digit count is computed by one thread. This measures pure single-core arithmetic throughput IPC, clock speed & cache performance of a single core.
  • Multi: The total digit count is split into independent single-thread computations, one per logical processor (up to 64), each pinned to its core.
All cores stay fully loaded for the entire runtime there is no serial end-phase where cores idle. This measures sustained all-core throughput: clocks under full load,
thermals and power limits become decisive. Multi mode has a second purpose: because every instance computes the same number of digits with the same algorithm,
all results must be bit-identical. MaxxPI³ cross-compares all instances (majority vote). If one core computed a different result, the run fails verification and the faulty
core is named a precise overclocking stability diagnosis.
 
What is decisive for the score?
 
The score is simply computed digits per second, displayed in thousands of digits per second: 
Score = (total digits ÷ runtime in seconds) ÷ 1000, rounded. Example: 256 million digits in 73.6 s → score 3478
 
What influences it:
  • Clock speed – in Single mode the maximum boost clock of one core; in Multi mode the *sustained all-core* clock (cooling and power limits included).
  • Caches and memory speed – at small digit counts the working set lives mostly in cache; the larger the digit count, the more memory bandwidth and latency contribute.
  • Core count – Multi mode scales with the number of logical processors, as long as clocks can be held.
Because the score is digits per second, results are comparable across different digit counts – but larger runs stress memory harder and hold the load longer,
so they are the tougher (and more meaningful) discipline.
 
How is the result verified, two independent mechanisms:
  1. BBP tail check (every run): After the computation, the last ~32 bits of the result are re-derived with the Bailey–Borwein–Plouffe formula – a completelyindependent method that can extract digits of π at an arbitrary position without computing the digits before it. Any error anywhere in the main computation changes the final bits; the probability that a wrong result still passes is about 2⁻³⁰. Only runs that pass ("OK !") can be submitted
  2. Cross-verification (Multi mode): All per-core results must match bit-for-bit; a majority vote localizes deviating cores.
The verification runs after the timed computation and does not count towards the score.
 
Can I abort a running benchmark?
 
Yes. During a run the Start button becomes Break. The abort is cooperative: the compute threads stop at the next checkpoint and the run is discarded
completely – no mix of old and partial results. (In the short finalize phase the abort takes effect at the end of the phase.)

What gets uploaded?

Only what you see and confirm: your hardware configuration (as detected by the scan), the benchmark settings and score, and a screenshot of the result window.
Upload is only possible for completed, verification-passed runs, and only once per run.