Compute Time Estimator
How long a computational job takes from the operations it needs and the hardware you have — devices, their peak rate, and the utilisation you actually achieve — with the devices needed to hit a deadline and the cost at an hourly rate.
Divide the work by the rate you sustain.
How the compute time estimator works
Divide the work by the rate you sustain. The catch is the rate: hardware is quoted at peak, and real workloads reach 30 to 70% of it depending on how well they keep the arithmetic units fed. A job of 5 × 10¹⁸ operations on eight 50-TFLOP/s devices at 60% is under six hours; at 20% it is a day.
The deadline row inverts the sum: how many devices to finish in the time you have.
Formula: time = operations / (devices × peak rate × utilisation)
Worked examples
| Inputs | Time to complete | Note |
|---|---|---|
| 5 exaFLOP on eight devices | 5.79 hours | under six hours |
| Poor utilisation | 17.36 hours | three times longer |
| A tight deadline | 5.79 hours | many more devices |
FAQFrequently asked questions
How do I estimate compute time?
Operations divided by the rate you actually sustain — devices times peak times utilisation. Everything else is knowing those three numbers honestly.
What utilisation should I assume?
Thirty to seventy per cent of peak for well-tuned numerical work; dense matrix multiplication reaches the top of that range, memory-bound or irregular code the bottom.
What is a petaFLOP?
Ten to the fifteenth floating-point operations. Training a small language model is a few exaFLOP — thousands of petaFLOP — and a large one is tens of thousands of exaFLOP.
Does adding devices scale linearly?
Only up to a point. Communication between devices eats into utilisation as the count grows; the Amdahl page shows the ceiling.
How do I find the operations in my job?
Profile it, or estimate from the algorithm: matrix multiply is 2MNK, a transformer forward pass about two operations per parameter per token.
Where these figures come from
- IEC 80000-13 — Information science and technology (quantities and units) — the decimal (kB, MB) versus binary (KiB, MiB) prefixes used throughout
- RFC 4632 — Classless Inter-domain Routing (CIDR) — the address-plan arithmetic behind the subnet calculator
- RFC 1918 — Address Allocation for Private Internets — the private ranges the subnet calculator recognises
- NIST SP 800-63B — Digital Identity Guidelines, Authentication — length over composition rules; the basis of the password guidance here
- NIST SP 800-57 Part 1 — Recommendation for Key Management — key-strength comparisons used by the key-space calculator
- National Cyber Security Centre — password guidance — UK national guidance on password policy
Last checked: September 2026. Units follow the SI decimal convention (IEC 80000-13 defines the binary alternatives); network and security figures cite the defining standard.