Estimate AI training compute, PFLOP-days, GPU hours, and training time from model size, training tokens, GPU count, and hardware throughput.
Calculator
Result Summary
Review the main answer, supporting values, chart breakdown, and calculation details.
Result Breakdown
Calculation Details
| Metric | Value | Note |
|---|
How it works
Estimate AI training compute, PFLOP-days, GPU hours, and training time from model size, training tokens, GPU count, and hardware throughput. Use the units shown in the calculator and review every assumption before using the result.
Formula
Training FLOPs ≈ 6 × model parameters × training tokens. Training time = total FLOPs ÷ effective GPU FLOPs per second.
Worked example
A 7B parameter model trained on 300B tokens needs about 12.6e21 FLOPs before accounting for GPU throughput and utilization.
Important assumptions
This tool is designed for planning and comparison. Real-world prices, manufacturer ratings, building requirements, lender terms, electrical codes, tolerances, and product specifications can vary.
Frequently asked questions
Is this calculator free?
Yes. It runs in your browser and does not require an account.
Are the results exact?
The math is deterministic for the entered values, but the result depends on inputs and assumptions. Verify safety-critical, financial, code-related, towing, fitment, and manufacturer-specific decisions with authoritative information.