Chinchilla Compute-Optimal LLM Pre-Training FLOPs & Token Ratio Calculator

Apply C = 6ND to size LLM training FLOPs, tokens, and H100 GPU hours.

Cluster Training Duration (Days)
24.8 days
Total Compute Budget (×10²¹ FLOPs) 96 ZettaFLOPs
Tokens-per-Parameter Ratio 250 tok/param
Chinchilla 20× Optimal Token Target 160 B tokens
Inputs
B params
Model parameter count (N) in billions.
T tokens
Total pre-training dataset tokens (D) in trillions.
TFLOPS/GPU
Effective achieved TFLOP/s per GPU (after MFU).
GPUs
Number of GPUs in training cluster.
Results & Graph
Cluster Training Duration (Days)
24.8 days
Total Compute Budget (×10²¹ FLOPs)
96 ZettaFLOPs
Tokens-per-Parameter Ratio
250 tok/param
Chinchilla 20× Optimal Token Target
160 B tokens

Cluster Training Duration (Days)

Graph 1 • Model Parameters Billions Sensitivity Recalculates the domain equation across 50% to 150% of the primary input while holding other parameters constant.
0 days 20.1 days 40.2 days 50%: 12.4 days 12.4 50% 75%: 18.6 days 18.6 75% 100%: 24.8 days 24.8 100% 125%: 31 days 31 125% 150%: 37.2 days 37.2 150%
Graph 2 • Primary & Supporting Metric Comparison Compares the primary output alongside supporting domain benchmarks.
Cluster Training Duration (Days) 24.8 days Total Compute Budget (×10²¹ FLOPs) 96 ZettaFLOPs Tokens-per-Parameter Ratio 250 tok/param Chinchilla 20× Optimal Token Target 160 B tokens
Graph 3 • Exact Output Table & Units Displays each calculated output in its exact domain unit.
Cluster Training Duration (Days)24.8 daysTotal Compute Budget (×10²¹ FLOPs)96 ZettaFLOPsTokens-per-Parameter Ratio250 tok/paramChinchilla 20× Optimal Token Target160 B tokens
Summary: Pre-training an 8B model on 2T tokens requires 96.00 × 10²¹ FLOPs, taking 24.8 days across 128 GPUs.
Summary: Pre-training an 8B model on 2T tokens requires 96.00 × 10²¹ FLOPs, taking 24.8 days across 128 GPUs.
⚡ Send result to: API Token Cost →Subnet & Password Entropy →Aspect Ratio & Storage →
Interactive chart & math engine powered by CalculatorGraph

Formula & Mathematical Methodology: Chinchilla Scaling Law Systems Engineering Formula

ComputeFLOPs = 6 × ParametersN × TokensD
GPUHours = FLOPs / (GPUTFLOPS × 10¹² × MFU × 3600)
Compute_FLOPs = 6 × Parameters_N × Tokens_D; GPU_Hours = FLOPs / (GPU_TFLOPS × 10¹² × MFU × 3600)
  • modelParametersBillions: Model Parameters Billions (B params)
  • trainingTokensTrillions: Training Tokens Trillions (T tokens)
  • sustainedGpuTflops: Sustained Gpu Tflops (TFLOPS/GPU)
  • clusterGpuCount: Cluster Gpu Count (GPUs)

Rounding: Binary/decimal capacities and engineering metrics retain full floating-point precision and round to 2 decimal places.

Worked Example: Architecture Sizing Example: Chinchilla Scaling Law

Apply C = 6ND to size LLM training FLOPs, tokens, and H100 GPU hours.

  1. Enter the workload parameters, model dimensions, or network link specifications.
  2. Compute the sizing equation: Compute_FLOPs = 6 × Parameters_N × Tokens_D; GPU_Hours = FLOPs / (GPU_TFLOPS × 10¹² × MFU × 3600).
  3. Inspect the primary capacity metric, headroom buffers, and 3 SVG visualizations.
Result: Calculates Chinchilla Scaling Law via Compute_FLOPs = 6 × Parameters_N × Tokens_D; GPU_Hours = FLOPs / (GPU_TFLOPS × 10¹² × MFU × 3600).

Frequently Asked Questions

How does the Chinchilla Compute-Optimal LLM Pre-Training FLOPs & Token Ratio Calculator compute capacity?

It evaluates Compute_FLOPs = 6 × Parameters_N × Tokens_D; GPU_Hours = FLOPs / (GPU_TFLOPS × 10¹² × MFU × 3600) directly in your browser and plots three custom SVG diagrams.

Can I compare different cluster or model configurations side by side?

Yes. Use the "Compare (0/3)" button in the results panel to pin up to three configurations.

Does this tool upload any of my architecture parameters?

No. All calculations run 100% client-side in your browser.

Developer Tools · AI & LLM Infrastructure

Chinchilla Compute-Optimal LLM Pre-Training FLOPs & Token Ratio Calculator: inputs, method, and how to read the result

Apply C = 6ND to size LLM training FLOPs, tokens, and H100 GPU hours.

Choose inputs for the case you are checking

The defaults are illustrative. Replace them with values for your case and keep the units shown beside each field.

  • Model Parameters Billions: Model parameter count (N) in billions.
  • Training Tokens Trillions: Total pre-training dataset tokens (D) in trillions.
  • Sustained Gpu Tflops: Effective achieved TFLOP/s per GPU (after MFU).
  • Cluster Gpu Count: Number of GPUs in training cluster.

What the calculation includes

With the default values shown, Cluster Training Duration (Days) is 24.8 days. The calculation follows Chinchilla Scaling Law Systems Engineering Formula; the formula and variable definitions above show how the inputs combine.

One-input check: With the other defaults held constant, changing Model Parameters Billions from 8 B params to 9 B params changes Cluster Training Duration (Days) from 24.8 days to 27.9 days. This is a comparison inside the calculator's stated model, not a prediction.

Where to use caution

  • Uses industry-standard IEEE, IETF RFC, and GPU/cloud hardware sizing conventions.
  • Capacity planning estimate; benchmark under production traffic patterns before finalizing cluster or hardware procurement.

Scope: IEEE / IETF / Cloud Systems Engineering Standard Check the assumptions and limitations against your situation before relying on the estimate.

More in Developer Tools: API Token Cost Calculator, IPv4 CIDR Subnet, Password Entropy & URL/Base64 Encoding Calculator, Aspect Ratio, Image Dimensions & Storage Converter.