FLOPS Estimator

Summary

Calculate theoretical peak FLOPS for a CPU/GPU or count FLOPs in a neural network layer — dense, convolutional, or attention.

How it works

1. Choose a mode: "Hardware FLOPS" to estimate processor peak throughput, or "Neural Network Layer" to count FLOPs for a specific layer type.
2. For Hardware mode: enter clock speed (GHz), number of cores, FP units per core, and FMA factor (2 for FMA-capable processors).
3. For Neural Network mode: select a layer type (Dense, Conv2D, or Attention), fill in the dimensions, then choose the batch size.
4. Read the result in FLOPS or FLOPs with human-readable and scientific notation.
5. Use the copy button to grab the number for further calculations.

Use cases

• Compare theoretical peak performance between processor models.
• Estimate the compute cost of a neural network forward pass.
• Plan GPU memory and compute budgets before training.
• Convert between GFLOPS, TFLOPS, and raw FLOPs counts.
• Teach students the relationship between clock speed, parallelism, and throughput.
• Sanity-check hardware specs published by vendors.

Frequently Asked Questions