GPU FLOPS Calculator

How to calculate TFLOPS

One TFLOPS (teraFLOPS) is one trillion floating-point operations per second. For a single GPU, peak throughput is:

TFLOPS = cores × boost clock (GHz) × 2 (FMA) ÷ 1000

Example: an NVIDIA H100 SXM5 has 16,896 FP32 cores running at 1.98 GHz, giving 16,896 × 1.98 × 2 ÷ 1000 ≈ 67 TFLOPS of FP32 performance. Tensor Core precisions (TF32, FP16, BF16, FP8) run far higher because the tensor units process whole matrix operations per cycle.

To size a cluster, multiply per-GPU TFLOPS by the number of GPUs, then apply a real-world efficiency factor (model FLOPS utilization, typically 40 to 60 percent for large-model training) to the vendor peak. This GPU FLOPS calculator does that math across mixed GPU and system configurations for every precision at once.

About the GPU FLOPS Calculator

This free GPU FLOPS calculator estimates total floating-point performance (TFLOPS by precision), GPU power consumption (TDP in kilowatts), and memory capacity for any combination of datacenter GPUs and pre-configured systems like NVIDIA DGX and GB200 NVL72. All figures use vendor-published peak TFLOPS adjusted by your chosen efficiency factor.