Glossary › PC Hardware › TFLOPS
What is TFLOPS?
TFLOPS (tera floating-point operations per second) measures how many trillion calculations a GPU can do each second. It’s a theoretical maximum on a spec sheet, not a measure of how fast games will run.
How it’s calculated
For NVIDIA cards, FP32 TFLOPS = CUDA cores × 2 × boost clock. An RTX 4070, for example, has 5,888 cores at 2,475 MHz, which works out to about 29 TFLOPS.
Why you can’t compare TFLOPS across generations
With the RTX 30 series, NVIDIA doubled the number of FP32 units per core block. TFLOPS roughly doubled on paper, but gaming performance didn’t. An RTX 3070 shows about 20 TFLOPS against about 13 for an RTX 2080 Ti, yet the two perform almost the same in games. TFLOPS is only a fair yardstick within one generation.
What to use instead
For gaming, look at real benchmark results. Our RTX GPU comparison ranks every card from the RTX 2060 to the 5080 on one averaged benchmark score, with full TFLOPS figures listed beside it. For AI work, TFLOPS matters more, but VRAM still decides which models fit at all.