FP32 Performance (Single-precision GFLOPS)

The theoretical computing power of the graphics card with single precision (32 bit) in GFLOPS. GFLOPS indicates how many trillion FP32 floating point operations the graphics card (GPU) can perform per second. The larger this number, the faster the graphics card is.

FP32 or "single precision" is a term for a floating point format which occupies 32 bits in computer memory and has a precision between 7 and 8 valid digits. It is laid down in the "IEEE 754" standard and defines how binary data are stored. The calculation according to FP32 is e.g. more complex than according to FP16 (half accuracy).

The FP32 raw performance is of a purely theoretical nature, as it is only a small part of a complex GPU. In addition to the memory equipment (graphics memory), the memory bandwidth or the memory rate also play a major role. The number of execution units is also an important indicator. Modern graphics cards are also very optimized and have different areas for different computing tasks, e.g. for calculating ML (machine learning) or image processing. Ray tracing cores also fall into this area.

However, the performance information does give an initial assessment of the expected performance of a graphics card. Since the FP32 performance can also be determined quite easily from game consoles (e.g. XBox Series X / S or Playstation 5) or internal processor graphics cards (iGPUs), the FP32 performance also enables a cross-system performance assessment.

Thanks to the manufacturer's FP32 performance information, it is also possible to compare graphics cards for which there are no precise benchmarks from games or specific applications. The raw performance is also hardly susceptible to future firmware or driver updates, which can significantly affect the performance values ??in games, for example.

FP32 (GFLOPS)

back to index