Are TFLOPS (Teraflops) a good indicator of graphical capability?

Brandi93 · June 27, 2023

I have a GTX 1070 laptop which according to TechPowerUp has 6.5 TFlops. I tried Resident Evil 4 Remake on it and barely got 50 fps on high preset with FSR2 to Quality.

I decided to give it a try on my PC, a Vega64 one, which according to the same website, has 12.6 TFlops. The same game performed miles better. With High Preset, no upscaling, I was getting closer to 75 fps. Even the 1% lows were 60+ fps.

So TFlops do make a difference right? When you have 6.5 vs 12.6, there is a clear winner. But has anyone compared two graphics cards, with same CPU and RAM, but for example a 7 TFLOPS one to an 8 TFLOPS one? How much is the difference?

r00tb33r · June 28, 2023

About as good as a car top speed comparison. Is it a good indicator of car's capability? Sometimes?

RONOTHAN## · June 28, 2023

TFLOPS are like clock speed, within a generation it can kind be an indicator of the better card, but there's way more that goes into the performance of the card that just relying on that piece of data alone is not a good strategy. Look at the 4070 vs. 6900 XT. The 6900 XT is the faster card in raster performance, but only by about 5-10%. If you only use the floating point performance though, it's 50% faster (~50 TFLOPS vs. ~30 TFLOPS). The 7900 XT vs. RTX 4080 is probably a better example, the 4080 is undeniably the faster GPU of the two, but it has roughly half the floating point performance of the 7900 XT (~100 TFLOPS vs. ~50 TFLOPS).

If you're just looking at 30 series cards or something along those lines, it can be a decent performance indicator, but at the point where you're comparing across generations or across brands it just doesn't hold up at all.

da na · June 28, 2023

Tflops are the raw computational power, which is far more relevant for GPGPU workloads (say, 3D rendering, crypto mining, machine learning etc) than for gaming. As has been previously mentioned, tflops is not all that matters.