A pet peeve of mine is when enthusiast PC hardware people say you can't really use FLOPS to compare video cards, due to some variation of the expression "not all FLOPS are equal", or "a FLOP on this card might be worth more than a FLOP on that other card." Huh?

"FLOPS" stands for "floating point operations per second". For example, "1.3 * 2.7" is a floating point operation. When we say that a video card can do 6 teraflops, that means it can do six trillion such operations, every second. If another card can do 4 teraflops-- 4 trillion floating point operations-- it is flat-out not going to be able to render as much geometry at equivalent framerate targets.

A FLOP is a FLOP. "1.3 * 2.7" is the same whether it's done on one video card versus any other. It's a totally apples-to-apples metric! They are created equal, they are the same everywhere, and it is a useful measuring yardstick for the overall performance of a card in terms of rendering geometry!

Is it the end-all of the performance conversation? Obviously not! Video chipsets have pipelines, where an immensely elaborate number of actions are all chained together, in parallel. One chipset may be able to do floating point math more efficiently, but could have a huge bottleneck elsewhere, which means it would render the same scene less efficiently in net, which means lower framerates.

And maybe this is the sentiment these tech people are trying to get across: you can't just look at a video card's FLOPS number and call it a day. But then, why don't they just say that, versus claiming that "2 + 2" is different when I do it, versus when you do it?