GPU performance - theoretical/marketing vs actual?

Your mom

>>55103209

your mom!

>>55103121
These measures are peak FLOPS.

Real-world problems rarely run at peak FLOPS. Converting from peak FLOPS to "real-world" performance isn't a simple linear translation like 20%, 40% or 90%, it's much more complicated than that, ie non-linear and highly algorithm dependent.

If it was a simple linear transformation like that then the Tesla cards would always be best.

>>55103121
I really want a game with the Cornell Box aesthetic. Those perfect white spheres and boxes with the gorgeous bounce lighting, hngg.

>>55103121

The bottleneck in real life applications (mostly games) is memory and not peak performance, followed by driver optimization. Latest Windows 10 store releases are an example of how horrible is to have a Ferrari and not being able to reach the pedals, with a 980ti performing almost on part with an Xbox One. Thank Nvidia for transforming GPUs on blackboxes for developers, and AMD for giving open tools.

>>55103716
mirror's edge

>>55103121
What about uncompressed pixels?
Why isn't /g/ laughing at this right now?

>>55103720
>Latest Windows 10 store releases are an example of
You mean what, like Forza Apex?

>>55103720
I'd say it's more closely linked to pixel and texture fill rate. Go look at those numbers if you want a reliable indication of what the hardware is capable of.

Rest is just driver optimization.

>>55103716
The effect is called Global Illumination/Indirect Lighting, and the reason you don't see it often is there's no good "Cheap" way to do it in real time computationally (most solutions are pre-baked). Nvidia tried with their VXGI tech, but I'm not sure how many games actually use that.

>>55106222

Gears Remaster and Quantum Broken. Forza is an exception, as it is Killer Instinct.