By using this site, you agree to our Privacy Policy and our Terms of Use. Close
vivster said:

 

This often quoted number relates to the Parker performance for FP16. In layman's terms FP16 stands for "half precision" which means that the GPU will process certain information less reliable. In games, as in contrast to graphic programs, precision is not important in all tasks. That means it's possible to use only half the precision and will still gain a good enough result.

 

Just a correction you are not entirely correct about this. Percision is not how accuratly the chip does the calculation but how big the instructions are, that is when you are dealing with programming and mathematics. Percision is how "big" the number is, putting it in decimal terms if you are figuring a percentage in some cases .16(16%) is will work well enough, but depending on different factors you may need more percision, .1641(16.41%). How reliable is "target", which depends on the information you are putting into the equation. You can be very percise but if your information isn't on target you will not get the right results. There was a entire chapter devoted to this when I took advanced chamistry in High School and how to determine your percision. Because at some point percision is just usless if the information and calculations don't get that percise, which may affect the target also.

Going back on topic, FP16, floating point 16, are floating point instructions that are only 16-bits big maximum. And FP32 are 32 bit max instructions. For some procedures you only need FP16 and if you optimize correctly you can run two instruction concurently depending on the chip, and from how nVidia is advertising this chip it seems to be the case. So in some cases and if you don't need 32-bit instructions you can process instructions faster this way.

This is one of the reasons why you don't want to look at just FLOPS, or anything else as is. I seen a person saying that the Xbox One runs at twice the speed if you do it in FP16 which may not be the case since the chip might not be able to do two FP16 instructions at the same time, though it may allow them.