Half, single , double describe the accuracy of a decimal representation. The more bits you have the more information you can store in a variable and the more accurate your decimals are in calculations. But since gaming is such a high performance task the difference in performance between using a float(single) and a double can be significant. A half might not have enough information (at 16 bits vs. 32 bits) to perform the tasks that game developers need to perform accurately. 

PS4 Pro GPU ISA is almost certainly GCN 3 ISA because GCN 2 and earlier didn't have native support for 16-bit float math ...

Some people here might want to look into the basics of how computers handle numbers. It's useful to know the basics of how computers operate, considering how much we're using computers of some kind all the time these days.

In practice, to get double the processing power, you have to sacrifice a lot of precision, even to an extent where it's too much precision loss for a lot of stuff. Just for the record: usually using double precision floating point numbers is recommended in programming. You use single precision FP numbers when you're sure it's know sufficient, and there's quite a limited range of situations where half precision FP numbers are useful.

In the end, this doubles the processing power in a very limited range of situations, but doesn't help most situations. It's an advantage but a limited one.

Yeah most tasks wont be suited for it, and even the ones that are, a lazy programmer wont bother (when in doubt just use the thing thats sure to work).

But its there for those that want to optimise their games.

Not only that, but this Op of this post already made a thread about the PS4's half precision capabilities.
http://gamrconnect.vgchartz.com/thread.php?id=220945&page=1

And you can replied my reply here (Which has the exact same relevence.):

http://gamrconnect.vgchartz.com/post.php?id=8087142

...You guys.. 🙈

Yeah, that's what I picked up from GAF too. I suppose it depends on what demands the game puts on the GPU?

Apparently it helps top down racers, perhaps also sufficient to have any other 2.5D game run in native 4K. Not suitable for HDR rendering I would think. That double flops statement is about as useful as MS doubling the memory bandwidth of esram because you can read and write simultaneously under certain circumstances.

There are other tricks that are useful in that article, improved culling (primitive discard accelerator) to save render workload, inflight compression (DCC) to save bandwidth, improved load balancing or work distributor to make better use of the available CUs, free ID buffer next to the Z-buffer to improve upscaling techniques and temporal AA.

Key in all this is:
"The point though is that these are techniques that can be implemented for a fraction of a per cent for the overall budget for the title," Cerny concludes - and with sample code supplied to developers, there should be relatively few issues in all future titles supporting PS4 Pro.



The big downside in the article, ps4 games will run the same without a pro patch, in compatibility mode (half gpu disable and at original clock speed) as I expected. And the extra 512mb memory is for 4K render buffers, no expectations for higher res texture packs.

🙄

Exactly. Some things need more precision while others can do with less.