![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Navigation
Install the app
How to install the app on iOS
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
More options
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
SM2.0b being phased out?
- Thread starter kcthebrewer
- Start date
Brent_Justice
Moderator
- Joined
- Apr 17, 2000
- Messages
- 17,755
FanATIc, Precision has nothing to do with speed, and all ATI cards from the 9700 up do vertices and texture coordinates in FP32 internally; only the pixel shader is FP24.
kcthebrewer
[H]ard|Gawd
- Joined
- Jun 14, 2004
- Messages
- 1,728
ATI and nVidia both would be better off supporting multiple levels of precision so that they can run it as fast as possible at only the needed precision. I wouldn't count against them both supporting 16, 24, and 32 precision with their new/next generation...
tornadotsunamilife
2[H]4U
- Joined
- Jun 26, 2004
- Messages
- 2,492
I'm fairly positive they won't, at least not until nearing the end of their cycle.^eMpTy^ said:how do you know they won't have a gig of memory?
kcthebrewer said:ATI and nVidia both would be better off supporting multiple levels of precision so that they can run it as fast as possible at only the needed precision. I wouldn't count against them both supporting 16, 24, and 32 precision with their new/next generation...
I think fp16 is overrated.
It was a nice hack for the FX series... but on a 6800, there's barely any performance-difference between fp16 and fp32. The quality difference however is quite apparent. The 6800 can also keep up with the fp24 stuff from ATi performance-wise.
And fp24 is okay for now, but once ATi goes 32 bit, I don't think fp24 will be interesting either... it doesn't have such severe quality issues as fp16, but the performance difference with fp32 will probably be even smaller.
So I wouldn't mind if they both cut out the 16/24 bit stuff and just made their stuff fp32-only, and just perform well.
Scali said:I think fp16 is overrated.
It was a nice hack for the FX series... but on a 6800, there's barely any performance-difference between fp16 and fp32. The quality difference however is quite apparent. The 6800 can also keep up with the fp24 stuff from ATi performance-wise.
And fp24 is okay for now, but once ATi goes 32 bit, I don't think fp24 will be interesting either... it doesn't have such severe quality issues as fp16, but the performance difference with fp32 will probably be even smaller.
So I wouldn't mind if they both cut out the 16/24 bit stuff and just made their stuff fp32-only, and just perform well.
fp16/fp32 is part of the dx spec...partial precision is useful because most, if not all shaders...don't need more than fp16...
and where are these "severe quality issues"...I get the feeling you're just making that up
^eMpTy^ said:fp16/fp32 is part of the dx spec...partial precision is useful because most, if not all shaders...don't need more than fp16...
The specs only say that partial precision is a hint, which means you require at least 16 bits of precision, rather than at least 24 (SM2.0) or 32 (SM3.0).
There's nothing out-of-spec about just ignoring these hints, and using a higher precision, like ATi already does.
and where are these "severe quality issues"...I get the feeling you're just making that up
Oh please.
Look around for some screenshots of HL2 in partial precision for example... Things like reflections and refractions become very blocky and grainy because 16 bits aren't enough to represent the full 360 degree freedom of a vector in 3d accurately.
Halo has even worse rendering bugs, the whole water goes wrong when forcing partial precision.
See here for example: http://forums.guru3d.com/showthread.php?threadid=115614&perpage=10&highlight=&pagenumber=4