Blade_Runner
Forerunner
Partial precision to be removed from DirectX 10
It's been the cause of many a debate and much controversy since its inclusion in the DirectX 9 specification, but it appears that with Windows Vista on the horizon, and a new version of DirectX to appear alongside it, that partial precision hints (aka FP16) are finally dead as of DirectX 10.
This probably won't come as much of a surprise to most, as we have already seen a tightening of the precision requirements between Shader Model 2.0 and 3.0 as FP32 became mandatory for such parts, so it seems that Shader Model 4.0 will simply finish the job by removing the requirement of or support for partial precision.
If nothing else, this change should make life somewhat easier for developers, who now only have to worry about targeting a single precision, rather than spending swathes of time chopping and changing shaders to see if the use of FP16 can improve performance without adversely affecting image quality. Now the performance penalties are much lower than we have seen in the past (although they are still present on existing parts, as we've seen in a number of our recent reviews), leaving FP16 behind entirely seems like a logical progression from here.
Source
It's been the cause of many a debate and much controversy since its inclusion in the DirectX 9 specification, but it appears that with Windows Vista on the horizon, and a new version of DirectX to appear alongside it, that partial precision hints (aka FP16) are finally dead as of DirectX 10.
This probably won't come as much of a surprise to most, as we have already seen a tightening of the precision requirements between Shader Model 2.0 and 3.0 as FP32 became mandatory for such parts, so it seems that Shader Model 4.0 will simply finish the job by removing the requirement of or support for partial precision.
If nothing else, this change should make life somewhat easier for developers, who now only have to worry about targeting a single precision, rather than spending swathes of time chopping and changing shaders to see if the use of FP16 can improve performance without adversely affecting image quality. Now the performance penalties are much lower than we have seen in the past (although they are still present on existing parts, as we've seen in a number of our recent reviews), leaving FP16 behind entirely seems like a logical progression from here.
Source
