Originally posted by Sazar
no doubt 16 bit IS useful but remember that mathematically.. the first 6 bits are already assigned for other purposes so for all intents and purposes 16bit is really 10 bits... when you look @ it like that you will see there is a lot more headroom with 24bit than 16bit FP... 32bit FP is good... no doubt but I would have to wonder whether in order to get playable framerates with info stored using 32bit FP is going to be feasible?
that being said I do agree that 16bit FP is not as bad as it is made out to be.. but there are distinct advantages as well...
the problem with all the pixel shaders precision
that both NV30 and R300 cards offer in Doom3
which are FP16 (64bits) , FP24 (96bits) and FP32(128bits) is that ..
1)Doom3 engine and game was designed in the Geforce3 ,so any extra in imagequality for the Nv30 and R300 will be just a minor tweaks,because they arrived much later in the development of Doom3.
just look at carmack own comments..
It is going to require fairly deep, non-backwards-compatible modifications to
an engine to take real advantage of the new features, but working with
ARB_fragment_program is really a lot of fun, so I have added a few little
tweaks to the current codebase on the ARB2 path.
There are some more things I am playing around with, that will probably remain
in the engine as novelties, but not supported features:
The R300 can run Doom in three different modes: ARB (minimum extensions, no
specular highlights, no vertex programs), R200 (full featured, almost always
single pass interaction rendering), ARB2 (floating point fragment shaders,
minor quality improvements, always single pass).
2)Doom3 is a dark game , a very dark one , absence
of light is everywhere ,so for that kind of games even
the precision of a Geforce3 or radeon8500 will be just
3)you need precision for lights ,not for darkness and
not for shadows , black is black .
highly reflective surfaces is when precision is a must ,
like the Nvidia TRUCK demo ,something is not going to
happen in Doom3 or any other game anytime soon.
4)realistic soft shadows are the result of precision in
lights , and doom3 doesnt have soft shadows ,and
most of the lights in the game are in motion,
(bath scene) or blinking on or off. so precision is useless there.
5)the NV30 can do 64bits precision (FP16) which was
what John carmack asked ,for his future games ,
and i dont think he was thinking in doom3
a PS1.3/1.4 game simulation in OPEnGl ,when he
asked for that.
the R300 can do 96bits(fp24) but it will be
invisible extra precision againt Nv30 64bits(Fp16)
that no one will be able to see ,not in a game like
Doom3 ,who was originally designed in a Geforce3
carmack asked for 64bits precision , because
he saw that 32bits was not enough for some
scenarios when there is FAR distance and lights can
can show some graphical errors. something that you
are not going to see in doom3 ,a close combat game ,
with close walls everywhere .
so i agree ,that for Doom3 its not a BIG! issue
which path video cards runs the game ,not for IQ.
the only issue will be for benchmarking purposes .
but i think that 64bits(fp16) its just more than enough
precision for ->doom3 ,the same way 32bits in a tnt2
was more than enough precision for QUake2.
so i think it would not be possible in the future to compare apples vs apples comparison between
ATI and NVdia in games that use FP color precision .
so the only thing we can do is play both cards
at a precision at the best IQ/performance combo.
and that is in the NV30 -64bits(fp16) and
R300 96bits(fp24),because as i have already said
FP16(64bits) is more than enough for the games
we are going to see in a couple of more years.
64bits ,96bits and 128bits precision and beyond it
will be more usefull in FAR FUture games ,with cinematic
stylist look and quality ,but thats not something we
will see anytime soon , being optimistic maybe until Aquamark3,Doom4 or UNREAL 3.
the only thing that will be more usefull in the
close future are a video cards with support for
very long pixel shaders /vertex shaders intructions.