ATI thinks they have the edge in
core and clock speed and they are saying that is going to make up for
some admitted feature ommissions and Nvidia is saying they have the
additional features that are going to make up for a slightly slower
core and clock speed.
Actually the improvements that SM3.0 brings (for gamers) are almost
all related to performance and speed and even that needs to be fully
tested because a) Some of the SM3.0 programming changes, like
branches, texture look-ups by VS3.0 may have a performance hit and b)
it's up to the hardware design (number of registers, FP24 or FP32 etc)
that'll dictate which hardware can run a certain shader faster.
The issue right now, is that nVidia with a likely slower core/clock
speed will try to be competitive with ATI because 6x00 cards may able
to do certain shader operation in single pass that 'll take longer on
X800 cards.
Don't forget that you can't just arbitrarily pick huge shaders that'll
put the GPU under a lot of pressure. Most of the FarCry shaders (that
were demoed by nVidia to show PS3.0 effect, which turned out to be
PS2.0 effects anyway) are hardly 10 or so instructions long.
You can read this interview of CryTek's CEO
http://www.pcper.com/article.php?aid=36
I am just a little bit wary of nVidia's marketing tactics. With NV3x
they came out touting CineFX and FP32 feature (which ATI still doesn't
have even with R420) but it turned out that the performance hit was
non trivial on any NV3x card and they had to drop down the precision
or pick DX8 shaders (in games like Far Cry or the upcoming HL2),
making those features practically worthless for NV3x series.
I really hope that this time nVidia has brought technology that
doesn't prove to be a handicap. My ultimate interest in this debate is
to see a fierce and lively competition between ATI and nVidia, so the
prices go down and I can get the card with best performance to price
ratio, whether it's from ATI or nVidia or someone else.
My 2¢