| View
previous topic :: View
next topic |
| Author |
Message |
Ailuros

Joined: 21 Jun
2002 Posts: 49
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
Did it installed it
successfully so I'm finally on 29.42 WHQL effectively.
Concerning anisotropic my findings seem though to go along
with those of Unwinder at nvnews:
| Quote: |
NV25 GPUs do have some AF related
issues. It looks like fillrate drops drastically when
more two (or more) textures are simultaneously filtered
with AF. In this case the efficiency of multitexturing
is equal to zero (i.e. multipass rendering performance
is equal to singlepass multitexturing). This was
already proved by 3DMark/SS fillrate tests. I've also
created my own test D3D application to ensure that is'
not the application's fault. I just rendered two
isotropic textures in both single and multitexturing
modes. The performance reduces significally (equal to
multipass rendering) when AF is used for both textures,
but it is MUCH higher when I disable AF for the second
texturing stage. Furthermore, the GPU must _not_ use AF
for isotropic textures at all, but NV25 do it (NV20
don't). I'm afraid that it's just a hardware feature
(bug?) of this GPU and it will never be fixed by NVIDIA
in new drivers. Don't tell me that it will be fixed
coz NV already fixed OpenGL driver. They didn't fixed
anything, AF was just optimized for all NVIDIA GPUs (and
it's very said that they did it after discovering the
problem only). The easiest way to work around this
problem (NVIDIA, listen ) is to allow us to force AF on
per-stage level. There is no need to use AF for all
texturing stages (or texture units in OGL therminology)
in the most of currently existing games. Why to filter
secondary textures (usually light/reflection maps and so
on)?! It's just a waste of fillrate, the difference in
IQ will be hardly noticable. I assume that they
already did such optimization in the OpenGL 28.90+, and
there are some facts that confirm it: 1. Both
anisotropy modes (slow/quality and fast) give the same
performance in singletexturing mode, fast mode gives
boost in conjunction with multitexturing only. 2. IQ
differences in Q3 are noticable (but it was very
difficult to find it ) on the lightmaps only. 3. It
looks like they've experimented with D3D driver too - it
contains encrypted D3D registry entry, which allows to
limit the maximum degree of anisotropy for specified
texturing stages for NV25 only. All this stuff
allows me to assume that in fast mode they simplify
filtering algorithm (disable anisotropy at all?) on some
textures (far from camra?), filtered by non-primary
texturing unit. | | |
| Back
to top |
|
 |
Blighter

Joined: 23 Jun 2002 Posts: 57 Location:
Helsinki, Finland
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
Gah. Sorry Ail
Thank the heavens that I got NV15
Crap I'm just kidding, I want a new
card. _________________ Blighter | |
| Back
to top |
|
 |
Ailuros

Joined: 21 Jun
2002 Posts: 49
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
ROFL
Unwinder is
guestimating. What he misses is that in many occassions the
drivers seem to simply shut multitexturing off, where you
don't notice a performance difference.
NV15?
Look at it like that:
q3a/demo
001/1024x768x32/bilinear
4xOGMS + texture sharpening
LOD bias at 0.8 64-tap performance aniso:
82.3fps | |
| Back
to top |
|
 |
EvilEngine GODMODE


Joined: 27 May
2002 Posts: 153
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
Hey AIL have you tried
using the GF2 compatibility mode for supersampling aa? I have
no idea if it works for GF4s or not. If not this is what
you do to do it:
 | |
| Back
to top |
|
 |
Ailuros

Joined: 21 Jun
2002 Posts: 49
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
Iīve seen it, and intended
to just give it a whirl to see how it performs. If I wanted
Supersampling I would have gotten a 8500
The tab is
not greyed out so it should be working. In fact maybe I should
try TNT compatibility too hahahaha
One
question: on default the drivers use 5mb of system textures
for PCI in openGL and 47mb for D3D. Should I set them lower
and if yes WHY?  | |
| Back
to top |
|
 |
EvilEngine GODMODE


Joined: 27 May
2002 Posts: 153
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
| Ailuros
wrote: |
Iīve seen it, and intended to just give
it a whirl to see how it performs. If I wanted
Supersampling I would have gotten a 8500
The tab is not greyed out so it should be
working. In fact maybe I should try TNT compatibility
too hahahaha
One
question: on default the drivers use 5mb of system
textures for PCI in openGL and 47mb for D3D. Should I
set them lower and if yes WHY?  | Depends on who you ask. Technically it
shouldnt have any positive or negative effects on a AGP card,
but some people still set it to "0"(as do I) Thats seems
to be more of a placebo effect, and I have honestly never seen
any changes that I can tell that come from changing this
setting. So leave alone if you want. Another setting like this
is the "use fast linear-mipmap-linear filtering" , and I
really have never seen a visual or a fps change from that
either. | |
| Back
to top |
|
 |
EvilEngine GODMODE


Joined: 27 May
2002 Posts: 153
|
Posted: Thu Jun 27,
2002 Post subject:
|
 |
|
|
The supersampling is still
handy to have since our vgas are flexable enough to do at
least two types of AA(not counting quin) Early games that
use lots of unoptimized alphas like MDK2, which really look
sweet when using the compatibility mode. Fps still are well
above 100+fps even with 2x+2x(maxed) and all the eye
candy. | |
| Back
to top |
|
 |
Ailuros

Joined: 21 Jun
2002 Posts: 49
|
Posted: Fri Jun 28,
2002 Post subject:
|
 |
|
|
Thanks EE.
All FSAA
modes are available on the NV25 including 4-9tap, which
admittedly combined with any degree of aniso is quite slow,
unless you drop resolution which I refuse to heh.
Between 2x/4x MSAA and Quin/4-9tap, I of course prefer
the former two modes by far, yet the latter are not completely
useless for all occassions. Quincunx does not blur as much as
it did in the past, but the blur is still there. Combined with
a balanced negated LOD setting though itīs not as bad as
everyone makes it out to be. People are probably just too
scared to admit it because of the criticism it received and
the post filtering problem that doesnīt allow to reproduce
accurate IQ in screenshots.
As far as I can tell the
bluring/clarity on screenshots differences appear mostly on
text and hardly on the scenery. Although SS:SE isnīt a game I
would play with quin on, notice how the framecounter or the
red number in the center are correctly reproduced, in contrary
to the statistical numbers on the right side (by the way
thereīs no LOD negated for that one):
The
output of the scenery looks identical to me as in in game
conditions. Oh yeah my monitor sucks as someone else would say
*wink wink*  | |
| Back
to top |
|
 |
EvilEngine GODMODE


Joined: 27 May
2002 Posts: 153
|
Posted: Tue Jul 02,
2002 Post subject:
|
 |
|
|
I see we have LOTS of
guests that have visited from B3D. Come on guys and gals your
welcomed to register, and post too
Anyways,
glad all of you stopped by to check this place out, now I have
to get my admin to pump up the bandwidth limits for this
domain LMAO. | |
| Back
to top |
|
 |
|
|