nVidia NV40, NV41, NV45 Information

http://babelfish.altavista.com/babel...%2F01-27_a.php


quote:

nVidia NV40, NV41, NV45 & CO. Information
27. January 2004/of aths & Leonidas / page 1 of 2



Around the NEXT gene diagram chip NV40 of nVidia so some rumor climbs,
about which however to date naturally purely nothing at all was not
officially confirmed. Unfortunately make also speculations round,
which were raised into the status of a rumor - although a statement
was to be in this connection never heard, not even unofficially. We
want to summarize today here our past knowledge over nVidia the NV40,
peppered also with a set of fire-new information, which was not to be
read so far yet anywhere.

To the classification nVidia of the NV40 of project: The zero in the
code name point to a completely new architecture concerning the
feature pallet, like that as this so far actually always with nVidia
the case were (NV10 = GeForce1 with DirectX7-Featurepalette, NV20 =
GeForce3 with DirectX8-Featurepalette, NV30 = GeForceFX with DirectX9
Shader 2,0 feature pallet). The NV40-Chip will be further DirectX 9,0
(and not DirectX 9,1), but however the Shader abilities on stage 3,0
to raise.

In accordance with actual nVidia planning is each year such an new
architecture into the market to come, as well as all half years an
appropriate Refresh chip. However nVidia so far actually only with
Riva TNT 1/2 and GeForce 1/2 could adhere to this planning, the
GeForce3-Chip and only right the GeForceFX chip was already late
substantially, which had then naturally also appropriate effects on
the following projects and has. By means of the delay of the GeForceFX
the NV40-Chip was planned in the long run for the autumn 2003.

This planning could be as well known just as little kept, why nVidia
also in the autumn with the NV38-Chip (GeForceFX 5950 Ultra) second,
actually not planned Refresh of the original NV30-Chips (GeForceFX
5800 /Ultra) on the market threw. Over the new date of the NV40-Chips
one was at times of the NV38-Launchs not yet in the clear, in the
meantime correctly knows itself we however that nVidia either on the
CeBIT (18 until 24 March) or the Gamers Developers Conference (22
until 26 March) will present the NV40, while one can count on first
purchasable NV40-Grafikkarte direction at the end of April or Anfang
May.



And thus to the technological change of the NV40-Chips to the current
NV38-Chip: It is to be accepted as the first strongly that Pixelshader
architecture was substantially improved with the NV40, and the Shader
substantially more efficiently works than at past GeForceFX maps. This
could be reached with two measures: Enlargement temp register files to
minimize (in order "empty blisters" in the Pipe) and allocation of the
arithmetic units of Vector4 in Vector3+Skalar (register Combiner and
the R300-Shader a clear achievement plus obtains) with this method
depending upon Shader. These interferences into architecture cost
transistors, but the achievement gain would than only justify that
more. Therefore we are rather sure us in this point that nVidia with
the NV40 will make these changes in such a way.

Probably the NV40 contains in addition at least twice as many
arithmetic and logic units for the Pixelshader, when the NV38 has.
That the Pixelshader achievement substantial was increased, anyhow the
performance would mean architecture improvement and widening together
taken with Shader programs of the version 2.0 and 2.X. the Shader
version 3,0 at security of bordering probability by the NV40 will also
be supported, but we assume the use of conditioned branch statements
in the 3.0er Shader slows processing down substantially. The
background for this: So far always four pixels are together computed
(either at the same time or successively). These dependence must be
dissolved with the Pixelshader 3,0, since the branch statements in the
Shader code for each pixel can fail differently.

Probably the NV40-Architektur does not permit an indication in
"pipelines x TMUs" any longer so correctly, we assumes however that
the chip behaves with the Multitexturing like a 8x2-Architektur.
Probably however 16 Z/Stencil tests can be implemented at the same
time per clock. Those plays, which set that on a separate z-passport
(Doom III), would benefit naturally. We would like to know this
statement of 8x2 however posed under special reservation!

The NV40 become approx.. 175 million transistors awarded, which in
view of the past rate of growth with a new architecture appears
reliable. According to our opinion thereby place is enough for
additional pixel processors and Texture Mapping unit. However thereby
the new transistor budget would be probably exhausted, so that except
improved achievement and support of pixel and Vertex Shader in version
are to be expected 3.0 hardly new features. It gives in the rest of
suggestions that the NV40 was again changed during the development to
convert and test (instead of equal the planned Design). Whether that
the final NV40 the good one or bad one handed, become the comparisons
with the R420 and/or R423, which NEXT gene chips of ATi, to show to
have.

In order to increase the total output, it is not sufficient naturally
to multiply simply the arithmetic power. The traditional filling rate
remains just as important, why we speculate evenly on a 8x2
architecture with the Multitexturing Setup (with still up to 16
textures per passport). Then however also a gigantic range would be
necessary, in order to be able to use this enormous filling rate: We
in this connection nVidia quite think capable of to set on 600 MHz
DDR2-Speicher to block or still faster memory. This would stand also
well in the tradition that nVidia regularly on the fastest in each
case available memory chips sets. It would be thus also conceivable
that there is a Ultra variant of the NV40 with up to 800 MHz storing
act. Secured is in this connection however still nothing.

The TMUs will probably still ever a bilineares SAM-POLARIZE per clock
to produce to be able. Trilinear and anisotropic filtering will be
implemented probably à la GeForceFX (while S3 with the delta chrome
control trilinear filtering from a MIP map, which will in the long
term probably become generally accepted). Whether it remains with the
NV40 further with 8x AF implemented, is at present still unclear, or
nVidia also 16x AF or more.

We mean that 8x AF is à la GeForceFX in the quality mode still on the
height of the time. Interesting naturally 16x AF would be in not more
"optimierter" GeForce4-Qualitaet, in order to have the direct
comparison with the Radeon. Whether nVidia again completely trilinear
filtering will permit, is also unclear. We fear that the driver forces
the map further to filter"brilinear " although the hardware would be
to full trilinear filtering able.

A short remark on the AF: "genuine text book quality" would be
extremely aufwaendig, why nVidia (and S3 in the delta chrome) starting
from 4x AF accepts a certain angle dependence. In relation to the
strong angle dependence with ATi's AF is to be however almost
neglected. We here quite accept certain weakness in 45° angles with
high-quality AF modes. The "optimizations" à la GeForceFX will
probably be further insertable however in the NV40, an option, whose
sense us is veilful, since they destroy more quality than achievement
bring in. Anyhow one is used of nVidia that the chip is evolutionary
continued to develop.

A further point, over which there is still absolutely no information,
is the question of the anti- Aliasings. In view of the good
6x-"sparsed"-Maske, which ATi offers since the R300 (Radeon 9500/9700
), an improvement in the GeForce series would be to be recommended
urgently. We do not know to this topic that nVidia with the NV40 at
least mode will introduce a new anti- Aliasing, however unfortunately
at all, as this is to be arranged: Comparable quality to higher
performance or however a better smoothing quality.

The NV40-Chip will be marketed probably still as "GeForce". Parts of
the nVidia crew wanted a new name, but allegedly marketing was to have
raised against it objection. Thus the GeForceFX series goes beginning
with the NV40-Chip probably into a further round, to our information
further with a nummerischen appendix four digit with "6".

We summarize again, which we know at present from the NV40 (or to know
believe):

nVidia NV40
175 million transistors, in 130nm manufactured
8x2 architecture, however 16 Z/Stencil tests per clock
DirectX 9,0 architecture, supports Shader 3.0
opposite NV38 and Shader doubled more efficient pixels
supports DDR1, GDDR2 and GDDR3
internal AGPx8 interface
exact clock rates: unknown; there is estimated 500 to 600 MHz chip
clock and 600 to 800 MHz storing act
Improvements with anti- Aliasing: (at least) a new mode; its SAM
polarizing sample is however unknown
Improvements with the anisotropic filter: unknown
Presentation: GDC or CeBIT at the end of March 2004
Market entrance: At the end of April or Anfang May 2004
Sales name: GeForceFX 6XXX
If we here already once a prognosis to the NV40 to dare may: At least
under the aspect of achievement nVidia will once more even exceed
itself. In order to know, whether it is enough to offer to the strong
competition Paroli we must however probably or badly wait for the GDC
and/or the CeBIT, because also of ATI at this time with R420 and R423
new HighEnd chips are expected.

Glad I followed the link, I was wondering what language that was badly
translated from.

On Wed, 28 Jan 2004 13:57:33 -0500, "PB" wrote:

Glad I followed the link, I was wondering what language that was badly
translated from.



German, using one of the translation programs....
BTN........

John Lewis

"John Lewis" wrote in message
...
On Wed, 28 Jan 2004 13:57:33 -0500, "PB" wrote:

Glad I followed the link, I was wondering what language that was badly
translated from.



German, using one of the translation programs....
BTN........


* BLINK *
* BLINK *

Nooooo, reeeeeealy?

Gee, guess I couldn't figure that out from following the link like I said I
did.

/sarcasm off

On Wed, 28 Jan 2004 13:57:33 -0500, "PB" wrote:


German .......




Glad I followed the link, I was wondering what language that was badly
translated from.