
We've covered
many current 3D Graphics Processors in the graph below, in
recent articles. However, we modified it here a bit,
to show you only the top end products available on the
market, leaving out the true mainstream or value oriented
cards.
 |
GPUs and VPUs Compared |
Fillrate and Bandwidth = Frame Rate |
|

There has been a
fair amount of controversy over whether or not the GeForce
FX (NV30 or NV35) can be called an "8 pipe" GPU. We
asked NVIDIA directly, with some very specific questions
regarding how their programmable graphics processor handles
things. Here's what we were told...
"If a
game is mult-textured both ATI and NVIDIA will run at 8/#
textured pixels per clock.
So...
Clock Speed Single Texture + Z
Dual Texture + Z Z/Stencil Only
Radeon 9800 Pro
380
3040
3040
3040
GeForce FX 5900 Ultra 450
1800
3600
3600
Basically, only frame buffer blending is done at 4 pixels
per clock, everything else is done at 8 ppc.
The main problem is the old terminology does not really
apply to the new programmable pipelines. GeForce FX
5800 and 5900 GPUs run at 8 pixels per clock for all of the
following:
a) z-rendering
b) stencil operations
c) texture operations
d) shader operations
For advanced applications (such as Doom3) *most* of the time
is spent in these modes because of the advanced shadowing
techniques that use shadow buffers, stencil testing and
next-generation shaders that are longer and therefore make
the apps "shading-bound" rather than "color fill-rate"
bound.
Only color+Z rendering is done at 4 pixels per clock, all
other modes (z, stencil, texture, shading) run at 8 pixels
per clock.
The more advanced the application, the less percentage of
total rendering is color, because more time is spent
texturing, shading and doing advanced shadowing/lighting.
When measuring in games that do not use advanced rendering
techniques, is it more important how many pixels per clock
you can do, or how fast the game performs?
Look at Serious Sam. We
win at that game and it uses a lot of color + z
rendering. If we can do the same amount of work with a
4x2 set-up as they do with an 8x1 set-up, then who should
the "shame" be targeted at? Shame on us for only
having a 4x2 pipeline or shame on them for getting beat by a
GPU with "half as many pipes"?
Clear as mud eh?
Well, suffice it to say that the NV35 sometimes acts as an 8
pipe GPU and sometimes a 4 pipe. The bottom line is
this, unless you are concerned about the technical side of
things, it's all about frame rate folks. So there's
not a lot of sense in burning up brain cells on this
mystery. The benchmarks that will follow here will
speak for themselves.
 |
A
New Family Of High End GPUs |
NV35 - The Real GeForce FX |
|
Here are a
couple of quick snaps from NVIDIA's NV35 / GeForce FX 5900
presentation, which cover positioning and future iterations
of board level products.

As you can see,
NV35 or the GFFX 5900 Ultra is now the flagship, with the FX
5600 and 5200s following up in the mainstream segments.
There is no sign of the NV30 or GFFX 5800 Ultra here.
It's clear NVIDIA is aiming to completely put the NV30 out
to pasture and to think, it was only in its infancy.

The real good
news is that there will be a couple of other flavors of the
NV35, with much more manageable price points.
Personally, we think the days of a $500 Graphics Card are
over but that's the ESP (estimated street price) of the
GeForce FX 5900 Ultra. We're hopeful that the "street"
will beat down the price a bit more on the 256MB variant
we've tested for you today. However, users will have
choices and the GeForce FX 5900, a 128MB version of the NV35
(core and memory clock speeds not noted as of yet), appears
to go head to head with the Radeon 9800 Pro's price point.
The real interesting card here, in our opinion, is the "NV35
Value". Depending on clock speeds and memory
components used, this could be the "enthusiast mainstream"
card of choice, along with some earnest overclocking.
Time will tell but NVIDIA is still obviously getting a
handle on all of the SKUs for retail.
 |
NVIDIA's UltraShadow |
Hardware Acceleration Engine For Next Generation
Shadow Effects |
|
Although the
NV35 shares much of the same feature set of the NV30, in
addition to its beefed up memory bus width, it does also
come with some extended capabilities. Most notably,
NVIDIA is now evangelizing their new "UltraShadow"
technology within the NV35's core. This is a hardware
assist for shadow volume processing. Think of it as a
sort of "Z-Cull for shadows".

Now then, as you
can see in the above diagram, there are certain areas of a
given scene, where rendering of shadow effects are required
to present the proper perspective, look and feel, to the
viewer. Then, very much like hidden surface removal or
"Z-Cull", there are areas or "clip planes" that can be
defined, where the effect is not needed as much, since it
will not be technically viewable to the user.
Depending on the light source position and volume, shadows
will be drawn or limited according to the pre-set conditions
of the programmer. Which brings us to where the rubber meets
the road with NVIDIA's new "UltraShadow" technology.
This is a feature that Game Devs must implement in their
engine, in order to take advantage of it. It's not an
"out of the box" feature. This is something that
obviously could take time to deploy, in next generation game
development and it is all dependant upon adoption in the
community. Having said this, like hidden surface
removal, it seems like a logical progression as complex
lighting and shadowing requirements scale, with next
generation titles like Doom3.
Tech Demos, Drivers and Image Quality
|