I'm really glad that you asked that question. Single-chip VSA-100
boards, the Voodoo4 line, will enable significantly faster and
better-looking content than the Voodoo3. I just mentioned the key
performance enhancements but let me explain more about their impact.
20% improved raster efficiency is pretty easy to understand: on
average it's 20% faster in 3D rendering with all other factors held
constant. Rendering two pixels per clock as opposed to one for
Voodoo3 gives another 50% to 60% real performance improvement on
single-texture apps. and scenes. I said that texture compression
took some time to implement, well one of the reasons is that we can
store textures in compressed format in not only the frame buffer but
also in the texture cache. That effectively increases the size of
the texture cache which offers a nice performance boost.
Applications that use either Microsoft DXTC or our FXT1 compressed
textures will see another 30% to 40% performance improvement. This
adds up to a 100% potential performance improvement even before we
talk about the new texture combines and alpha blends and the
OK so you've got the Fill Rate contest won. What about image
quality? We are hearing so much about "cinematic effects",
how are the new scene enhancements like motion blur, and depth of
field coming along? Are they going to add new level of realism or
are we not quite there yet? Will game developers be able to
implement these new effect easily?
Remember that full-scene AA is the single most important feature
that can be added to improve 3D image quality. No one should be a
victim of bad graphics with jaggies and scintillating pixels. The
other cinematic effects will be introduced in titles at different
times. While I can't say for sure, motion blur is likely to be the
first feature that developers use because it's artistically
intuitive and programmatically straightforward. We've been demo'ing
modified Quake3 with motion blur for a while now: making fast-moving
characters blur is a very nice touch. I want to be clear that
there's no announced commitment from Id to add motion blur to
Quake3. Soft shadows and reflections might be next because again
they are artistically intuitive and the code changes are not that
difficult. It's pretty obvious when you need to have a realistic
shadow or reflection in a scene. The feature that I'm personally
most excited about is depth of field blur or focal blur. It takes a
real sense of cinematography to use it effectively but when employed
correctly the results can be stunning. After coding the T-Buffer
demos our in-house demo developers said that they wanted to go to
film school to learn how to better use the effect! One truth since
the beginning of game development, though, is that you can never
predict exactly how a developer is going to use your hardware
features. These people are so creative that they'll code incredible
things that we'd never think of. The core philosophy of the T-Buffer
then is that by putting powerful tools in the hands of creative
developers, amazing things will happen.
image for larger view)
FXT1 Texture Compression - Did you need to change the hardware to
Yes, definitely. FXT1 and DXTC took a lot of time to code.
3dfx has admitted that the VSA-100 is based on the a same
architecture as Voodoo1? How do you keep milking it for more power?
I guess it's the same architecture in the same sense that a Model-A
and a Z8 both have four wheels and an internal combustion engine.
We've been able to successfully refine and optimize over many
generations of products.
Is it unusual for a graphics company to continue with products based
on the same basic architecture for 5 years?
When we set out to design a chip we first ask, "What is it that
the end users will really need, and how do we give them more than
they expect?" The architecture that's chosen is derived from
the features and performance required. We've been able to meet the
users needs for a long time with the same core architecture by
adding the right features at the right time. That's a testament to
the foresight in the original design, that it could be radically
upgraded over a long period of time while maintaining a large degree
of software compatibility. When Voodoo5 hits the market next year it
will have been about four years since Voodoo Graphics. If you look
back at the history of graphics accelerators a four year run is not
that unusual. Everyone knows that we've been working on a new
architecture for a while, but that's a different subject.
If it is the same architecture, aren't you just perfecting the
Voodoo1, or just adapting it to changing market demand?
Again, there were one million transistors in Voodoo Graphics and
there are fourteen million in VSA-100. Whether it's the same
architecture or not probably isn't relevant. The real question is
whether Voodoo4 and Voodoo5 are going to deliver features and
performance that make the end users incredibly happy. There's no
question, they will.
What do you have to do physically to a chip to add new features?
The chip first exists as a design data base. Many of the components
of that data base are literally software programs that describe the
functions of different parts of the chip. When we go back to add new
features and upgrade existing features it's a lot like writing new
programs and re-writing old ones. There are specialized
computer-aided design (CAD) tools that enable us to do convert or
"compile" the programs into a series of logic gates. Once
we're sure that all of the features are properly working we use
another set of CAD tools to create the physical description of the
chip, literally the drawing of every transistor and connection.
How different does the die of VSA-100 look physically to the
Voodoo1? How about the Voodoo3?
We start the physical design from scratch for every product so aside
from the fact that all three are chips, or two chips in the case of
Voodoo Graphics, they look totally different.
Well guys, I think that just about sums it up. In short, this isn't
your Daddy's ol' Voodoo, is it? Are we still on track for an April
Thanks a lot for your time, Dave, we really appreciate it. We're
still on track for shipment in the April timeframe as we said at
Comdex. We'll have news about stuff like pre-orders after the
Thanks for your time as well gents! We will be waiting patiently?
(yeah, right) :-)