She 7 or 17?
Is speed everything for
some video card zealots?
Many of us
have heard repeatedly that the Voodoo3 is the fastest card on
the market for gamers today. If a Voodoo3 and a TNT2/ATI Rage
128/TNT/G400MAX/S4 were all clocked the same for the memory and
core, I would claim that the Voodoo3 is indeed faster than all
of them in many situations. The
real issue, however, is how much faster the Voodoo3 is
than the other cards. This is something that many 3dfx "zealots"
do not want to consider at all. They are so emotionally attached
to their beloved graphics card that higher brain functions seem
to have stopped. Does the negligable speed advantage the Voodoo3
have justify it's lack of other features current video cards support
that the Voodoo3 lacks?
Let me summarize
what I have said so far for the benefit of the 3dfx zealots. I
stated clearly that the Voodoo3 is faster than the TNT2, ATI Rage
128, and the other cards. The issue is how much faster is the
Voodoo3. This question is extremely important when one actually
considers that the TNT2, ATI Rage 128, Savage4, the currently-unavailable
G400MAX, Neon 250, and others offer more features than the Voodoo3
including up to 32MB of on-board memory, up to 2048x2048 texture
sizes, 32-bit rendering, true and working AGP support, etc. The
outdated i740 and the Riva 128ZX even support features that the
Voodoo3 do not. But let us focus on just the popular V3 versus
benchmarks that I have seen, the Voodoo3 is usually faster than
the TNT2 by about five to 15 frames per second in OpenGL Quake
II benchmarks (e.g. demo1.dm2 and others). In D3D games, the results
seem to indicate a tie, with Voodoo3 being faster in some resolutions
on some CPUs while the TNT2 is faster in some resolutions on other
CPUs. With results like that, how can anyone claim a clear and
distinct winner? Some might say that the TNT2 has the edge in
D3D games but I do not really see that; it's really just a virtual
OpenGL scores, however, are quite different. Not counting crusher.dm2
and mon2.dm2 (both in 16-bit), it seems consistent that the Voodoo3
provides higher frame rates than the TNT2. The bottom line is
that the V3 is the Queen of Speed (or to be more accurate, it
is consistently faster than the competition) in OpenGL games.
point, you're probably asking yourself, "Okay, you already said
that the V3 is faster. You have now hammered that idea into my
head. So what's the point?"
Do you think
that a five to 15 FPS boost justifies buying a card that offers
considerably fewer features than the competition? That is the
point, or to be accurate, one of the points. For many people,
that potential 15-FPS boost is all the justification they need
to purchase a card with less features but faster performance.
For these people, they do not care that the slower card may be
providing "speeds exceeding 60 frames per second". They want that
extra 15 FPS and the unnoticable performance increase that comes
issue - the most important issue - is to determine if that 15
FPS boost is real or "synthetic". As we all know, the miniGL is
used for all of the Voodoo3 benchmarks in OpenGL games. The reason
for this is that the Voodoo3 is not a fully compliant OpenGL card.
Does using the miniGL create an illusion of superior performance
in OpenGL games? Let
me give you a hypothetical example using the Kingpin Alpha demo.
For the sake of argument, let us pretend that at 1600x1200x16,
the Voodoo3 performance with fog enabled is 80 frames per second.
The TNT2 at the same resolution and with fog enabled is 63 frames
per second. So clearly the V3 is faster than the TNT2 by approximately
17 FPS per second in this hypothetical example.
happens when you disable fog? The Voodoo3 performance with fog
disabled is still 80 frames per second. The TNT2 with fog disabled
is now up around 73 FPS. The V3's incredible performance advantage
mysteriously shrinks to just 7 FPS per second. Why? The
answer to the question is that the V3's miniGL either does not
support OpenGL fog or it is only extremely partially supported.
Therefore, enabling fog has no impact on its performance. As far
as the V3 is concerned, fog is always disabled. When you see Kingpin
benchmarks, which situation do you think is being reported? Are
people reporting benchmarks with fog enabled or are they reporting
benchmarks with fog disabled? Would it surprise anyone that fog
enabled benchmarks are the most common? Kingpin
is a good software to use for this simple test because we can
enable/disable fog at the menu. Otherwise, I would need to know
all those console commands to figure out what to enable/disable.
In short, Kingpin made it easy. I probably would not have noticed
the performance discrepancy of fog if it was not for the menu.