> "It was fairly impressive when it came out but its specs are grossly
> exaggerated."
>
> I disagree again. It's specs like polygon count were downplayed. unlike every
> single PC 3D accelerator chip maker of the 1990s, who totally stretched and
> exaggerated their specs.

Nobody downplays their specs. Performance specs are typically very simplistic calculations based on available hardware resources that don't take into account the interplay of different components over time. Look at processor IPC numbers for a good example.

> Pro-1000s, but one Pro-1000 with 2x geometry and 2x rasterizer, same difference.

My understanding is that the Model 3 is a scaled down Pro-1000. Real Pro-1Ks were large standalone boxes and packed more punch than Model 3.

> Dreamcast arrived with comparable performance in 1998-1999. and even then,
> Dreamcast did not completely rival Model 3 in every way.

The Model 3 doesn't rival the DC. I've never seen the Model 3 do graphics that were better than the DC's.

> It takes a Gamecube or an Xbox to rival & surpass the quality and performance of
> MODEL 3, in practice.

No, PCs before the current generation of consoles were already far ahead of the Model 3.

> could rival Model 3. the entire TNT/TNT2 line lacked geometry processors,

When the GeForce came around, hardware T&L was available on consumer level cards.

My Supermodel dev machine is a mere 500MHz Pentium III with a GeForce II and my card is far better than the Model 3. It would be capable of tossing that geometry around without many problems I suspect if it wasn't for the Model 3's unfriendliness towards modern APIs like GL and D3D.

Supermodel's graphics problems are due to the fact that state changes (enabling/disabling textures, changing blending modes, toggling lighting, etc.) are very expensive and Model 3 specifies state information on a per-polygon basis. The usual programming model is to set up various state attributes, render a big chunk of polygons, then change states when absolutely necessary. I'll bet most hardware has per-polygon attributes but this isn't accessible from any API.

> I didn't say per scene / per frame, I said per SECOND.
>
> "Try a few thousand."
>
> yeah, per scene / per frame. huge difference.
>
> "I remember Ville once hooked up a polygon counter to his D3D engine in
> Supermodel and in Sega Rally 2 there were around 9K polys per frame."
>
> 9,000 polygons * 60 frames per second - 540,000 polygons per second. which is in
> line with Model 3's 1,000,000 polygons. Sega Rally 2 does not hit the peak
> performance, but does more than a few thousand polygons/sec, obviously :)

Note that Sega Rally 2 was Step 2.X Model 3 hardware which had an upgraded R3D GPU and a much faster CPU. The 9K figure is quoted off the top of my head and is an upper limit. The actual figure may be anywhere from 3K-9K. I'm curious now so when I get home later tonight I might hook up a polygon counter to Supermodel.

> but Model 3 had that fast-access VROM that DC did not have.
> Model 3 games have lower res texture, but usually more of them. more variety.
> Model 3 games look "texture heavy" and maintain 60fps.

A cursory glance of the DC specs reveals that both systems have an equal amount of memory for textures.

> the quality of Model 3 graphics is very high even though the amount of geometry
> is small by PS2 and even DC standards.

What exactly does "high quality" mean if the geometry count is lower, the graphics pipeline was far less flexible, various effects were simply impossible (such as anything requiring multi-pass rendering), and the resolution was lower?

> but every Model 3 game ran at 60fps for

Console and PC games since have been able to maintain 60FPS or higher while looking better. Some games toss around lots of geometry or effects or just aren't written well.