🚀 go-pugleaf

Obviously UNL turns out idiots for college grads.

Daniel Chee <cchee2@bigred.unl.edu> writes:

> The V4 does do Multitexturing although it's done in a shit ass
> way. 

Umm. It has 2 full texture units per VSA-100. Thats like the TNT2.

> Each processor chip has it's own memory space and loads the same
> texture. Then there's the frame buffer. So in reality, you don't get
> the whole 32MB to load textures. Say 8 MBs go to the FB, then you've
> got 24MB left for texture right? Wrong! That's 24MB divided by the
> number of processors. If there's 4, then 6MBs each and they all load
> the same texture! Stupid ain't it? There's nothing new in the V4,

You are obviously an idiot. See other posts to see how the Voodoo4
uses texture memory efficiently even while duplication is
unfortunately necessary for higher memory bandwidth. 

Firstly, the greater the number of VSA-100 chips, the less memory used
by the framebuffer since its split among each cip. 

Secondly, each VSA-100 can texture upto 128 scanlines at a time thus
encompassing some polygons completely (in their scanline block). This
allows for less texture duplication as the other VSA-100 does not need
to load those textures at all. So in a high polygon game, the Voodoo5
architecture utilizes its memory more efficiently if you set your
rendering blocks to be large enough (which is probably driver
adjustable).

Also, each VSA chip gets its own 32 MB of texture/framebuffer space
except for the V5 5000 board. This is as much memory as any other
gamer board available (more in effect because of framebuffer size
getting smaller as you add more chips).

> it's basically a V3 with support for what in the competitors
> products, 32bit color, etc.... In plain simple english, it's a piece
> of shit.

In plain english you're completely clueless.


> Dave wrote:
> 
> > I am pretty underwhelmed by the specs of the V4/v5. With 14 million
> > transistors per chip, why are we only seeing v3 speeds? Where have
> > they spent those transistors. I am no 3d graphic whizz but how many
> > transistors would it have cost to add a second texture unit to the
> > chip? This would have doubled the figures 3dfx zealots could throw

There are 2 texture and 2 pixel pipelines per VSA-100. 2 VSA-100's
give you 4 texture/pixel pipelines per board. I am not sure why the
increase from 9 to 14 million when all they added was:
- 1 pixel pipeline
- enhanced texture pipeline
- 32 bit rendering
- 32 MB support
- 2048x2048 textures
- 32 bit texture support
- AGP 4x + sidebanding
- caches? (this is possibly the biggest reason for the transistor
  count increase but I'm just speculating)
- asynchronous core/memory clocks? (speculation)

> > around the newsgroups and provided a handy boost to any game utilising
> > multitexture, which is any decent game these days. Surely 3dfx would
> > be better off even ditching MPEG2 support and using those transistors
> > saved for the second TMU (I've no idea how many transistors MPEG2
> > suppors takes), anyone buying a v4/v5 would likely have a p400+ class
> > CPU anyways and would a hardcore gamer care? Imagine the hype if the
> > V5 boasted a 2.9gigatexel/sec fill rate?
> > Comments??

MPEG support probably does'nt take up too many transistors and is
necessary to get any OEM sales. A 2.9 GPixel/s fillrate is limited
more by memory bandwidth than anything else. Of course i wish 3dfx had
their core on 0.18/0.22u like the Savage2k and clocked at 200 Mhz with
200 Mhz memory. (mmm... drool...)

Cheers,
salim