Why have there been no VSA-like architectures?

eddieck

Gawd
Joined
Dec 13, 2009
Messages
1,010
The VSA (from 3dfx for the 2 of you that don't remember) provided some amazing scaling, better than SLI and CF from what I understand/remember. Why haven't we seen any more architectures like this?
 
Well, perhaps because VSA sucked. You might recall that 3dfx folded after that. It was not at all a successful venture.

However another part of the reason is likely that cards now do geometry, as well as pixels. Back in the old days all they did was process pixels. All the transformation and lighting was done on the CPU, they just handled texturing and rasterization more or less. That sort of thing is easy to break down in to arbitrary levels. Geometry is not so easy to break down, and current cards deal with it.

Also I think you'll discover that modern chips do scale in most of the same ways the VSA chips did, it just isn't useful to do so. Things that VSA was good at that modern cards are also good at:

1) Trilinear/AnIso filtering scaling. Knock together a bunch of VSA chips and you could get much better filtering than normal. Nice, but not needed now. A single modern chip handles 16x AnIso with no real speed hit.

2) Better FSAA. This is suppose is kind of useful, but not enough for most people to pay tons for it. You discover it works well though, in particular with SSAA. Take 4 cards and they'll render 4x SSAA at about the same speed 1 card would render no FSAA. You basically just offset each of them to handle a different part of the sub pixel.

3) Higher resolution. This is something the VSA targeted since it was designed for simulators. Again, we find modern chips scale well in that regard. AT low resolutions, SLI isn't so impressive. Scale up the rez, it scales up nicely. However again this is not such a big deal as we find single cards driving 1920x1200 and even 2560x1600 with ease.

Regardless, it isn't easy or cheap to make a multi-chip solution that scales real well. If it were, nVidia would probably do it as they bought 3dfx. Things are just more complex now.

Also designing dynamite multi-chip solutions isn't as good an idea as doing faster single chip solutions. More chips means more expense. It is cheaper to do 2*X on one chip than X on two chips.
 
VESA?

Disputed claim:
VESA has been criticized for their policy of charging for published standards. VESA charges non-members for some, but not all, of their published standards. Some people believe the practice of charging for specifications has undermined the purpose of the VESA organization.
 

No Vsa:

vsaslot.jpg


The predecessor of AGP.

Sycraft pretty much nailed it.
 
No Vsa:

vsaslot.jpg


The predecessor of AGP.

Sycraft pretty much nailed it.

Aren't you funny :D

http://en.wikipedia.org/wiki/Voodoo_5

3dfx built VSA-100 because they were a failure as an integration company, and rather than wait on Rampage, they released a hackneyed chip that was the VSA-100.

And although the AA performance (and quality) was impressive for the time, today's chips can do exactly the same thing with no tricks required.
 
Back
Top