Stepped/Progressive SLI

Was talking to Vimeous (and a rather worse for wear TiG) about this.

What if someone launched a set of PCI Express cards that offered different technologies.

ie you buy the first card, and it offers up to DX9, but another addon card offers different technology, not SLI as such, just more stuff?

So at £100 you have a DX8.1 card, and then later on you can buy a card that handles DX9 for an extra £100.

Or whatever the new subject is.

Or the add on cards take AA and AF effort onboard themselves. So if you have a 17" TFT and you cant run at 1600x1200, youmight as well have AA and AF maxed instead. So you buy the relevent card with relevent chip.

Or you buy one that is simply a load more ram....it costs you £50 and simply increases your frame buffer.

Some games would work better with that than others, and you'd read a review and go that route.

Or you buy the fastest pair you can afford, and then later on upgrade only one of them to d the new texturing just announced.


Its all me dreaming really....just a thought. Scalable add on cards.

Bloody marketing nightmare though.

It's a technical nightmare and would never happen. Nice idea, but logistically impossible for most vendors, and a technical can of worms.

CWI

Can of Worms Interleaving.....you're right....nightmare to sell to the developers

its possible, setting aside the hardware developers disadin over the idea.
gpu's could use LGA type sockets.. could use dimm slots for memory. some kind of 'northbridge' to tie the various components to the gpu.
I often wondered about using a cpu for processing graphcis seperately from the main cpu.

A lot of things are very hard to separate from the main GPU - AA/AF is one. AA is intrinsically part of the pixel pipeline you can't really do it in an addon. About the only thing you could put in addons is either more pipes (so you take your 8 pipe card and make it 16 or whatever), or more memory.

it was just me musing really. Cleverer people than me have alreay told me NOPE YOU NUMPTY, get back to cars ;)

I reckon in the future the gpu will be more general purpose and it will be possible to
programme its setup... So instead of dedicated pipelines they will all be the same and
you can assign them to be either vertex or pixel or whatever.