Rebirth of the daughterboard - PCI-Express' lovechild

Sometimes, when the powers that be are feeling nice, an idea, usually accompanied by lightning and pain, forces its way into the brain. After much rubbing of temples and murmurings of ‘ow’, many contemplate the gift that has been delivered to their

Sometimes, when the powers that be are feeling nice, an idea, usually accompanied by lightning and pain, forces its way into the brain. After much rubbing of temples and murmurings of 'ow', many contemplate the gift that has been delivered to their pumping neurons. Neurons probably shouldn't pump, but pump they do, and you should feel enlightened to have brain cells doing something they shouldn't. Naughty neurons.

This is exactly what happened to me one day.
So, here is the pitch. Why not have video cards with removable RAM and graphics chips – just like motherboards. This, of course, was possible with older video cards, like the S3 Trio. Obviously it wasn't too successful back then, but surely there is a market for it now, considering how much cards cost these days and the similarities in chipset series. Cue laughter – or stunned, amazed silence, in which case you are a God to me. For the time being, let us call these new 'mini-mobos' daughterboards.

If you think you've heard the word daughterboard before, you'd be correct. The word is attributed to old sound cards – you could attach a wavetable board to your existing card to expand the amount of sounds the card could produce. We can be flexible, so let us extend this term to include boards with removable chips, rather than circuit boards. If we can do it with sound cards and motherboards, why can't we do it with video cards? Makes sense in a way, doesn't it. We spend the odd $500, or more, on a new video card every 4-6 months. So, cut out the middleman -- the third-party who buys the chips and sticks them on a board -- and be done with it. Let them make special, revamped boards, in differing styles. Like they are now, with silver PCBs and glamorous, but over-the-top names.

Your TSOP or BGA DDR RAM is fine (for now) - you just want the latest chipset. As long as the models are the same (ie. GeForce4 Ti, GeForce2) it should be possible. Think about it. When you buy a new video card, you not only buy the new chip and RAM, you also buy a another chip for TV-in/out and another heatsink – not to mention all the additional circuits, capacitors and ports. You don't buy a new motherboard every time you buy a new CPU (well, if you have an Intel motherboard, then maybe you do) or want extra RAM, so why should you with a video card?

At this point, you may be asking 'Someone has to have thought of this before.' You are probably right and, maybe, it has been discussed over and over again. In the interests of saving this article from an early, flaming demise, thoughts should be aimed at this not being the case. If you still think it's dumb, allow yourself to politely fall asleep.
Fight this urge though. Please.

The likes of NVIDIA and ATI are easily comparable to Intel and AMD, with companies such as SiS (who have made a foray into the 3D card market with its Xabre graphics chip) easily identifiable as VIA with its C3 processor. A development such as this would open up possibilities for smaller manufacturers to focus on producing killer chips and not on securing companies to stick their hardware onto boards.

There are a few things that could quite possible stop this from becoming reality, other than NVIDIA/ATI's unwillingness to move to such a model. Technological and design-related problems would most certainly arise. Pin arrangements and memory incompatibilities would have to be considered, as memory speeds just keep getting faster and RAM packaging and design is always undergoing change. Not to mention the inclusion of new features, such as hardware T&L. However, it is hard not to fall back on 'if it works with CPUs, it can work with GPUs'. Considering that everything is integrated on-die, ie. there is no controller chip on the board to worry about (such as the Northbridge or Southbridge found on a motherboard) it should be a simpler task to achieve than it is with CPUs/motherboards. Unless there is a major change in pin arrangement or features, you could upgrade you GeForce4 MX to a Ti by simply purchasing a new chip and hence gain those valuable pixel shaders – the ones John Carmack loves. As with anything computer related, however, we would have to upgrade eventually.

A development such as this would allow the usage of new technologies, and a chance to rethink the whole ZIF socket – assuming that this would be the system used to swap-out graphics chips. The main concerns I have with the current socket system are two-fold. One, it is still very easy to bend pins and render a perfectly good CPU useless. Two, a new socket design would allow companies to rethink cooling methods.
Personally, the socket system used with the old 286/386 CPU would be good – no pins to bend, you just click the chip into place. You can go here to see for yourself:
http://oldenburger.linuxtage.de/Oldenburg1996/picture-16.html
Look for a tiny gold chip in the bottom left corner of the motherboard. That's the socket, with a CPU installed.

Some thought would have to go into an easy way of removing the chip, as the old way involved a flat head screwdriver being inserted into the groove at the top left corner of the socket. It worked the same way for the old S3 Trio, not to mention a swag of other ancient display cards.
This would be an evolution, not an innovation. I believe video cards have developed enough for this to be taken seriously. We should slow down our race upwards; stop and look around; and make the platform we are on stable for those who follow behind. If the race continues, we will simply run into the same problems that plague the Wintel platform – bandwidth (IDE), conflicting standards (DDR, Rambus) and slow advancement into fresh technologies (32- to 64-bit). Sure, things like IDE are being replaced by Serial ATA, but how long has it taken for this to occur, and how far does it depart from traditional methods. We still have the master/slave restriction, and you can only have as many drives as you have ports – something that places SCSI above ATA. It's game over for S-ATA in this regard -- a sixteen to one victory, albeit a small one.

This idea of sub-component upgradability, or SCUB for short -- a little word born of this article -- doesn't have to be restricted to video cards. It can work for anything. The other computer subsystem that would see the benefits would be storage. Why buy a new hard drive – just buy an expansion card and add another five or ten gigabytes to your system. With the change over to flash memory, something that is bound to happen as the technology matures, and the introduction of PCI-Express, will create the necessary springboard or, at the very least, a wooden plank on a tire to propel SCUB forward. And no one said it would be easy. The plank may break.

If you've been inspired in some way, then cool. Awesome actually. But please don't attempt to pry the GPU off your video card. For now, you'll just end up with a scratched and pitiful piece of silicon and tearful eyes.

Logan Booker