Why not add GPU sockets AND video-memory banks to the mainboard?

[MadRat]

It would be easier to cool the videocard if it came socketed to lay flat rather than standing on its side in the AGP slot. Memory could be plugged in just like it is for main system, in specialized banks. The AGP slot just seems to be holding us back as it is not getting us to the resolutions we deserve.

 

[shadow]

madness!!!

pure MADNESS!!!!



how ya doing mad?

 

[DarkMajiq]

It's a crazy idea, but it could work, but I don't think that it will happen anytime in the near future.

 

[Peter]

With the insane clock rates that are now used on graphics boards, memory slot or socket designs have long been electrically impossible.

You'll lose quite a few MHz of RAM clock with the added line length and capacitance of such a design.

Regards, Peter

 

[ChurchOfSubgenius]

Just imagine if you had spent an extra $100 on a board with Voodoo 3 onboard last year, how pissed would you be?:| Half the fun of having a computer is changing stuff.

 

[MadRat]

Why is that? The GPU's operate at alot lower speeds than CPU's. The trace length between CPU and GPU would drop considerable, too. By standardizing the design of the GPU everyone benefits. System RAM could be used when there is no "video RAM" installed, allowing barebone systems to be ultra cheap.

 

[ChurchOfSubgenius]

nevermind, double post, see signature for explaination.

 

[shadow]

system ram ain't fast enough for todays vid chips

 

[DaddyG]

Madrat, I think that Intels Hub based Architecture could be easily adopted to accomodate a GPU hub. You would need a controller to handle the HUB interface and the GPU, but the high bandwidth of this architecture would make it ideal for your suggestion.

 

[MadRat]

SDRAM is fast enough for 800x600. How much larger resolution does a consumer grade system need? The basic system would use the mainboard's SDRAM or DDR RAM. The more advanced user may want to upgrade to a $200 module of 5gb/sec 64mb RAM.

 

[Peter]

Now big surprise, guess what approach the GPU and chipset makers followed many years ago?

Yes, exactly what you suggest - standardize a method of having an onboard graphics controller with no RAM, and give it access to system RAM. The method is called UMA, and we all know it sucked.

Then they pulled the graphics engine into the chipset to give it DIRECT and FAST access to system RAM. We're currently enjoying the third generation of that approach. By the way, some of these chipsets, namely those by SiS, allow giving the graphics unit some local RAM.

Regards, Peter

 

[DaddyG]

Peter, just because UMA sucked, it doesn't meant that a Hub based GPU would suck. Its called progress !!