geforce memory limitations, how to alleviate???????

[Trone]

Supposedly nVidia's Geforce cards all suffer from the same problem ranging from the original to the Geforce 3, in that they all have memory bandwidth problems. Correct me if this is not right. No one overclocks the core of the card because it would do little, but the ram on the other had will do a lot. If you could double the ram's speed say from 500mhz to 1ghz, would that double performance? If so, why do manufacturers/nVidia do this. Instead of equipping cards with 64meg of ram at say DDR 250mhz, which usually at that amount is pretty overkill, why not run 2 sets of 32meg of ram at 250mhz, which equals 500mhz, which due to the fact that its DDR yields 1ghz??????? Is that a correct assumtion? Can that be done? I vaguely remember reading something awhile back about motherboards offering double the ram bandwidth, but you half the amount of ram. Im guessing that this technique cannot be done for some reason, otherwise some rouge manufacturer would put out a super duper geforce 3 that is twice as fast as anything out there. Any comments on this? What is limiting manufacturers from doing this right now? Thanks!

 

[Richard]

The technique you're referring to sounds very similar to 3dfx's SLI.

The only problem is that it requires two independent GPUs to go along with the RAM. In the case of the GeForce3 that would be one heck of an expensive card.

It is a nice idea though.

Another technique would be dual channel ddr, but that too would be both expensive and a monumental feat to assemble. The number of traces coming off of the GeForce3 GPU is already high. Imagine doubling it.

 

[Trone]

Richard999 (Jul 24, 2001 11:27 p.m.):
The technique you're referring to sounds very similar to 3dfx's SLI.

The only problem is that it requires two independent GPUs to go along with the RAM. In the case of the GeForce3 that would be one heck of an expensive card.

It is a nice idea though.

Another technique would be dual channel ddr, but that too would be both expensive and a monumental feat to assemble. The number of traces coming off of the GeForce3 GPU is already high. Imagine doubling it.

Heh, just go with like a 10 layer board!!!!!!!! J/K Actually that makes sense, i guess that is why they have yet to implement that feature on any cards. Yeah, the dual channel ddr is what I was talking about, but I guess that wont work. I guess we will have to just wait for 2ns ram!

 

[Richard]

Hang in there.

QDR, is right around the corner.

Trone (Jul 25, 2001 12:52 p.m.):

Richard999 (Jul 24, 2001 11:27 p.m.):
The technique you're referring to sounds very similar to 3dfx's SLI.

The only problem is that it requires two independent GPUs to go along with the RAM. In the case of the GeForce3 that would be one heck of an expensive card.

It is a nice idea though.

Another technique would be dual channel ddr, but that too would be both expensive and a monumental feat to assemble. The number of traces coming off of the GeForce3 GPU is already high. Imagine doubling it.

Heh, just go with like a 10 layer board!!!!!!!! J/K Actually that makes sense, i guess that is why they have yet to implement that feature on any cards. Yeah, the dual channel ddr is what I was talking about, but I guess that wont work. I guess we will have to just wait for 2ns ram!

 

[Trone]

Richard999 (Jul 25, 2001 02:30 p.m.):
Hang in there.

QDR, is right around the corner. [/b]

Well to tell you tha truth, for me it will be a lot longer, as I only have enough money to usually spend about 125 dollars on a video card. The card I currently have is a Geforce2 GTS. I got it last week too. But yeah, QDR would be cool, I guess they could implement it in budget level cards? Or would the price make it restricted to high end?