The effective significance of memory bandwidth?

[claptrap22]

To all:

I'm continuously trying to become adept with respect to the more technical aspects of pc building and gaming since I built my first rig last summer. In this vein, I've become interested in the effective implications of memory bandwidth. Let me ask my question by reference to an example. Take an HD 7950 with a 384-bit bandwidth and a GTX 680 with a 256-bit bandwidth. It is a matter of fact that the 680 outperforms the 7950. My hypothetical is the following: is there any scenario where the 7950 trumps the 680 on account of its 'superior' bandwidth? Asked another, more cynical way: is the higher number for an AMD card something that provides window dressing but in effect is overkill for current and near future gaming needs? Many thanks in advance - I always find the folks here so helpful.

 

[claptrap22]

Sure... when the 7950 is overclocked it can beat a 680 pretty easily at 1080p or higher.

Anyways, you can't compare specs like memory bandwidth across architectures. You can really only compare specs in a relevant way by comparing different cards of the same architecture (e.g. 660ti vs 670 vs 680 which are GK104 chips or 7950 vs 7970 which are Tahiti chips).

The only way to know about the performance of graphics cards is to read reviews. Lots of reviews.

Thanks for the reply, BigMack70 and I get what you're saying. By trade I am a philosopher and a theorist so here I'm trying to get a grasp as to what these specs are telling me. I have friends who occasionally ask me for tech advice. As a way of learning more myself, I try to imagine what I might say to someone who was shopping for a card and said something like, "The bit number on this card is bigger than the one on that card. What does that mean? Doesn't it make that card better?" And so on.

Given what you've said, BigMack70, do you have any leads I might turn to to take up your advice? The implication of your response is that an answer to my question depends on further knowledge of how the chip's architecture is designed with respect to memory bandwidth. I find that reasonable and would like to know how I can learn about that. (Yes, I suppose I could google it, but since we're having a conversation here...)

 

[claptrap22]

I should have been more precise. I meant: where could I learn more about the chip design with respect to architecture? Not something terribly technical but technical enough to get a decent understanding. I've got places for reviews properly in hand 😉

 

[swifty_morgan]

google.................?

 

[bigshootr8]

Well google can put you on the path to sites that will give you information regarding the significance of bandwidth and or a cards architecture. To my knowledge the significance of a memories bandwidth or bit is the amount of information that can transferred over a lane i.e, pixels, colors or back in the day there was a term called pixel pipelines I think its a more modern way of painting the picture.

Also, yes the 7950 has the ability to be overclocked a bit. However you need to take into consideration that not all GPU's are created equal and there is a thing called the silicon lottery each chip has a different threshold for heat and how far you can push it. I also believe that to get to 680 levels and even 7970 and or 7970 ghz levels you would end up with quite a bit of heat and you would at that point would want to consider something like superior air cooling top tip shape cooling and or water cooling in a custom loop setting.