need info about memory bus (gpu)

[MisterWhite]

So i've heard those metaphors about memory bus being a "tunnel" and memory being the "water" or something like that and that if the "tunnel" is narrow less "water" can flow, but if it flows fast the "tunnel" could be narrow. I guess what i really need is some kind of calculations on how many bits on a video card you need to access all its memory and what speeds you need to counter the "narrow" tunnel. And maybe some general facts because i couldn't scrap the information i need on the web.

[TidaLWaveZ]

7 minutes ago, MisterWhite said:

So i've heard those metaphors about memory bus being a "tunnel" and memory being the "water" or something like that and that if the "tunnel" is narrow less "water" can flow, but if it flows fast the "tunnel" could be narrow. I guess what i really need is calculations on how many bits on a video card you need to access all its memory and what speeds you need to counter the "narrow" tunnel. And maybe some general facts because i couldn't scrap the information i need on the web.

 

What an ambitious question.

[Kryptyx]

9 minutes ago, MisterWhite said:

So i've heard those metaphors about memory bus being a "tunnel" and memory being the "water" or something like that and that if the "tunnel" is narrow less "water" can flow, but if it flows fast the "tunnel" could be narrow. I guess what i really need is calculations on how many bits on a video card you need to access all its memory and what speeds you need to counter the "narrow" tunnel. And maybe some general facts because i couldn't scrap the information i need on the web.

Why do you need this? What are you trying to figure out?

[MisterWhite]

1 minute ago, TidaLWaveZ said:

 

What an ambitious question.

Sorry

[Guest]

The way it's been explained to me is that the memory width is like lanes on a highway; basically the more lanes the more information can pass through. But there are more factors determining the effectiveness of said bus width (architecture, speed, CUDA/Stream processors, MHz, VRAM, etc) so it can't be the only specification you look at when browsing for a GPU. 256-bit seems to be pretty standard, though I think AMD cards tend to be a bit higher. It plays a factor for sure, but again there are so many other contributing factors it's kinda hard to keep track.

 

In terms of calculations, I got nothing. Perhaps someone wiser than me on the forum knows.

[MrLomax]

ok, so you just sat there waiting for us to tell yo a bedtime story

[MisterWhite]

2 minutes ago, Kryptyx said:

Why do you need this? What are you trying to figure out?

Some guy in Facebook group brought up that "video card is better if it has wider memory bus" and he still does not believe a word i say about it. During my search i got curious on how much memory can be accessed depending on the bus width

[typographie]

Really what is more likely to matter to actual gaming performance is the memory bandwidth, of which the memory bus width is one contributing factor. To borrow your metaphor above, it's less important whether the tunnel is wide or the water flow is fast, because in the end you're getting a final measurement of water volume per second. A wider tunnel and a faster flow are two ways to arrive at the same end.

 

To use a very broad generalization, cards with higher memory bandwidth tend to tolerate higher resolutions and higher levels of antialiasing (memory-intensive tasks) with somewhat less performance loss than other cards with lower bandwidth. But I urge extreme caution trying to draw conclusions about one card versus another from a different architecture based on memory performance alone. Actual real-world benchmarks are what really matter, not the theoretical figures. Better memory performance doesn't necessarily always make one card faster than another.

[MrLomax]

normally video memory bus is important in chipsets and processors, also gpu chips but a bus is a non temporal measure relative to the amount of data in a segment of transfer material

[TidaLWaveZ]

7 minutes ago, MisterWhite said:

Some guy in Facebook group brought up that "video card is better if it has wider memory bus" and he still does not believe a word i say about it. During my search i got curious on how much memory can be accessed depending on the bus width

 

Does widening the expressway alone change your cars ability to drive faster?

[MrLomax]

5 minutes ago, typographie said:

Really what is more likely to matter to actual gaming performance is the memory bandwidth, of which the memory bus width is one contributing factor. To borrow your metaphor above, it's less important whether the tunnel is wide or the water flow is fast, because in the end you're getting a final measurement of water volume per second. A wider tunnel and a faster flow are two ways to arrive at the same end.

 

To use a very broad generalization, cards with higher memory bandwidth tend to tolerate higher resolutions and higher levels of antialiasing (memory-intensive tasks) with somewhat less performance loss than other cards with lower bandwidth. But I urge extreme caution trying to draw conclusions about one card versus another from a different architecture based on memory performance alone. Actual real-world benchmarks are what really matter, not the theoretical figures. Better memory performance doesn't necessarily always make one card faster than another.

+1

just as you said,  A wider tunnel and a faster flow are two ways to arrive at the same end. also temperature

[agent_x007]

There is no corelation between bus width and maximum memory.
We were getting more memory per bit-width every other GPU made it's debut (in case of NV : starting with 6800 Ultra in 2004 128MB/128bit, 8800 GTX 256MB/128bit, GTX 480 512MB/128bit, GTX 680 1024MB/128bit, GTX 980 2048MB/128bit).
HBM changed that, HBM 1.0 = 128MB/128bit.

If technology would be available earlier, U could have got 512MB on 6800 Ultra, or 3GB on 8800 GTX.
+ Virtual memory makes "MB/bus bit limitation" useless anyway.

The "narrow tunel" or memory bandwidth bottleneck, is highly program and GPU architecture dependant.
Basicly : It's not about getting rid of it (since it's not possible), but instead making it almost non-existent.

Memory bandwidth is needed the most with higher resolutions, AA/AF levels and/or general graphics settings.
BUT if u play on Low or HD Ready it's pretty much useless (compare for example GTX 285's 512bit bus to GTX 750 Ti's 128bit and then compare actual performance difference between the two).

PS. Remember that "1 Bit of bus width" is not eqaull to other "1 Bit of bus width".
Prime example : 1x GDDR5 = 2x GDDR3 (GDDR5 bus has twice effective bandwidth of equall width GDDR3 bus).