looking for official crossfire info relating to vram and can NOT find it

[toyota]

sometimes I come across people who think that running two cards in crossfire or sli will double the amount of vram. right now there is a guy on another forum who says his 4850 1gb crossfire setup is showing 2gb of available vram.

well I have spent 30 minutes trying to find one shred of official info to prove that vram does not get doubled. so does anybody have a link that I can give this guy or anybody else that thinks that vram gets doubled in crossfire?

 

[happy medium]

Do SLI or CrossFire double the memory ?
This is one the questions that many people make mistake understanding it .
The answer is NO , SLI or CrossFire doesnt double the memory , You have a 512MB card , adding another card wont make your memory 1GB, it still will be 512MB,so both cards will use 512MB of RAMs but the memory won't double.
Another example :
A game needs 512MB card to run at maximum settings , and you have a 256MB card , and you think if you add another 256MB card , your memory will be 512MB and thats not true.

You may see this mistake in many sites.

http://www.tomshardware.com/forum/245454-33-crossfire-faqs


some more info...............


http://www.slizone.com/object/slizone_ask_mmm013.html


and more...............

The caveat here is that CrossFire X will settle on the lowest core GPU clock, memory clock, and video RAM size to determine the operative clock speeds and effective memory size. As a result, a Radeon HD 3870 X2 paired with a Radeon HD 3850 256MB would perform like a trio of Radeon HD 3850 256MB cards. And, of course, that means the effective memory size for the entire GPU phalanx would effectively be 256MB, not 768MB, because memory isn't shared between GPUs in CrossFire (or in SLI, for that matter).

http://techreport.com/articles.x/14284

 

[toyota]

thanks happy :thumbsup:

 

[betasub]

This is a result of Alternate Frame Rendering (used by both Crossfire and SLI): each GPU is working on a different frame, and so RAM on different card/GPU is not cumulative (i.e. is not available to other GPUs).

 

[pandemonium]

sometimes I come across people who think that running two cards in crossfire or sli will double the amount of vram. right now there is a guy on another forum who says his 4850 1gb crossfire setup is showing 2gb of available vram.

They're wrong in thinking your memory is shared between the GPUs and thus being added, but ultimately the point is that you're getting nearly double the processing power. Which in a lot of cases is true when running multiple GPUs. Most cards today scale between 70-105%.

To reiterate, it's a mute point if you're doubling your GPU memory. Memory amount isn't everything. Most cards don't even get bottlenecked by how much RAM they have. This is also a clever ploy by manufacturers to price higher memory cards higher; when it doesn't actually improve performance. I'm not saying it's like that always, just most of the time. Some cards do actually need more RAM than originally designed with.

 

[deimos3428]

A game needs 512MB card to run at maximum settings , and you have a 256MB card , and you think if you add another 256MB card , your memory will be 512MB and thats not true.

Well, yes and no. Your memory in that example is 2x256MB - and that's the way it should be written for reasons of clarity. There is twice the memory available, it's just not pooled across the GPUs and that imposes limitations on what you can do with it.

Each GPU can only use its own memory. Since each card has 256MB, you can't run the application that requires 512MB. However, you can (theoretically, with perfect scaling) run an application that requires 256MB at double the frame rate because each card has 256MB. Those instances run in parallel until the output is stitched together; it's still possible to make use of all of the physical memory in the system.

It's quite like adding additional servers to a farm. The total amount of RAM in a server farm composed of identical machines scales with each new machine. You can run multiple instances of an application in parallel. (Depending on what you're doing, you could also stitch the output together.) But no single server in the farm can run an application larger than its own physical memory, regardless of how many machines are racked next to it doing the same thing.