VRAM Usage - 6970 (2 GB) vs. GTX 580 (1,5 GB) vs. GTX 580 (3 GB)

[Codex]

Originally Posted by Buchon

There a lot of confusion in this thread about how SLI/Crossfire work.

SLI/Crossfire cards and systems dont share VRAM between GPUs.

That's not entirely true. It depends on how the developer has set up the buffers in DirectX. While yes they physically don't share memory, but with DirectX you can map the memory to appear as one huge chunk.

http://msdn.microsoft.com/en-us/libr...=vs.85%29.aspx

[madrebel]

Originally Posted by Codex

That's not entirely true. It depends on how the developer has set up the buffers in DirectX. While yes they physically don't share memory, but with DirectX you can map the memory to appear as one huge chunk.

http://msdn.microsoft.com/en-us/libr...=vs.85%29.aspx

no, you can't. each GPU has X memory. gpu1 can't access gpu2's memory. if GPU2 even wanted to access GPU1's buffer then this would slow down the frame draw process which is why it isn't done.

[Codex]

Originally Posted by madrebel

no, you can't. each GPU has X memory. gpu1 can't access gpu2's memory. if GPU2 even wanted to access GPU1's buffer then this would slow down the frame draw process which is why it isn't done.

I know. I was replying to Buchon's comment about VRam and System Ram.

[GOZR]

The best way for SLI is split rendering half the screen and the other for the other half..

Split work better no matter what .. NVIDIA profile now do not have this option but i tried all and it's still better for high textured sim

[Armatian]

Split frame doesn't share the load evenly, there's no perfect method, you can easily have one half having a huge HDR/shading workload and the other half showing the floor.

On topic: It looks clear in luthier statements that the engine need resource optimizing (should improve memory use), and more fine threading, because the same thread that renders, needs to load finer textures when you'r close to ground, hence horrible stutter you all see.

[Slechtvalk]

Originally Posted by madrebel

no, you can't. each GPU has X memory. gpu1 can't access gpu2's memory. if GPU2 even wanted to access GPU1's buffer then this would slow down the frame draw process which is why it isn't done.

this is setup picture from someone with 5970 (2 gig memory), when i use same settings (with my 1 gig memory card) you see video memory usage over 100%, and he's only using 51% of his memory (its says on picture below) which means he does got 2 gig video memory?

And why would ATI sell that card saying it has 2 gig when you only can use 1 gig anyways?

[doghous3]

@Slechtvalk

The HD5970 is a duel GPU card. Each GPU is assigned 1Gb of RAM.

I can confirm that CoD does not invoke the second GPU. Hopefully the upcoming patches will resolve this.

[Slechtvalk]

Originally Posted by doghous3

@Slechtvalk

The HD5970 is a duel GPU card. Each GPU is assigned 1Gb of RAM.

I can confirm that CoD does not invoke the second GPU. Hopefully the upcoming patches will resolve this.

But why as you see on picture I posted his memory usage is only 51% (with his 5970 card) while my 1 gig card with same settings is using over 100% video memory?

[Kankkis]

Originally Posted by Slechtvalk

But why as you see on picture I posted his memory usage is only 51% (with his 5970 card) while my 1 gig card with same settings is using over 100% video memory?

That s good question, i don't know, i think because windows see it 2gb that's the reason, but im sure games uses only 1gb memory, i mean they uses 2gb but there is both memorys with same data, working likely somehow that RAID hdd's, so if faster that way.

Kankkis

[Kankkis]

http://www.hardocp.com/article/2010/...b_performance/

The Radeon HD 5970 is AMD’s "fastest" video card. This is not a single-GPU video card like the Radeon HD 5870. The Radeon HD 5970 video card consists of two Radeon HD 5870 (down-clocked) GPUs connected via CrossFireX on a single board. While both GPUs retain all the specifications internally as a Radeon HD 5870, the performance has been reduced by core frequency reduction and memory frequency reduction. But worth mentioning is that many people find the 5970 very easy to overclock by large margins. There is also only 1GB of GDDR5 available per GPU. This memory video RAM limitation can cause high resolutions with antialiasing to run out of local video memory, since memory is not shared on the video card. Each GPU needs its own frame buffer space. The only way to fix this issue is to use more memory per GPU, but the 5970 doesn't have it.

In our evaluation of the GeForce GTX 580 last week, we did not compare it to the Radeon HD 5970. The reason being is that pricing on the 5970 was higher when we began the evaluation process by a full $160. The 5970 ended up being directly comparable to the GTX 580's pricing towards the end of our process. The GeForce GTX 580 is currently going for $499 online. The Radeon HD 5970 can be had now for $469.99 after $30 MIR. This is an excellent deal compared to past pricing, and we wanted to see how performance stacks up against the new GTX 580.