Why GFX card with more RAM?

[Codex]

Originally Posted by SEE

1. Is CloD one of the exceptional tiltles that requires more than 1Gb even for a modest resolution and why?

No. VRAM usage is increasing with many newer games, particularly with more games using tessellation and advanced shader routines.

The thing with VRAM is that its where the developer wants all the scene assets to live, because it's traditionally the fastest RAM on a PC, also DirectX has "direct" access to it. So what the devs do, assuming they use Direct X routines, is load all the 3D models, Textures etc into that memory space and do all the pixel, vertex and tessellation processing on those assets in that memory space.

Remember the actual buffers that hold the final image are fairly small, for example:

1920 x 1080 = 2073600 pixels
2073600 pixels x 32bit pixel (RGBA 8bit each colour) = 8MB (roughly)
DirectX default buffering is double = 16MB of VRAM space

It's all the scene assets that take up the majority of space.

Originally Posted by SEE

2. There seems to be an industry practice of using identical chipsets but by disabling or changing a peformance parameter (i.e No of shaders, reducing clock speeds, etc) marketing a budget version. I.e a Nvidea GTX 2GB 560 is slightly cheaper than the 560ti version yet bench tests show only a 5% reduction in performance across the most demanding games which can be adressed by oc'ing. Would someone looking to upgrade on a budget be better going for the better chipset with lower RAM or the budget version with more RAM?

This is no different to how CPUs are badged and priced. It may surprise you that your 560 GPU would most likely have come from the same silicon wafer that produced a 580 chip. The difference is the 560 chip may not have performed to meet the 580 specs so the factory would have locked down the clocks, the number of shader units etc etc and made it a 560.

The performance gains from having extra VRAM memory is more than likely having enough space to store assets and preventing scene assets from being copied over to system ram. I've only dabbled in rendering and animating basic 3D objects in Direct X but there's a lot of stuff happening in background that needs to be appreciated.

[icarus]

Originally Posted by Codex

No. VRAM usage is increasing with many newer games, particularly with more games using tessellation and advanced shader routines..

Except I have not ever seen any games use over 2 gb vram even at 2650 x 1600 res with AA at 16x super and I have seen Cod use well over 2 Gb with no AA or AF. That seems exceptional to me.

[Codex]

Originally Posted by icarus

Except I have not ever seen any games use over 2 gb vram even at 2650 x 1600 res with AA at 16x super and I have seen Cod use well over 2 Gb with no AA or AF. That seems exceptional to me.

There are a few reasons why I think this may the case. Mind you I'm only speculating here:

1) I believe the code base of CoD lives in the managed world, i.e. .NET and it makes unmanaged function calls to the DX API to render the graphics. This style of programming model is inherently a more memory intensive operation and slower (only slightly) as there's lots of storing of memory heaps and stacks and buffers going on. But it means more manageable code, no need to worry about memory leaks as much as unmanaged code and easier to update / modify.

2) We're dealing with a flight simulation that needs to create a land mass which is not only accurate but vast as seen from the sky, this means more memory is needed than the average game to store the environment. On top of all that, you have textures and buffers (pixel, vertex, shaders etc.), and 3D models with a higher than average poly count.

3) Optimization. I don't think CoD is properly optimized, hence why it's going through a complete rewrite at the moment.

[icarus]

Originally Posted by Codex

There are a few reasons why I think this may the case. Mind you I'm only speculating here:

1) I believe the code base of CoD lives in the managed world, i.e. .NET and it makes unmanaged function calls to the DX API to render the graphics. This style of programming model is inherently a more memory intensive operation and slower (only slightly) as there's lots of storing of memory heaps and stacks and buffers going on. But it means more manageable code, no need to worry about memory leaks as much as unmanaged code and easier to update / modify.

2) We're dealing with a flight simulation that needs to create a land mass which is not only accurate but vast as seen from the sky, this means more memory is needed than the average game to store the environment. On top of all that, you have textures and buffers (pixel, vertex, shaders etc.), and 3D models with a higher than average poly count.

3) Optimization. I don't think CoD is properly optimized, hence why it's going through a complete rewrite at the moment.

Agreed, which makes it more memory intensive than other games or even other sims. I've seen it in Evga Precision peak at 2.5 Gb vram! That is exceptionally high and most likely a result of #3 not so much #2.

[Liz Lemon]

Originally Posted by icarus

Except I have not ever seen any games use over 2 gb vram even at 2650 x 1600 res with AA at 16x super and I have seen Cod use well over 2 Gb with no AA or AF. That seems exceptional to me.

So what? That doesn't change the fact that there are games that use over 2gb of vram - and cod is one of those titles.

Also keep in mind that the game is using a deferred rendering pipeline. That means that the game is rendering to multiple buffers - all of which take up space that is directly tied to resolution.

[icarus]

Originally Posted by Liz Lemon

So what? That doesn't change the fact that there are games that use over 2gb of vram - and cod is one of those titles.

Also keep in mind that the game is using a deferred rendering pipeline. That means that the game is rendering to multiple buffers - all of which take up space that is directly tied to resolution.

Name one that uses 2.5 gb vram with no AA or AF applied. Crysis2 and BF3 use 1-1.5 gB with 16x AA and 16x AF and there is no sim that uses 2.5 Gb with no AA and AF. CoD uses so much, not because it is so complex, it is because it is not optimized. That is why they are redoing the graphics engine.

BTW, I'm not being negative here, I'm being positive. When this is optimized there will be hope for those with less than 3 Gb ram.

[SEE]

The discussion regards Vsync and Triple Buffering interested me based on the frame rate integer jumps that Ataros mentioned.

I have always had Vsync enabled and triple Buffers set but decided to test a MP session with Vsync disabled and Triple buffers Off. Apart from the screen tearing, I was acheivineg 80+ fps at altitude and game play was much better even in the hotspots were FPS sink badly - it seemed much smoother even when fps went below 30.

I am torn wether to put up with the screen tearing (which isn't so bad that its unplayable with headtracking) or go back to Vsync capping at 60hz

EDIT: After writing this I did a bit of googling and found this interesting article regards Vsync/Triple Buffering and the advantages/disadvantages.


http://www.tweakguides.com/Graphics_9.html


This link explains how Triple Buffering works - why it isn't supported in DirectX3D, etc.

http://www.anandtech.com/show/2794/1

[Codex]

I agree with that tweak guide. It's comes down to personal preference. We've all got different hardware and its a matter of trying different settings.

[Chivas]

I have less than 3 gigs of vram and have never had a CTD, or Launcher exe problem, or frame rate issue. It tells me the code isn't all bad, but an optimization/rewrite of the code it still urgently required. Hopefully the next patch will fix most of the performance issues, and the developers can put more resources in fixing the game play issues.