Multiboxing Hardware Science!

[Sam DeathWalker]

I don't belive that the gpu actually renders anything that is not visible on the display.

Thats not to say it dosn't load the textures or set aside the memory for all clients, just that the gpu dosnt do the calculations, even if the data is available if its not going to the display. I am sure it loads the textures for all clients all the time though.

In the first screen yes the GPU is rendering all 15 clients at the full resolution, but at 10fps.

In the 2nd screen the GPU is only rendering 1 client but becuase your fps and effects are higher (are they?) more GPU load is being used. And in the 2nd screen more of the world is showing. Although the 2nd screen is also shown in the first I am fairly sure its at 10fps in the first screen and 60fps in the 2nd.

My point is that in the 2nd screen with only one client showing the other 14 clients put ZERO load on the GPU.



Well that video card is something for sure. Excellent information keep up the good work.

300Mb for first client with most effects and 65M for addional lower alts is excellent infos.

Maybe 2G cards are not needed, it seems that there is a lot less video information needed by the card then I had thought although the fact that there are lots of 4K read and writes would tend to support that very little video information is being used. They have made wow to be played on some really poor computers.


Sure wish we could find out if it uses the vram that not "in use" in the gpuz display as cache.

These people clearly are not stupid obviously if something can be done to speed up game play it probably has.

http://developer.amd.com/documentati...ure_Topics.pdf

That is very bad for stable performance and MultiGPU (SLI/Crossfire) scaling but it was a manageable
solution at the time. Streaming allowed us to stay within the 32 bit limits with run‐time
data requirement sometimes exceeding the limits. On 64 bit and enough main memory
or when using half resolution textures the texture streaming is not necessary and performance is more stable.

Thus the conclusion is that the more Vram (and system ram) you have the better even if gpuz shows only some being used I am sure the rest is being used for cache purposes. Might have something to do with "lod" from the above artical.


The LOD Bias affects MIP-Mapping in determinating the MIP-Level.

MIP-Maps are collections of one and the same textures in different resolutions. To allow perspective effect and avoid underfiltering, objects in the back of a scene need to be displayed smaller and the textures in use need to have a lower resolution. The MIP-Map Detail Level globally determines if MIP-Mapping happens with the best possible resolution (and therefore best imagequality) or with best performance in mind through all MIP-Levels.

Now LOD Bias means global displacement of the MIP-Level determination. With this value set below zero, textures of higher resolutions will more likely be applied, while with a LOD Bias greater than Zero MIP-Mapping will "prefer" textures of lower resolutions.



gxTextureCacheSize is a CVar that affects the D3D9 API; if you're already using DirectX 10/11 mode, this CVar will not have any affect. This variable decides how much memory - in megabytes - to use for caching textures. The default value is '0', meaning the game will handle this on its own. Some players have found better performance - particularly when turning - by manually selecting a value for this. As a general rule of thumb, ~70-75% of your VRAM is best. For example, if you have a 1.5 GB GTX 480, a good value to try would be 1152 MB. The syntax would look like this:

SET gxTextureCacheSize "1152"

For quick reference, these are 75% values. Feel free to use more or less as depending on specific configuration settings.

128 MB: 96
256 MB: 192
384 MB: 288
512 MB: 384
768 MB: 576
896 MB: 672
1 GB: 768
1.125 GB: 864
1.25 GB: 960
1.5 GB: 1152
2 GB: 1536


Ok that answers the shared question also. 75 percent of the video ram is for texture cache, which no doubt is available to all clients.

They want 25 percent free assuming only one client so with 10 or 20 box maybe less texture cache should be set aside according to a 80M per client rough estimate.