Video Card... Is more Vram or stronger GPU better?

[Buetzel]

while writhing a reply in http://www.dual-boxing.com/showthread.php?t=33260 , i remembered this discussion was going on here.

do we have someone here on the forum who has definitely too much money and who could (and would) test this?

as far as i see it, there would be a simple (yet expensive) way to test this.
one would need a nvidia 460 with 1gb ram, one nvidia 460 with 2gb ram and one amd 6870 (better performance as far as some benchmarks from testing sites go).

my guess to the topic would be, that indeed 2gb should be a remarkable improvement to 1gb, IF (that's a BIG if) wow really utilizes it.

my reasoning: most sites indeed test graphics cards only for one monitor and more important only for one game instance. so for example tomshardware has tested the 460 2gb as a bit slower performance wise as the 460 1gb model and of course both models slower as the 6870. i think the 2gb model of the same card is a little slower because the bigger memory has to be managed. and if a game or benchmark doesn't utilize the memory of that size, only this management overhead will be noted.

now... when we use 5 instances of a game, the textureloading is rather important for performance (especially in big hubs like lagaran was in wrath - and i suppose it was loadaran, not lagaran, as i suspect the low performance in the city to stem from a lot of texture loading all the time, either from hdd/sdd or from system ram, which still is slower as videoram and needs computing power to be moved from system ram to video ram, not from lags).
this probably is also the case in bigger raids and battlegrounds. the bigger the number of players, the more textures have to be loaded for the characters.

as someone mentioned, i also don't believe that wow is multiboxing optimized in that it can use a texture loaded for one instance on another instance of wow. so all textures have to be loaded 5 times to the graphics card's video ram. also - this reduces the video ram to 200mb per instance, which is clearly a lot less than the whole 1gb of a standard card these days.

as i said earlier - wow has to really utilize the vram still. and the reason for me posting now was actually this point. i remember an addon which i used in early wrath, which dynamically adjusted graphics details and which had also a way to tell wow how much video memory it should be using for textures.
anyone here who remembers that one or knows what i'm even talking about? vital for really testing the performance of a graphics card with more vram would be to adjust wow, so it uses the bigger vram. if wow is set to only use ... for example 64mb vram for textures, then of course the card with a faster gpu would flat out win and card with the same gpu but bigger vram would show no improvement.
i will definitely look for the settings, that addon changed. hopefully i find that stuff again...

edit2: found it, i think. http://www.wowpedia.org/CVar_gxTextureCacheSize
notably: only works with dx9 api... so perhaps... not too much help with the gpu's we're discussing... at least not if one wants to use the added benefits of dx11 in cataclysm.

edit3: also: http://www.wowpedia.org/CVar_textureCacheSize
32mb or 64mb for textures in the system ram? if that's the case, then of course the graphics card has to load textures from a drive and graphics performance does not benefit from system ram that much. also explains drive activity, which sam pointed at.


on the other hand: of course also more graphics effects like spells and stuff need more performance. so for example in raids/battlegrounds performance is also heavily influenced by the power of the gpu.

so imo the questions have to be:
- is it (still?) possible to adjust wow, so it uses the given vram effectively?
- is there a performance gain with more vram and same gpu?
- if yes, is this performance gain bigger than the performance gain of a better gpu (possibly only slightly better gpu, when using the gpu's i mentioned)

edit: just got another thought: of course the bigger vram would make itself noted only after some time playing. if you first log in to dalaran, still the textures have to be loaded from hdd/ssd. but maybe after cycling once around dalaran, all textures should be loaded and THEN performance should be better than with less vram. so... it won't be a performance gain which is always and instantly noticeable, thus perhaps pushing the benefits of a more powerful gpu a bit more.

@sam: perhaps lax is right after all. at least when wow doesn't use more than 1gb vram while running 5 instances.


i for one am also really curious about this, as i need a new graphics card in the near future (geforce 8800 gts 512 doesn't cut it anymore. me wants myself some dx11 eyecandy!). so if anyone would sacrifice the money (or maybe more like "if someone already has this cards lying around or uses them for different machines or something"), i'd definitely like to hear about it.


and lastly - i hope my long sentences with too many commas and brackets don't cause you headache. i know it does to me. sorry for that.

[Sam DeathWalker]

http://forums.worldofwarcraft.com/th...48996420&sid=1

If it's a 2 GB card, each core has 1 GB to itself and you should maybe give it something under that, like 800 if you're doing nothing else on the card. If you multitask and you're on Vista or 7, it's best to use a lower number or let the game manage it on its own. Keep in mind that gxTextureCacheSize only functions in d3d9ex mode.


Well this is most interesting to be sure.

WoW only using 32/64MB per client of system ram for textures in default.

But you can set it at much higher, but you have to watch the 2G limit as your game will lock up if you run out of memory.




Ok here we are the 3rd post in this thread is gold:

http://forums.nvidia.com/index.php?showtopic=182340



Posted 07 October 2010 - 01:45 AM
Obscene CPU overclocking is technically one way to get smooth performance in Dalaran, but it's hardly worth looking until optimizing game behavior, which is far more reliable and significantly less risky, no longer yields measurable gains. The distance between those points can greatly minimized with a few simple adjustments, though.

First thing, however, is to accept the degree of World of Warcraft's CPU limitation. Achieving 60 constant frames per second while in the middle of a populated Dalaran is very difficult because of the hundreds of client-server interactions occurring between all of the players. My personal frame rates will dip to about 50 FPS at high-traffic times, but outside of Dalaran and chaotic Wintergrasp fights, rarely is visual performance ever an issue.

A quick look-up chart for the processAffinityMask CVar in the Config.WTF would look something like the below graphic, which has eight common configurations. Adjust accordingly by the core binary values at the foot of the stack if you don't feel your ideal circumstance is listed. Keep in mind that, with the exception of manual core allocation, the process affinity mask variable is no longer required to be present or explicitly defined (automatic detection/allocation is "0" which runs the application across all available logical cores) within Config.WTF. Being a 32-bit application, the processAffinityMask CVar is only capable of addressing 32 cores with a maximum CVar value of "4294967295."


Changing the gxTextureCacheSize variable is less straightforward because of the numerous factors that affect graphics memory consumption, like anti-aliasing, shadow volume precision, resolution, and the buffering model. Graphics cards like the GTX 480 are powerful enough to drive the largest monitors without the possible likelihood of supersaturating its memory capacity, so you could set your CVar to match your VRAM size directly. As pointed out above, however, there is still more to do regarding your performance overall.

For in-game settings, the biggest concern lies in shadow quality. There are six levels of detail, but only a few need to be kept in mind. The first level draws only the traditional shadow blobs for on-screen actors (players, NPCs, creatures, etc.) with baked (pre-generated) shadows for world geometry. The second level draws geometry- and lighting-accurate shadows, but only for your character and those within very close proximity; baked world shadows are still used. The third, fourth, and fifth levels use dynamic shadows for world geometry as well as actors, with varying level of detail ranges for both (dropping actor shadows before world,) increasing in range and resolution as the slider is pushed farther to the right. At the sixth level, the maximum extrusion range and precision are used for both subsets, and shadows may overlap. As you can probably tell, the maximum value for this will be exceptionally demanding and should be avoided; at the very most, use the second-highest setting.

A few other pointers would be to never use anything beyond 4x in-game multisampling, enable vertical synchronization and triple buffering, and never run in Windowed mode. These all contribute to your performance in their own fairly appreciable ways.

As for NVCP settings, try the following.
Quote
Ambient Occlusion: (Off)
Anisotropic Filtering: (Application-controlled)
Antialiasing - Gamma Correction: (On)
Antialiasing - Mode: (Application-controlled)
Antialiasing - Setting: (Application-controlled)
Antialiasing - Transparency: 2x (Supersample)
CUDA - GPU’s: (All)
Extension limit: (Off)
Maximum Pre-rendered Frames: (0)
Multi-display/mixed - GPU acceleration: (Single display performance mode)
Power management mode: (Adaptive)
SLI rendering mode: (Single-GPU)
Texture Filtering - Anisotropic sample optimization: (On)
Texture Filtering - Negative LOD bias: (Allow)
Texture Filtering - Quality: (Quality)
Texture Filtering - Trilinear Optimization: (On)
Threaded Optimization: (Auto)
Triple Buffering: (Off)
Vertical Sync: (Use the 3D application setting)


That post is so good I'm going to put it up on my forum, so it dosn't get lost.



IsBoxer wants windowed mode though.