» Sun Jan 16, 2011 7:07 pm
Like absolute poop from a can. When fallout launched an we started modding it. We noticed they had these really high resolution versions of weapons for 1st person view. So first thing we did was switch it to use the 1st person version in 3rd person, or in the world model. This made things look great. However at this point it steered the modder the wrong direction. Into thinking that DXT compression was making the images look of less quality. Into thinking the only way to get high resolution is to use large sized texture maps. Basicly ignoring technology produced to further gaming development over the last 20 years. Fallout 3 doesn't really need AA, besides it conflicting with HDR lighting in many games (meaning it will fry your card) The AA only comes in handy under a few rare situations. AF however is quite useful (texture filtering) an 2X-4X AF can clean up the vanilla textures a whole lot, so much that you might not need a texture pack.
So yah that's why the texture packs have performance issues. This is rooted in the mistakes we made a long time ago. Once the noobys saw a few people with 4096 and 2048 texture maps, then they just did all their texture in that size. LMAO the other day I found a pair of earings that dragged my framerate down to 3 FPS. Heh another time I found a Rifle that the texture was over 16,000Kb, when most 1st person vanilla textures hover around 600kb. Funny huh, it gets worse. Lots of folks don't know what a mipmap is or they were too busy smoking meth to make sure they encoded mipmaps when they saved it. Then DXT compression is a computer language made just for grapics cards. A image encoded in DXT looks the exact same to us, but to the grapics card a DXT-1 image looks 8 times smaller as far as the video memory, 4 times smaller for DXT-5.
Once you get into becoming a texture god though you'll see that we should have went the opposite direction with our texture process. Yes you still need a 4096 or 2048 map to make HD textures, however that's just used to create the Conceptual if you wana call it that. Then you shrink it down, as you gain sharpness when you shrink an image, an lose sharpness when you enlarge it. Those 4096 and 2048 maps will keep 99% of the detail when brought down in 1/2 steps 4096,2048,1024 in 96 pixels per inch supersampling (or whatever is native to the monitor) however once encoded in DXT with mipmaps what used to be 5,000Kb of video card memory is now 600Kb just like the Vanilla stuff. Thus allowing mostly all the detail of the huge map, but also allowing an order of magmatude of more chaos to be running along with that content.
Camera view factors into the size needed the most. For example the player charicter body. Because of the camera views you get with the body, 2048 maps are the only ones that look HD, 1024 looks great too, but only if the head, body, hands are all 1024, an you would only really need a 1024 if you were lacking in hardware. Can't hurt to spend a little system resources on making your charicters body look good. So somewhat the player charicter is a exception to the rule. Moving on.
2nd thing that factors into the size needed would be the actual in-game size of the item. Something the size of the player charicers hand, doesn't need a 1024 map, it barely needs a 512, an can look fine with a 256 map. Something like this 256 map is only going to be like 15Kb LMAO For example some of FOOK2 is 1,200Kb for a hand size. This is important because many of the player created weapons are made up of many different parts. So on some weapons you could end up using 4-5 1024 maps for hand sized items. When it could look the same if the texture matched the actual size of the items. Which depends, however you can't go wrong with 1024, it's only 600Kb even with a 35% increase due to having the mip-maps encoded into the texture.
Now that we got that out of the way. TEXTURE PACKS well those folks had a tough job. Just about every ground/terrain texture is a 256 map, it's a very small texture, maybe they are 512, idk but if you 1024 that you just doubled the amount of resources you need to run the game. Hence why they run so poopy/laggy. That being said you can take the largest textures max res for high end machines then shrink them down to vanilla sizes for the same type item. Because Bethesda was very smart about optimization it's a good rule of thumb. Anyhow you now have the same high res great looking texture, but in a package that cut the fat so that it now performs well, plus it gained some sharpness from the shrink, and you know for sure it's in the correct DXT depending on if it has any transparency in it, even further you know for sure that the mipmaps are encoded.
A real world example of this would be MMM. Yah sure it has great textures, Mart's work is pure magic. However if the textures had been properly optimized Mart could have set the default spawn rates double what they are, you know more chaos more fun, an at the very least it would provide a buffer for stuff like MMM gone wild, or natural selection, hunting an looting, which are problems with MMM to a degree. Feral Ghoul rampage for example they also use a high res texture map for a creature that's sole purpose is to look like crap. Seems like a waste, an that's why feral ghoul boggs down in certian areas. This is why the South east part of the wasteland map bogs down when you use FOOK2 raider armors, with MMM. The sheer amount of raw data being rendered an culled. Even having culled huge texture maps slows the framerate down, and you don't even see it. As much time as I put into installing the mods I also have to put into fixing them. Maybe 60-70 hours for 150 mods. That number doubles when you have to optimize the textures.
