as a user of the free tools gimp and paint.net with some experience i want to add that the compression algorythm for dxt1 and dxt5 (most common formats for dx9 and games like skyrim le and oblivion) in paint.net is superior. never use gimp for saving your dxt1 or dxt5 textures! gimp is really nice for working with images but i recomment to save the files as lossless tga and save it compressed with paint.net. especially for saving mipmaps the paint.net supersampling algorythm is superior and much better quality. the quality of mipmaps is heavily underrated. most of the time the game uses the mipmaps and not the fullsize texture. omitting mipmaps or generating them with inferior quality will waste vram and slow down gameplay and degrade image quality ingame.
dxt1 and dxt5 are lower quality but much faster to decrompress compared to newer bc7 dx11 format available for FO4 and SSE. if bc7 hardware support is implemented this may be not relevant.
one thing in the description is wrong: the recommendation to use texture sizes like 1536 and 2560. this is misleading. it may work but it will slowdown graphics or cause incompatibilities because standard decrompression format and mipmaps use sizes of 128, 256, 512, 1024, 2048, 4096p and so on.
normal map textures suffer the most from compression artifacts due to the included directional info that will be distorted and i'm not shure if compression of this type is always better than downsizing and saving uncompressed. my eye normal map mod for skyrim is a good example of the negative impact of a low res vanilla texture showing its dxt1 compression artifacts in game and how it can be cured. bc7 seems to be a good compromise and very close to the original. i would recommend to use bc7 for games like fo4 or sse supporting this format nad if high res color is essential (eg. skin, skin gloss). the paint.net bc7 plugin works nice and is for free. bc5 was designed for normal maps but i'm not shure if it works flawlessly in all cases and with all games. if you are not shure and you need best quality (eg. for body, face) you can use bc7 also for normal maps.
Okay, the amount of things wrong with what you've said both in your comments / replies to people and on your main post is staggering, but ill try and break it down bit by bit.
To start with, your Criticisms of kkthebeast and Gambit77 are uncalled for, especially saying that kkthebeast doesn't know what hes talking about regarding meshes and texture resolution. I do allot of work in the background with the Raider Overhaul / Super Mutant Redux team, and that work includes extensive testing of various texture tricks and odd sizes for textures. By definition BC or Block Compression requires resolutions divisible by 4 in order to be able to create mipmaps, however if a texture doesn't have a height and width divisible by 8 at a minimum, it WILL make the game CTD, 100% Guaranteed. Height and Width needs to be divisible by at least 8 for the creation engine to function. Additionally, packaging textures that are not standard sizes into BA2 files is a total crapshoot, Ive done it, and it works for some people, but not for others, the cause of this is unknown, and they will work if unpacked, however loose files are their whole own set of issues for performance etc.
Regarding your comments about this only being relevant for people with low amounts of vram, again i have to disagree, also i highly doubt that any videocard with 1GB of VRAM can ever run the game smoothly or without swapping textures in and out of memory constantly, realistically FO4 needs at least 2GB to have any shot at smoothness at even the lowest setting, to add to that I can easily make FO4 use all 8GB of memory in my 1070 and still be running out. So vram is an issue for every FO4 player, not just those with low VRAM. and any amount you manage to free up, it will find a way to use.
Next thing, DXTn / ATI1 / ATI2 should be being used for NOTHING anymore, they all have BC equivalents now and DXTn was deprecated 8 years ago when we got Windows 7. To list them off... DXT1/DXT2 = BC1 DXT3/DXT4 = BC2 DXT4/DXT5 = BC3 ATI1 = BC4 ATI2 = BC5
To add to that, BC formats use a new extended header in the DDS file which allowed for more formats then the original four character code formatting allowed, anything that uses DX11 knows how to read the extended header, and by extension, knows how to use BC4/5/6/7.
BC6 and BC7 are new with DX11 and are far superior to BC1/2/3, which BC7 replaces. BC6 is irrelevant for this conversation because it is used only for HDR textures. now, BC7 allows for up to 7 different methods of compressing each 4x4 block of the texture, allowing it to select the best method for each block versus being forced to apply the same method across the entire texture, this leads to massive quality gains regardless of texture resolution. You can read about those modes on MSDN here: https://msdn.microsoft.com/en-us/library/windows/desktop/hh308954(v=vs.85).aspx
As you can see here, BC7 does an amazing job with changes in color.
What you see on the left is an enlarged copy of a 24 bit RGB test pattern, with gradients, on the right the same thing converted to BC7, there is some difference but the images look nearly identical, now, if we look at the same image converted to BC1 or 3....
We see a significant amount of chunkyness in both images, in addition we also see that BC1 / 2 / 3 all produce identical color output since the increase in size is only because of the increase in Alpha Bits, something that BC7 varies on a per block basis as needed, allowing it to use the unused alpha bits for additional color definition
Now, you really, really seem to like BC1 for some reason, and think nothing should use BC5/7 even though its the same size as BC3. With the availability of BC7, there is literally NO reason to use BC3 at all, it has a lower quality at the same file size. In addition to that, i argue there is no reason to even use BC1 regardless of the memory savings, the memory savings you are looking for can be easily achieved by resizing some of the existing BC5 Normal and Specular maps to less insane sizes.
In addition to that, just making a blanket statement that artifacting isn't noticeable until textures are below 1k is totally wrong, it depends entirely on the SIZE of the item the texture is applied to, for instance, deathclaws are basically 1 massive texture over their whole body and more or less need a 4K diffuse and normal map to look good, even at 4k some artifacts are visible simply due to the massive amount of surface area that texture is applied to. Meanwhile in the HD texture pack Bethesda made every ammo box model 4K for diffuse normal and specular maps, even blown up on a screen its near impossible to tell the difference between 4k and 2k, in game even 1.5k would look just as good, however the normal map could be tiny, like 512x because its just a box, and about 1k for the specular to prevent loss in the map detail. Similar thing as the deathclaw example goes for allot of walls and large objects, the more surface area something has, the larger the texture needs to be to keep quality.
Another note about textures, Normal and specular maps are BC5 for a reason, BC5 allows for 2 channels to be made at 8 bits / channel instead of 4 bits / color channel, allowing for a much higher quality to be achieved then would be possible even with BC7 at a given resolution, converting them to BC1 in the name of memory savings causes a even larger quality loss then doing the same for a diffuse texture.
for reference Ive included the PNG test pattern so you can run your own tests, just in-case you don't believe me.
RGB Test Pattern
Here is also a Comparison table noting some of the differences between the BC formats. Anything saved with DXTn formatting or ATI1/ATI2 is using the older DDS Header which is now deprecated in favor of details saved in the Extended header, introduced with DX10. Sadly the nearly all photo editors only support the older formats, which is also why BC7 files fail to open in nearly all editors.
As far as your crusade for vram savings goes, i would take a good look at how important each map is to a certain model, and fine tune each one to get the balance you want, you can use any size that is divisible by 8, however you may have issues packing the texture into a BA2 unless its using traditional sizes that are n2 based.
In closing
Don't assume that people that have spent 100's of hours creating fallout mods don't know what they are talking about.
BC5 for normal / specular
BC7 for everything else
- a guy who has spent 100's of hours working with the DDS format.
There is a much higher degree of intelligence in your post when compared to your comrades. For the most part, I was and continue to be in agreement. And just so you know, all my textures have always been divisible by 8, I just felt divisible by 4 was too limiting especially with regard to either best performance/visual by byte or best visual/performance per byte which is which is why I was recommending resolutions that were divisible by 8 without explicitly stating that they had to be divisible by 8. In any event, I have altered the direction of the project to more so initially only deal with specular maps and normal maps since these were the primary improvements in the high resolution texture pack. Both BC3 and BC7 deliver on smooth gradients with latter being twice as good, but to say that smooth gradients are always necessary is complete rubbish. So for diffuse maps we have BC1 when smooth gradients are not required and BC7 when they are required; do anything else and you are wasting precious vram. As for normal maps, it's safer to use 2048 BC7, but sometimes 2560 BC1 will look almost just as good and will use less memory (most be using the correct software). It's slightly more complicated for specular maps, if they are highly detailed BC7 will win each and every time, however if its a "full light/darkness or full light/darkness with barely visible detail (faded detail)" in that case BC1 is best bang for buck because the specular map is crap to begin with. Loose files are best because nexus mod manager is junk, I have noticed no anomaly with BA2, just with nexus mod manager. I will update guide in several days to be much more precise. I'll just point out that kkthebeast could not understand that: The game engine supports textures with resolutions divisible by 8 and that software can create and attach mipmaps to these divisble by 8 textures. He was 100% certain that could not be the case, and he was 100% wrong. When he couldn't see why he was wrong he spoke out of term and for that reason I said what I said. An intelligent person thinks before they reply, and he did not.
Wanted to say thank you for this post. Found it via Google and it helped me navigate the Intel photoshop plugin as a noob. Please consider adding this post to the forum so it can be better documented under your own control. :)
Have you had any luck with diffuse face textures and BC7? They come in BC3 by default and any attempt at using BC7 instead immediately causes the brown face bug over here.
This information is not current. For diffuse textures use BC7, for normals and specular maps use BC5. These are the newest best compressions with the lowest loss in quality.
The title of my mod "Strategies to simultaneously maximize visuals and lower VRAM". The keyword being simultaneously. Yes BC7 produces less artifact than DTX1 and yes BC5 produces less artifact than DTX5 and DTX1 however BC7 and BC5 consume double the memory. The loss or artifacting you are talking about is only visible at 1024 or lower resolution. This is why for diffuse I say to use 2560, so that the parent and child (1st mipmap) contain no discernible artifact. Since specular is shine, it cannot reveal an artifact or does a terrible job of it, therefore its safe and better to use DTX1. For normal maps, I specify if its under 1024 resolution save as BC5 because you are absolutely correct it would look like s#*! as DTX1 in that case, but if the resolution is above 1024 I advise to upscale in this particular case to 1536 (Perfect Resize) and save as DTX1. That combination outperforms the mere standard that BC5 is. Micro-managing (tricks) is better than macro-managing (standards). I specify some exceptions like overlaying 2 normal maps; in that case I say save as BC5 because there more artifacting in this case then mere resolution can fight. But most of the time, (higher resolution + old standard) wins vs (new standard + lower resolution).
The tutorial on demystifying textures (which sounds like you're trying to teach people that don't know anything about textures) should probably at the bare minimum include information on what you refer to as the new standard is, yet you don't mention the compressions that are the new standard at all. There is a reason they are the new standard. The main one being that current video cards have lots of power and lots of vram at a low cost to consumers. That's why the goal of maximizing quality should be discussed in this type of tutorial. Sure there's nothing wrong with telling people how to optimize textures for a potato, but you should also mention how to also properly compress textures for a current computer. The larger file size for the newer compressions also come with a large increase in quality, and with a current video card you have the vram to spare for the newer compressions. test pattern uncompressed test pattern compressed in dxt5/dxt1 test pattern uncompressed side by side with test pattern compressed in bc7
No, do NOT use 2560, Non standard sizes even if they are divisible by 4 and 8 are not compatible with various aspects of the Creation Engine,
...and unless you are willing to test every type of mesh with those screw sizes ie (FACE Data and other) and give a use-able list of what HAS to be 1024 or 2048 with the HD texture pack, vs the 1k normal/spec maps. unless you like the black face bug, and other issues.
Ignore his comment it is unsubstantiated. 95% of meshes do not require base 2 textures. Generally speaking all textures within a given category either all work or all fail. He is talking out of his arse premium user or not. I know that face and body textures often require base 2 textures, but armor and almost everything else does not. Do not speak of what you do not know 100%.
Regarding Gambit77,
I'll accept the criticism that new standards should be mentioned and discussed. Keep in mind, there is a large population that uses relatively older video cards, specifically 1GB video cards. I should have probably mentioned that that is one of the main target audience that this guide is written for. Obviously, if you have 4GB or more VRAM new standards make more sense; you are 100% correct about that. However, I should mention that merely Macro-Managing (using new standards) is poor practice; yes you are maximizing visual fidelity, but at what cost. My guide is about Micro-managing; that is the key. The part of your criticism I disagree with is the severity of artifacting with regard to diffuse and specular maps. Artifacting is only visible @ 1024 and below. Example: A 4K diffuse DTX1 vs a 4K diffuse BC7 at full resolution will contain no discernible difference. Where BC7 would shine is with mipmaps @ 1024 and below, that's when you will see an advantage. The ideal is its cheaper (vram) and (superior quality) to keep the resolution above the artifact threshold than it is to use the new standard. So a 25%-50% i.e., 1536 increase in res using old standard is better than 1024 res of the new standard.
Don't know 100%? Try running your game with those screwed up sizes in packed in a BA2 with out missing textures, or CTD. Good luck. I suggest you do more research.
By ensuring the texture dimensions are a power of two, the graphics pipeline can take advantage of optimizations related to efficiencies in working with powers of two. For example, it can be (and absolutely was several years back before we had dedicated GPUs and extremely clever optimizing compilers) faster to divide and multiply by powers of two. Working in powers of two also simplified operations within the pipeline, such as computation and usage of mipmaps (a number that is a power of two will always divide evenly in half, which means you don't have to deal with scenarios where you must round your mipmap dimensions up or down).
Textures are not always square nor are they always powers of two. The reason why they tend to be powers of two is usually to increase compatibility with older video cards that imposed that restriction. As for non-square textures, that's not usually a problem. To summarize:.
Textures should be powers of two because many older graphical cards still require it. The reason for that restriction was so that they could perform some optimizations to the texture mapping operations. Most modern graphic cards don't have this restriction either.
As you have stated you are rebuilding these for a potato rig 1GB VRAM. The non ^2 textures are not ideal.
When I started to learn modding, information which provides new users or modders incorrect data was one of the things that wasted so much of my time and made it that much harder for me to learn the right way to do things. It took me a long time to learn the correct way to generate and save textures files, and I have now made hundreds of custom textures for F4.
In the interest of not making things harder for other people, it would be very helpful if you could please clearly state at the beginning of your comments that: 1. These are your personal suggestions, which do not match the correct standards which Bethesda established for this game. 2. Your suggestions only apply to optimizing textures for a very low end computer.
Finally, for anyone reading these comments, please be aware that saving normal and specular maps in any format other than BC5, will cause problems.
So what would you suggest for skin texture to obtain very realistic appearances?
Example would be how some people will post shots from Daz poser and the textures will be hyper realistic. Any way to obtain that level of detail in fallout?
If you want people to look any more realistic, buy Far Cry 5, the details on the people's faces in the trailer, assuming it was in game graphics and not cut scenes, was SICK.
We're just not going to have that kind of realism in Fallout 4, lol.
17 comments
dxt1 and dxt5 are lower quality but much faster to decrompress compared to newer bc7 dx11 format available for FO4 and SSE. if bc7 hardware support is implemented this may be not relevant.
one thing in the description is wrong: the recommendation to use texture sizes like 1536 and 2560. this is misleading. it may work but it will slowdown graphics or cause incompatibilities because standard decrompression format and mipmaps use sizes of 128, 256, 512, 1024, 2048, 4096p and so on.
normal map textures suffer the most from compression artifacts due to the included directional info that will be distorted and i'm not shure if compression of this type is always better than downsizing and saving uncompressed. my eye normal map mod for skyrim is a good example of the negative impact of a low res vanilla texture showing its dxt1 compression artifacts in game and how it can be cured.
bc7 seems to be a good compromise and very close to the original. i would recommend to use bc7 for games like fo4 or sse supporting this format nad if high res color is essential (eg. skin, skin gloss). the paint.net bc7 plugin works nice and is for free. bc5 was designed for normal maps but i'm not shure if it works flawlessly in all cases and with all games. if you are not shure and you need best quality (eg. for body, face) you can use bc7 also for normal maps.
To start with, your Criticisms of kkthebeast and Gambit77 are uncalled for, especially saying that kkthebeast doesn't know what hes talking about regarding meshes and texture resolution. I do allot of work in the background with the Raider Overhaul / Super Mutant Redux team, and that work includes extensive testing of various texture tricks and odd sizes for textures. By definition BC or Block Compression requires resolutions divisible by 4 in order to be able to create mipmaps, however if a texture doesn't have a height and width divisible by 8 at a minimum, it WILL make the game CTD, 100% Guaranteed. Height and Width needs to be divisible by at least 8 for the creation engine to function. Additionally, packaging textures that are not standard sizes into BA2 files is a total crapshoot, Ive done it, and it works for some people, but not for others, the cause of this is unknown, and they will work if unpacked, however loose files are their whole own set of issues for performance etc.
Regarding your comments about this only being relevant for people with low amounts of vram, again i have to disagree, also i highly doubt that any videocard with 1GB of VRAM can ever run the game smoothly or without swapping textures in and out of memory constantly, realistically FO4 needs at least 2GB to have any shot at smoothness at even the lowest setting, to add to that I can easily make FO4 use all 8GB of memory in my 1070 and still be running out. So vram is an issue for every FO4 player, not just those with low VRAM. and any amount you manage to free up, it will find a way to use.
for reference:
vanilla game in Lexington: https://gfycat.com/CreativeTenderCrocodileskink
same area with the HDTP activated: https://gfycat.com/PartialGiganticArmyworm
Next thing, DXTn / ATI1 / ATI2 should be being used for NOTHING anymore, they all have BC equivalents now and DXTn was deprecated 8 years ago when we got Windows 7. To list them off...
DXT1/DXT2 = BC1
DXT3/DXT4 = BC2
DXT4/DXT5 = BC3
ATI1 = BC4
ATI2 = BC5
To add to that, BC formats use a new extended header in the DDS file which allowed for more formats then the original four character code formatting allowed, anything that uses DX11 knows how to read the extended header, and by extension, knows how to use BC4/5/6/7.
BC6 and BC7 are new with DX11 and are far superior to BC1/2/3, which BC7 replaces. BC6 is irrelevant for this conversation because it is used only for HDR textures.
now, BC7 allows for up to 7 different methods of compressing each 4x4 block of the texture, allowing it to select the best method for each block versus being forced to apply the same method across the entire texture, this leads to massive quality gains regardless of texture resolution. You can read about those modes on MSDN here:
https://msdn.microsoft.com/en-us/library/windows/desktop/hh308954(v=vs.85).aspx
As you can see here, BC7 does an amazing job with changes in color.
What you see on the left is an enlarged copy of a 24 bit RGB test pattern, with gradients, on the right the same thing converted to BC7, there is some difference but the images look nearly identical, now, if we look at the same image converted to BC1 or 3....
We see a significant amount of chunkyness in both images, in addition we also see that BC1 / 2 / 3 all produce identical color output since the increase in size is only because of the increase in Alpha Bits, something that BC7 varies on a per block basis as needed, allowing it to use the unused alpha bits for additional color definition
Now, you really, really seem to like BC1 for some reason, and think nothing should use BC5/7 even though its the same size as BC3.
With the availability of BC7, there is literally NO reason to use BC3 at all, it has a lower quality at the same file size. In addition to that, i argue there is no reason to even use BC1 regardless of the memory savings, the memory savings you are looking for can be easily achieved by resizing some of the existing BC5 Normal and Specular maps to less insane sizes.
In addition to that, just making a blanket statement that artifacting isn't noticeable until textures are below 1k is totally wrong, it depends entirely on the SIZE of the item the texture is applied to, for instance, deathclaws are basically 1 massive texture over their whole body and more or less need a 4K diffuse and normal map to look good, even at 4k some artifacts are visible simply due to the massive amount of surface area that texture is applied to. Meanwhile in the HD texture pack Bethesda made every ammo box model 4K for diffuse normal and specular maps, even blown up on a screen its near impossible to tell the difference between 4k and 2k, in game even 1.5k would look just as good, however the normal map could be tiny, like 512x because its just a box, and about 1k for the specular to prevent loss in the map detail. Similar thing as the deathclaw example goes for allot of walls and large objects, the more surface area something has, the larger the texture needs to be to keep quality.
Another note about textures, Normal and specular maps are BC5 for a reason, BC5 allows for 2 channels to be made at 8 bits / channel instead of 4 bits / color channel, allowing for a much higher quality to be achieved then would be possible even with BC7 at a given resolution, converting them to BC1 in the name of memory savings causes a even larger quality loss then doing the same for a diffuse texture.
for reference Ive included the PNG test pattern so you can run your own tests, just in-case you don't believe me.
RGB Test Pattern
Here is also a Comparison table noting some of the differences between the BC formats.
Anything saved with DXTn formatting or ATI1/ATI2 is using the older DDS Header which is now deprecated in favor of details saved in the Extended header, introduced with DX10. Sadly the nearly all photo editors only support the older formats, which is also why BC7 files fail to open in nearly all editors.
and some other reading on the topic.
http://www.reedbeta.com/blog/understanding-bcn-texture-compression-formats/
As far as your crusade for vram savings goes, i would take a good look at how important each map is to a certain model, and fine tune each one to get the balance you want, you can use any size that is divisible by 8, however you may have issues packing the texture into a BA2 unless its using traditional sizes that are n2 based.
In closing
- a guy who has spent 100's of hours working with the DDS format.
For the most part, I was and continue to be in agreement. And just so you know, all my textures have always been divisible by 8, I just felt divisible by 4 was too limiting especially with regard to either best performance/visual by byte or best visual/performance per byte which is which is why I was recommending resolutions that were divisible by 8 without explicitly stating that they had to be divisible by 8.
In any event, I have altered the direction of the project to more so initially only deal with specular maps and normal maps since these were the primary improvements in the high resolution texture pack. Both BC3 and BC7 deliver on smooth gradients with latter being twice as good, but to say that smooth gradients are always necessary is complete rubbish. So for diffuse maps we have BC1 when smooth gradients are not required and BC7 when they are required; do anything else and you are wasting precious vram. As for normal maps, it's safer to use 2048 BC7, but sometimes 2560 BC1 will look almost just as good and will use less memory (most be using the correct software). It's slightly more complicated for specular maps, if they are highly detailed BC7 will win each and every time, however if its a "full light/darkness or full light/darkness with barely visible detail (faded detail)" in that case BC1 is best bang for buck because the specular map is crap to begin with. Loose files are best because nexus mod manager is junk, I have noticed no anomaly with BA2, just with nexus mod manager. I will update guide in several days to be much more precise. I'll just point out that kkthebeast could not understand that: The game engine supports textures with resolutions divisible by 8 and that software can create and attach mipmaps to these divisble by 8 textures. He was 100% certain that could not be the case, and he was 100% wrong. When he couldn't see why he was wrong he spoke out of term and for that reason I said what I said. An intelligent person thinks before they reply, and he did not.
They come in BC3 by default and any attempt at using BC7 instead immediately causes the brown face bug over here.
test pattern uncompressed
test pattern compressed in dxt5/dxt1
test pattern uncompressed side by side with test pattern compressed in bc7
...and unless you are willing to test every type of mesh with those screw sizes ie (FACE Data and other) and give a use-able list of what HAS to be 1024 or 2048 with the HD texture pack, vs the 1k normal/spec maps. unless you like the black face bug, and other issues.
Ignore his comment it is unsubstantiated. 95% of meshes do not require base 2 textures. Generally speaking all textures within a given category either all work or all fail. He is talking out of his arse premium user or not. I know that face and body textures often require base 2 textures, but armor and almost everything else does not. Do not speak of what you do not know 100%.
Regarding Gambit77,
I'll accept the criticism that new standards should be mentioned and discussed. Keep in mind, there is a large population that uses relatively older video cards, specifically 1GB video cards. I should have probably mentioned that that is one of the main target audience that this guide is written for. Obviously, if you have 4GB or more VRAM new standards make more sense; you are 100% correct about that. However, I should mention that merely Macro-Managing (using new standards) is poor practice; yes you are maximizing visual fidelity, but at what cost. My guide is about Micro-managing; that is the key. The part of your criticism I disagree with is the severity of artifacting with regard to diffuse and specular maps. Artifacting is only visible @ 1024 and below. Example: A 4K diffuse DTX1 vs a 4K diffuse BC7 at full resolution will contain no discernible difference. Where BC7 would shine is with mipmaps @ 1024 and below, that's when you will see an advantage. The ideal is its cheaper (vram) and (superior quality) to keep the resolution above the artifact threshold than it is to use the new standard. So a 25%-50% i.e., 1536 increase in res using old standard is better than 1024 res of the new standard.
By ensuring the texture dimensions are a power of two, the graphics pipeline can take advantage of optimizations related to efficiencies in working with powers of two. For example, it can be (and absolutely was several years back before we had dedicated GPUs and extremely clever optimizing compilers) faster to divide and multiply by powers of two. Working in powers of two also simplified operations within the pipeline, such as computation and usage of mipmaps (a number that is a power of two will always divide evenly in half, which means you don't have to deal with scenarios where you must round your mipmap dimensions up or down).
Textures are not always square nor are they always powers of two. The reason why they tend to be powers of two is usually to increase compatibility with older video cards that imposed that restriction. As for non-square textures, that's not usually a problem. To summarize:.
Textures should be powers of two because many older graphical cards still require it. The reason for that restriction was so that they could perform some optimizations to the texture mapping operations. Most modern graphic cards don't have this restriction either.
As you have stated you are rebuilding these for a potato rig 1GB VRAM. The non ^2 textures are not ideal.
In the interest of not making things harder for other people, it would be very helpful if you could please clearly state at the beginning of your comments that:
1. These are your personal suggestions, which do not match the correct standards which Bethesda established for this game.
2. Your suggestions only apply to optimizing textures for a very low end computer.
Finally, for anyone reading these comments, please be aware that saving normal and specular maps in any format other than BC5, will cause problems.
Example would be how some people will post shots from Daz poser and the textures will be hyper realistic. Any way to obtain that level of detail in fallout?
CBBE for CBBE compatibility.
Wonderbody for simplicity; it just works.
We're just not going to have that kind of realism in Fallout 4, lol.