which resolution is your normal map?
the normal map for this weapon in particular is 2048x2048 (as are all the other maps as well)
My personal rule of thumb is to bake everything 2x the size i want it to be in the end (im aiming for 1024x1024 maps for the 92fs) so when i downsize it, i get a nice AA effect.
how high is your edge padding value?
edge padding? wat
edit: polycount wiki has told me what edge padding is. I would assume my edge padding is at default values, since i havent touched it. I usually do my own edge padding inside photoshop once my textures are finished.
oh and another thing mildly related to this subject; i suggest you disable mip mapping when exporting your first person weapon textures. First person weapons are always near the camera, so mip mapping wont save any resources, and in personal experience ive found mip mapped textures blur annoyingly when viewing your models surface at an angle (and it destroys all the precious detail)
do u use smoothing groups? and if yes how do u overcome the seems?
yes i use smoothing groups. My workflow is to first create the UV map, then use the textools script to create SG from UV shells (i used to do it by hand but the script does it in 1 click so its awesome). It also has other awesome tools like the shift overlap feature which is awesome when you have mirrored mesh on your model sharing the same UV coordinates.
There are 3 rules to seam placing. The first one is easy: split at 90* angles. I usually split at a bit less just for good measure.
the second rule is almost as easy. As you probably know, your model is 3 dimensional. Uv maps are 2 dimensional. When you try to unwrap a 3 dimensional surface on to a 2 dimensional plane, you will get stretching and warping. the amount of stretching and warping that i allow on a surface depends on how visible the surface is and what i intend to texture on it. If i want to have some text or a recognizable pattern there, i dont want any stretching on the UV. If the part will be covered by just a base texture, some stretching is fine. So basically you will have to figure out a way to split your UV's so you dont get too much stretching.
alright, then theres rule 3. This one is a bit more complicated.
Ill start by explaining how normals work. lets take an example: 2 square planes joined at a 40 degree angle and sharing a single smoothing group. The smoothing group will make the surface look round, even though in reality it is 2 flat surfaces. from now on i will refer to this as the unmodified normal surface. When you place a normal map on top of this unmodified normal surface, the normal map will try to "fix" it with the values stored in each pixel of the map. So basically, each pixel in the map will grab the unmodified surface normal of its spot, and modify it with the RGB value the pixel holds (the RGB value is converted to a normal vector but i wont go in to that too much) Basically this whole thing works sort of like the blending modes in photoshop. It takes the underlying layer (unmodified normal surface) and adds to it the values from the overlaid layer (normal map). If you wanted to make the surface flat again, without splitting the smoothing groups, the normal map would have to have the same round surface as the unmodified normals except mirrored, so when the two are combined, they would cancel each other out.
If you understand that, congratulations.
When you're placing your smoothing groups, keep an eye on the unmodified surface normals. They should be smooth. The smoother they are, the better your normal maps will be able to cope with them. If there are sharp edges in your unmodified surface normals, the pixel density of your normal map will not be enough to smooth it out, and you will get artifacts in your end product.
heres one last example:
the object is quite simple. the angle of the edges is very blunt, and it could easily be mapped to a single SG/UVshell without any stretching.

but as you can see, if you place a single SG on the whole object, the unmodified surface normals become very sharp. This is due to the abnormal topology.
The fix is very easy. you just split the smoothing group, and voilá

this example is a bit exaggerated, but it demonstrates the point nicely.
thats rule 3.
do u put everything in one normal map? or do u have extra maps for mag, sights, attachments?
It depends. I usually map objects that are unique to the current weapon in the same texture sheet. This means that if the magazine is unique to the weapon, and wont be used by any other weapon, it can share the normal map. Sights on the other hand are usually shared between weapons (atleast in crysis) so its worth having a seperate texture sheet for them.
Depending on the quality of the asset, i sometimes split my models in to seperate sub-materials by surface type. This lets me tweak the shader settings on a per-surfacetype basis. I always have a seperate texture sheet for each sub-material unless there is a very good reason to share the texture sheet between submats..