Optimisation (aka: Poly gone!)
I have looked carefully around the scene and reduced the poly count of many of the items in the scene.
- Telegraph wire lofts (each): 2400 faces to 428 (this is still high but allows for a smooth curve in the wires).
- Scaffolding frame cylinders (each): 216 faces to 20.
- Corrugated paneling (each): 8784 faces to 8.
The scene has gone from over 900,000 faces to only 130,000 without the replacement paneling in place, but luckily this won’t take it back up to what it was.
I think that these are great. Getting the materials to align was easier than I thought but getting the curves on the edges to look even half decent took a few hours of experimenting in Photoshop.
Replacement window structures
The other item that was too high on the poly count was the window structures. Each one was made of individual cylinders. This new one is far lower in polys and was made using splines and extrusion.
This is a render of the one that I am going to use. I initially forgot about the quick slicing of the Jambs, Transoms and Mullions which was creating stretched effects on those parts which shows as extra unwanted shading. I seem to be unable to create a window frame that doesn’t have irregular normals. Whilst the AO material overide is on you can’t see the problems, but under normal conditions the sides and top outer faces are transparent which indicates that their Normals are the wrong way around. I have tried to reverse them by selecting each polygon but not only do I seem to end up with a mess, I get the feeling that this shouldn’t be happening at all. I am going to ask my supervisor about sorting this out. I have a fix for the time being: Force 2-sided in the Render options gets rid of the problem completely which is ok for this project. In future when I might be designing for a game where the polygons will be moved around in realtime this isn’t an appropriate solution as I imagine it doubles the numbers of polys in some situations leaving the graphics card / renderer with the work load.
Replacement corrugated panels
I have tested two low poly methods of making corrugated panels. The two are reasonable in their appearance. I have chosen to use the single plane (left) as it looks more convincing than having a visable join of the other method (right). The only thing that I will have to remember with this is that the other side of this panel is transparent. This won’t be too much of a bother – wherever this issue arises I will clone a duplicate and spin it around 180degrees and use it as the reverse side.
8000 faces
This is to log todays tutorial with my supervisor.
I took my model in as I was having severe difficulties keeping max open with a file with close to a million polys in it. Quite why it didn’t occur to me that this might be the problem is embarrasing. I think I thought that Max would be able to handle several million polys. Some of the really good models I have seen sure do look like they have that many in them. Maybe they do. However at the meeting it was made plain to me that 1 million is the upper limit. My computer agreed. When I got home I deleted the offending parts of the scene to find that it would let me work with it again instead of kicking me out of Max back to the desktop. I looked at one of the corrugated roofing panels and it had over 8000 polys in it. With altered duplicates of this element all over the scene – no wonder Max kept giving up.
I am going to make better roof panels and window cages. The telegraph line lofts are already sorted out (2400 faces to 428).
I was shown a technique whereby the items in front of the camera can be made to face the camera without relocating (by rotation) called Orientation Constraint. I tried it this evening and its easy to set up and will be useful for creating background “flats” for my scene. It will save me having to rotate or reposition the background as I move the camera around.
I need to start using layers to help me more easily hide and show scene elements for making composites.
Using Photoshop for post production is standard, so I can use that for Depth of field effects instead of the Camera DOF in Max.
Now that I have removed the corrugated sheeting the file has lost 3/4 of a million faces.
New workstation
I have recently bought a two year old Dell from a friend. After upgrading virtually all of it apart from the Harddisk and Processor I’m now running the following system:
Pentium Core 2 Duo 64bit
ASUS P45 chipset motherboard
8 gigs of Corsair Heatsinked DDR2 RAM
500gb HD Sata
NVidia GeForce 9600 GSO 1.5 gigs of GRAM
DirectDraw 9 Pixel shader 4 (whatever they do!)
92 Graphics cores
Apple keyboard and Logitech G9 mouse
Windows XP Professional 64bit Supreme Version
… and what a difference this has made! everything renders! Even the huge scene that brought the laptop I was using to a grinding and soul destroying Mental Ray and then Max shutdown. Its early days, but XP64 is SO SO SO much more efficient with RAM. It looks like I will be able to render A3 300dpi images now which is what I have been wanting to do all along.
After using Vista for a while XP64 is a dream.
Material Scales / Memory Use
I’ve been wanting to get my head straight about materials. Knowing that they need to be square and in binary incremental sizes eg 64×64 / 128×128 / 256×256 / 512×512 / 1024×1024 2048×2048 etc. I decided to do an experiment with a soft drink can that I’m currently using. I knew that this object would be small in the scene so I made the material 512×512 instead of 1024×1024 or 2048×2048 that I have previously been using.
As you can see from the picture I began with the 512 map and reduced it in increments, initially without reloading the image which is why the first 4 on the top row (from the left) get increasingly worse in quality. Realising that this was happening I reopened the original file and scaled it to 64×64 with the Nearest Neighbour Interpolation switched on so that the image would retain its sharpness at the smaller size, this it did but as you can see it also made the map blocky. So finally I chose to reopen the original file and make a 128×128 map using BiCubic Sharper Interpolation, this has resulted in an image quality closer to the 256×256 image size but with half the data file size, so less memory used at rendertime. This is an important find when you consider that I am using CM as the units in this file and can measures 62x67metres! So I can transfer this knowledge to buildings and objects that should be this large.
In conclusion, my question about whether the size of an object in max bares any relationship to the print size of the materials applied to it has been answered: No, not at all! In Max you can use a first class stamp to material the side of a skyscraper. It could be argued that I still don’t understand whats actually going on, however frankly, knowing that I can save memory without losing render quality completely makes up for any fracture in understanding. (As I am regularly reminded I am doing an MA not an MSc.)
1 hour very well spent!
I welcome all comments on this post especially tips on saving memory when creating high poly count scenes.
Memory issue solved
After going through all the materials used in the most recent file that I thought was corrupt and altering them to either 1024² or 2048² as appropriate; the scene renders like a dream. It looks like buying the additional RAM was still a good idea, but at least the file actually renders instead of crashing out.
Lesson learnt.
The memory chips have arrived…
… but they haven’t made too much difference as it seems that my files are as corrupt as the mafia!
Mental Ray is not able to allocate enough storage to render the scene. Max is moaning about not being able to allocate memory to random materials that have been used.
I’m going to composite different scenes together, but even that seems a long way off when I can’t seem to get a basic scene to render.
This is all after installing 4gb ! Thats what annoys me the most. It seems like installing more memory has just given the computer more rope to string itself up by. <foul language> Is this Vista? </foul language>
64K RAM SYSTEM 38911 BASIC BYTES FREE READY…
4Gb of RAM is on its way to me right now. No more running out of memory with Mental Ray or complete closures of 3D Max.
(Well, we’ll see.)
Maxed out on memory
I’ve been fighting with 2GB of RAM getting this scene to render. I’ve figured out that I had a lot of rubbish running in the background and I can render at the moment but it feels like this laptop is on borrowed time.
My laptop is on a pair of cooling fans too. I’ve not noticed any difference when using max.
Heading towards 5% sky, at least at this angle. I’m trying to add interesting bits to the scene to. It seems to be working. I’ve quite impressed myself. This is 5-6 hours work from the previous stage to this. Thats having ignored the three hours of work this morning where the central building was built using using Extrude and Bevel and I ended up with the top of the building with inverted Normals !! How I have no idea! but I was watching time go by with frustration so I deleted it and approached the build in separate layers rather than as one whole poly.
9/9/9 Tutorial
Points discussed during the tutorial include:
- Camera placement for scene composition should aim to include Tertiary, Secondary and the Primary object of focus.
- I should grid my scene by the square inch (roughly) and ensure that there is valuable content of interest in each square. I need to aim for 5% sky too !
- I should refrain from overworking the structure and use a simpler and quicker method of creating the shell of the buildings that I want in the scene (see Gallery images).
- I should use the repetition model of the hotel with some kind of separation device (eg a pathway and fence) in front of it to create the perception of the perimeter of the hotel grounds.
- In the scene the left will focus predominately on the structure the right can focus predominately on occupation.
- To fin the mr Rendering error (see previous post) I can either remake the electric lines and use the Generate Mapping Coordinates tick box or use the Normals modifier to unify the normals in the loft.
- Ideas for occupation: Trellis, plants, motorcycle etc. Each flat has a different person in it! Make it look that way!
- Remember to make an ambient occlusion composite of the scene just before adding the people in PS.
- When making windows, delete the polys to create the gap and then use the Border poly editing tool to select the edges of the window and then use Bridge to create the window cill rather than extruding the inside through the gap and then usiing Vertex Snap to align the verts in to place.
- Realism is for documentary images where accuracy is important.
- Believeability is for pushing the boundary of what people will consider to be Dramatic and impressive.
- “That must have taken YOU ages” vs “Thats fantastic! Look at how the ….” The image has presence and demands to be noticed in its own right. (eg without reference to the artist.)
- I should bring my Balchelard work together with my Modelling. (Its currently miles apart!!!) (damn!)
- Remember to use “Interpolation steps ” when creating bezier wall lines for extruding. This prevents blocky ambient occlusion too.
I was set an exercise to create “crazy spaces” these are the results. I have found it more fun to create exterior spaces rather than interior as you can see.
During the tutorial we briefly discussed the issue that I am modelling exterior sspace yet reading Balchelard who wrote about interior space. I need to write, justify and illustrate what I understand about Interior vs Exterior (good title) prior to the presentation for this section.
These are the spaces I created during the exercise. Each one shows the shape first and then a couple of views from the inside. All with AO as the overiding material so that I can see the corners more dramatically / effectively.
Number 13 is the best composition in terms of the camera angle and interest available in the scene. Ambient Occusion creates a fantastic effect!
