Terrain textures: acceptable quality vs. crashes

[chromebird]

Problem:

After compression terrain textures noticeably degrade in quality. while using uncompressed textures or scaled up and compressed textures both cause "out of video memory" error and crash on Windows XP 32 bit systems. Support of low-end systems is vitally important for the project since a lot of our players use them.

Description:

terrain dimensions: Grid 8193x8193, Surfaces: 32x32, Step: 1 Height 400; surface diffuse, mask and normal = 256x256; coarse diffuse and normal = 1024x1024. As shown on not_compressed.jpg and compressed.jpg there is no difference in FPS. However there there are visible difference in quality. Compression artifacts are even worse from close distance: see compressed_tex.jpg and uncompressed_tex.jpg .

 

Uncompressed textures caused directX9 "out of video memory" error on following systems (both ATI and Nvidia videocards):

CPU: Windows XP (build 2600, Service Pack 2) 32bit

Intel® Core2 CPU 6400 @ 2.13GHz 2133MHz

GPU: NVIDIA GeForce 8600 GT 6.14.12.6099

System memory: 2047 Mb

Video memory: 512 Mb

 

Windows XP (build 2600, Service Pack 3) 32bit

CPU: Intel® Core2 CPU 6320 @ 1.86GHz 1868MHz

GPU: ATI Radeon HD 2600 PRO 6.14.10.6925

System memory: 3327 Mb

Video memory: 512 Mb

 

These are two more-less acceptable solutions out of many that we tried to fight crashes and visual quality degradation:

 

1). increased mask and diffuse image sizes to 512x512 and compressed them. image quality remained still unacceptable; game loaded but crashed almost immediately on attempt to increase the quality of materials (user controlled option) with errors in log "could not create texture..."

 

2). we did not compress textures, image size remained unchanged 256x256 but we added mip-levels. Image quality was acceptable on high graphics settings (as expected), and it was possible to play on low graphics settings that use lower mip-levels.

 

The question remains: even though 2nd solution looks successful is this solution valid?

Shouldn't we use compressed textures no matter what?

What are "hidden" possible problems that we did not "catch"?

Can anyone suggest a better solution?

[chromebird]

it looks like only I am having troubles with new terrain. it would be great if somebody could share their successful results - screenshots, some tips about possible optimizations...

[binstream]

We are to release "Valley" demo for customers within the next update - it's a nice reference of terrain usage.

[ulf.schroeter]

it would be great if somebody could share their successful results - screenshots, some tips about possible optimizations...

 

just some screenshots DISI-X terrain

[chromebird]

Thank you for the screenshots.

Ulf, may I ask what software you used to generate the height map?

[ulf.schroeter]

Ulf, may I ask what software you used to generate the height map?

 

Hi chromebird,

 

terrain heightmap was generated with TerraVista from PRESAGIS baseded on real-world DTED elevation data and NASA BlueMarble satellite imagery. While being used widely in the simulation area, TerraVista it's quite expensive and normally overkill for game terrain modelling.

 

One of the best tools for procedural terrain creation is World Machine 2. I used this tool for terrain mask generation. It takes some time to get used to procedural terrain building, but then it's really powerfull and it can be used for creation of terrain heightmap, diffuse texture and different types of mask textures (e.g. for WorldClutter, ObjectGrass) based on terrain parameters (e.g. height, slope, erosion flows, etc)