So I decided to go back and make a larger version of the planet image I created previously for use as a desktop. Basically that involved just resizing an slightly recomposing the scene and rendering it. To add a bit of a half-assed atmospheric effect, I added an Outer Glow to the transparent planet layer. I think this still might have some value in combination with a good set of cloud and atmosphere spheres in Blender.
Next, I decided to create a flat map-style image suitable for gis tools and the like. The Cycles renderer is not terribly fast on my computer. The panoramic camera only makes the situation worse. So I made some modifications to the settings to try to reduce rendering time considerably. Hopefully without compromising the quality of the map image.
I have zipped up all(or most) of the blend files I used on this here
I set the resolution to 2048×1024, but any 2:1 aspect ratio will work(the X and Y values labelled as Aspect Ratio control the shape of the individual pixels; 1:1 square pixels are perfect). Output is an 8-bit RGB PNG with no compression. Samples were reduced to only 32 with no speckling I could see. It could probably be further reduced. I went with Direct Light for light paths. I’ve limited Light bounces to one for diffuse and glossy and zero for transmission, volume and transparency. Viewport is Static BVH for “faster render”, but I don’t think that will effect rendering outside of the viewport. Under Acceleration structure, Use Spatial Splits is checked for longer build time but faster renders. The build time at the start seems pretty short in any case, so it shouldn’t cause problems and in any case this should make larger resolutions a lot quicker.
I also made some changes to the node graph for shading.First, I disconnected the Bump node. It had no visible effect with the light coming from right in the center where the camera also resided. I also tried to take out the Glossy BSDF shader, but although the specular effects weren’t visible I found the water areas to be too dark.
The comparison with my previous PlanetCell map render is not perfect as the shader tree for PlanetCell was quite complex, and I was using a lot of OSL script nodes, which slow things down, but my previous PlanetCell renders took… HOOOUUURRRSSS. No. Literally. They took hours. Nearly an all day render. My render for this only took 31 minutes and 42.88 seconds.
As a further experiment, I plugged the Color Ramp into a Brightness and Contrast node and connected that and the original ramp into an RGB Mix node with the Greater Than and Multiply nodes used to drive the Mix Shader node plugged into the face node. Although I’d hoped that this would result in significantly faster renders, it took 32 minutes and 53.42 seconds to render. Basically a wash.
For a third experiment, I also reduced the Glossy light bounces to zero. That… also seemed to slow things down. It took 33 minutes 28.07 seconds to render. I don’t understand that at all, but whatever.
Finally, I simply lightened up the water colors in the Ramp itself and reduced the render light samples to eight. This time the render time was 7 minutes, 35.81 seconds! I also increased the tile size to 128×128. I wonder if there is some overhead in rendering and moving tiles? Let’s test that.
No other changes. I’m just increasing the tile size to 1024×1024. That separates the rendered image into two sections, so that my dual-core CPU can handle them in parallel. Rendering time 6 minutes and 3.62 seconds. Apparently, there is some benefit to the larger tiles, but apparently additional samples are expensive. Still pretty quick and decent render.
Now, we try a larger render. 4096×2048 with a tile size of 2048×2048. That took 21 minutes 49.22 seconds.
I tried an 8192×4096 render with a tile size still of 2048×2048 with 5 and 8 samples. The 5 sample version took 58 minutes 10.82 seconds to render, but it was much too grainy. The 8 sample rendering took 1 hour 33 minutes and 27.97 seconds, but still seemed a bit grainy.
Finally, I rendered an 8192×4096 render at 10 samples. This took me 2 hours 1.16 seconds to render. The improvement to graininess was minimal. I think I’d stick with 8 samples in the future, because the time cost was just too high for the negligible improvement.
Next week, I’m going to Disneyland on my daughter’s Make*A*Wish trip, so I won’t be able to maintain the pace of postings I have lately. After that, I’ll get right back on the horse.