ContextCaputre - What is the benefit in creating big tiles?

Question

Hi All 
 I was doing a small time benchmark to see if there is any benefit regarding time using big tiles during production: 
 
 Block size 13.4 MegePixel 
 all settings set to defaults. 
 System setup - GTX980TI with 64 GB ram. 
 3mx creation (both for the exact same area) 
 One big tile: size 53GB >>>> initial production time: 5h 30m. >>> convert to other CRS >> 28min 
 Many small tiles - 10GB>>>> initial production time: 3h . >>> convert to other CRS >> 18min 
 
 So regarding time, it seems small tiles are better. 
 is there any benefit in big tiles? 
 
 Thanks 
 Ori

Casey Bentley · Answer

Big is usually better than small, up until a point. There a couple things at play here. First, when you do lots of small tiles you end up repeating certain calculations which will take more time. However, when working with larger tiles, you will not repeat those calculations as many times. 
 
Second, the global level of detail that Pascal is referring to is the scalable functionality. Basically when you have one large tile, you are loading one large pyramid, which gives you one small low resolution node at the top to load. When you have lots of tiles, you will have lots of nodes at the top of each tiles pyramid to load. This will take longer than just loading one large pyramid. Furthermore, these nodes might be larger than if you just had 1 large pyramid which will give you longer loading times. This can be somewhat negated using the unique mesh export, however for this to work, the tile size must be uniform (meaning you cannot use adaptive tiling). 
 
Finally, if a large tile were to fail, it is going to make much more time to reprocess than a small tile. Unless there was a specific reason, I would avoid tiles that large. I tend to use somewhere in the 16-30 gb range.