johnwhelan
Well-known member
Instinctively you may think that TRS2004 would run better than TS22. Much of the content is less detailed and the computers available were less powerful. Single core processors were not uncommon at the time so run it on a modern multicore processor and you're only using perhaps a quarter of the available CPU cores. Some modern CPUs have as many as 18 cores by the way.
The other thing to think about is the instruction length. Trainz uses the X86 64 bit instructions on the CPU, but GPU is optimised to process graphical data only initially they used an 8 bit instruction set so in crude terms could process 8 times as fast as a CPU. These days they use more complex instructions but the general principle is they can use a simpler set of instructions that can be processed much faster than a GPU can. TS12 used the GPU but did much of it's processing on the CPU. TANE uses multiple CPU cores and off loads much of the graphical processing to the GPU. TS22 optimises the code and with the same content at the same viewing distance you should get better frames per second.
Then you get into the content itself. It TRS2004 the viewing distance was limited and shadows none existent unless they were built into the mesh. In TS22 the default viewing distance is much greater so there are more objects in the scene. The problem comes with an object 14 kms away without LOD Trainz has to work just as hard as if it was ten meters away. LOD works by having a less detailed mesh to display when things are in the distance. Other changes over time include meshtable for scenery objects which offers less computer impact than xyz.lm.txt LOD method which was used on earlier versions of trainz. Build 4.6 or TS19 has some improvements to the meshtable so that will help squeeze a few more frames per second. So typically an asset that has a later build number will perform better than one with an earlier build number. In content manager you can preview the content and it will show you the polycount and the LOD levels. Avoid high polycount models. Speedtrees do have a performance impact so on a very low end machine I'd avoid them.
Basically for each scene your computer can display a certain number of polygons. Repeating an asset has a much lower impact, so two different houses will have a greater impact than two house repeated. Middleton for laptops has m;ny of these optimisations built in. It's UK based and uses the same undercarriage for all wagons, van and coaches. Just the 17_5 tops which typically are 54 or less polys vary, the underframe is around 1,300 polys but is repeated.
So running the latest version of trainz will help performance but set the viewing distance back. Also there are performance sliders that can be set to lower values. In the developer tab there is a profiler setting that can display your frames per second. Film displays at 24 fps, so 24 or above should be acceptable.
Laptops may run into overheating problems which will throttle the CPU and GPU so the real world performance maybe less than expected which is why I prefer a desktop or a refurbished workstation such as a Dell refurbished one. Middleton was designed to run on a old i5 with integrated graphics by the way.
Have fun
Cheerio John
The other thing to think about is the instruction length. Trainz uses the X86 64 bit instructions on the CPU, but GPU is optimised to process graphical data only initially they used an 8 bit instruction set so in crude terms could process 8 times as fast as a CPU. These days they use more complex instructions but the general principle is they can use a simpler set of instructions that can be processed much faster than a GPU can. TS12 used the GPU but did much of it's processing on the CPU. TANE uses multiple CPU cores and off loads much of the graphical processing to the GPU. TS22 optimises the code and with the same content at the same viewing distance you should get better frames per second.
Then you get into the content itself. It TRS2004 the viewing distance was limited and shadows none existent unless they were built into the mesh. In TS22 the default viewing distance is much greater so there are more objects in the scene. The problem comes with an object 14 kms away without LOD Trainz has to work just as hard as if it was ten meters away. LOD works by having a less detailed mesh to display when things are in the distance. Other changes over time include meshtable for scenery objects which offers less computer impact than xyz.lm.txt LOD method which was used on earlier versions of trainz. Build 4.6 or TS19 has some improvements to the meshtable so that will help squeeze a few more frames per second. So typically an asset that has a later build number will perform better than one with an earlier build number. In content manager you can preview the content and it will show you the polycount and the LOD levels. Avoid high polycount models. Speedtrees do have a performance impact so on a very low end machine I'd avoid them.
Basically for each scene your computer can display a certain number of polygons. Repeating an asset has a much lower impact, so two different houses will have a greater impact than two house repeated. Middleton for laptops has m;ny of these optimisations built in. It's UK based and uses the same undercarriage for all wagons, van and coaches. Just the 17_5 tops which typically are 54 or less polys vary, the underframe is around 1,300 polys but is repeated.
So running the latest version of trainz will help performance but set the viewing distance back. Also there are performance sliders that can be set to lower values. In the developer tab there is a profiler setting that can display your frames per second. Film displays at 24 fps, so 24 or above should be acceptable.
Laptops may run into overheating problems which will throttle the CPU and GPU so the real world performance maybe less than expected which is why I prefer a desktop or a refurbished workstation such as a Dell refurbished one. Middleton was designed to run on a old i5 with integrated graphics by the way.
Have fun
Cheerio John