Thinking about improving FPS, a path to optimizing?

yeah the radical hasn't been fixed yet..

 

Hi,

I didn't follow this. What is the problem with the radical ?

Thank's

 

A new slider on UI, to dictate 4k skins and an even more detailed lod_0, called Ultra. If Full is selected, it loads 2k skins and the new lod_1. If Ultra is selected, it loads 4k skins and the new lod_0.

 

Check your player.json for texture quality overrides... has nothing to do with LODs. Textures use MIPs for distance resolution.

 

When I check DX11 back-to-back with DX9 with the same set and amount of cars, in the same weather conditions and at the same TOD it gets quite obvious that it isn't nessecarily a problem of higher detailed models and textures but how the core engine is processing that content. In DX9 the content runs smooth, in DX11 it is running out of realtime and producing a slideshow, no matter how low I set the graphic options wich makes the graphic options themself pretty much redundant in DX11 for me, atleast with the newest content. So i would rather hope that they focus on the core engine instead of downgrading beautifully crafted content that still needs to look good for VR. My 2 cents

 

About my thoughts on LOD:

Can you explain a bit more? What I see on screen does not show LOD distances evolve if I change the opponent detail. The polygons and LOD switching distance do not appear to change, but the textures will become more blurry (cockpit view).

A good car to use as an example is the bugged LOD for the Radical GT3 as an opponent, where you can see the exterior mirror go transparent when it switches to a lower LOD. That distance never changes no matter what opponent detail I select, thus my conclusion that LOD switching distance is not affected by opponent detail.

 

Internal cockpits usually don't have any LODs, as you will always only see ONE cockpit - the one in which car you are sitting in. Textures can be affected by settings though, as they are primarily there to save memory.

Each cars external parts typically have defined details for each setting (chosen seperately for player and opponent cars) in the .GEN file, <LOW>, <MED>, <HIGH>, <MAX> - these activates the desired LOD-GMTs and uses different LODIn/LODOut-values.

Texture detail is also affected by these settings AND by distances in general per material, and will offset the chosen MIP - each MIP will be a quarter resolution of it's parrent. Say the high res is 2048x2048, the next MIP will be 1024x1024, then 512x512 and so forth...

I cannot tell you if there is a bug in the Radical, as these files are encrypted, but I know for a fact that LODs are being used for both driver and opponent cars.

There are some settings in the Player.JSON to play with, controling texture quality, independant from the LOD-details.

"Opponent Texture Override":-1,
"Opponent Texture Override#":"For opponents' vehicle textures: -1=use Opponent Detail, 0-3=override value",

"Player Texture Override":-1,
"Player Texture Override#":"For player's vehicle textures: -1=use Player Detail, 0-3=override value",

 

Thanks, WiZPER!

I'm looking at an ISI gen file and see that the LODs are explicitly set. So, for instance, the medium detail GMT LODIn is at 20 meters for the HIGH and MED levels, which means one wouldn't notice any difference between those two levels. It wouldn't be until LOW level is selected that the medium detail LODIn changes to 4 meters.

So it's starting look like if S397 tweaks the GEN file on their releases we'd get better fps options when selecting opponent detail. Better in the sense that player selections would have more impact than they currently do.

 

well mostly the horrid ffb and the fps hungry models. I'm on a 1080 and dropping a crap load of frames.

 

I tried very quickly the radical. FFB is very different than other ISI cars. I have to test again. When I have to test FPS I use ac cobra bevause there is whith it a huge cost of performance. If fps is ok with ac cobra, low sun : settings are ok

 

Ok, thanks. Interpolating from my results with GTX 1050 Ti, you should easily be able to maintain 75+ FPS at start of race with 32 cars on your GPU. As it's not the case, it means performance doesn't scale with GPU power and that there is a bottleneck. So LOD tweaking might be of limited benefit.

 

Are you running custom in-game graphical settings and if so, what are you running rF2 at?

 

I made a new test with a fresh profile now. I can't bother to run with maxed settings as there are nonsensical things like soft particles that does nothing other than cost FPS.

Matsusaka 20 GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX off/low: 55 FPS
Matsusaka 20 GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 49 FPS

According to chart below, which shows average from 20 games of different GPUs, a GTX 1070 should give 219% the performance of GTX 1050 Ti. So you should be getting a minimum FPS of over 100 on Matsusaka with 20 cars on default settings.

 

The issue is certainly with rF2 and how it works. As I've stated above, R3E, AC and pCARS 2 all run well over 60fps in all conditions & at max settings on my PC. AC benchmarks at 120+fps at max settings and runs a full grid of cars without any issues at all. pCARS 2 runs a 31 AI grid of GT3 cars at Monza in the dry at an average of 109fps (again, max settings).

One other factor could be the OS; I have Win10 Pro x64 with the latest "creators" update. This update broke the AA in AMS, requiring users to run in a windowed mode for any AA to work and I'm wondering if something funky is going on with rF2.

 

Just a couple of years ago, my vid card would not even be on that list. I upgraded as much as I could for rF2. Still closer to the low score than the high scores.

 

R3E and AC, does'nt HAVE "all conditions" - rf2 is a dynamic graphics engine, there are reason why other titles choose static conditions.

 

I understand pCARS and rF2 have dynamic weather.

The fact is the GT3 cars run poorly compared to the other ISI/S397 made cars and the URD & ASR paid for mods. The following fps is for the player at the rear of the grid.

Matsusaka 20 C6R's, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 186 FPS
Matsusaka 20 URD T5 cars w/2016 skins, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 156 FPS
Matsusaka 20 ASR 3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 145 FPS
Matsusaka 20 Nissan GT500 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 135 FPS
Matsusaka 20 Radical RSX cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 120 FPS
Matsusaka 20 Merc GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 87 FPS

How this is fixed is up to S397. I love the GT3 cars in rF2 and would like to see performance akin to the other cars available in the sim. I understand there are more polygons in the GT3 cars and the models are the same as AC but the performance isn't close to the other rF2 cars.

 

The fact is also, that GT3s use brand new shader technologies with visual body damage - and maybe other awaiting features.

 

Thats the point. Some people think, they have enough fps so there is no problem but the performance decrease compared to other content is not acceptable for me because the whole engine is poor performing compared to other and that combinend with such kind of heavy load content is the kill for many people

 

They are promised further graphics enhancements. Without any change to the current performance the performance will be even lower compared to the performance now.

Travis got 87 FPS on his benchmark. Now if they won't improve the performance in the upcoming soontm rain build. Since they are patching in real time reflections on the puddles and whoever knows what else. That number will be even lower.

I believe the new visual body damage has nothing to do with the performance on the grid since the cars aren't damaged.