Cars NFS Tournament Class B 1.6

[AlleyViper]

993 in n4S was not US version

Please go back and see the showcase and intro video again. For the showcase pics, they are a mix of both euro/us models. Cockpit camera also had a speedometer in mph, topping at 180. But then the low poly exterior model looks euroish without bumperettes on the rear, but has a narrow US plate without extra space on its sides (where a euro would, as it misses the small bumpers).
You'll find other inconsistencies in the original game showcase like the C4 ZR-1 pre/after facelift.

@Andrew_WOT if anything, it shouldn't be about mip maps (it's almost never about them, AC by itself generates them when loading textures missing them, taking a bit more time to load; many included textures on official content miss them for this reason) it should be about draw calls on VR in the main thread if you're not GPU limited. Compare DIP numbers on the render stats app (it goes up with more objects and materials on scene, taxing CPU).
If you're only using these 3 cars, the ZR1 has no lods so you'll have the same number of objects in the hundreds at all distances, and the Viper has above average objects on the last 2 lods if you go strict by the book, so a bit of performance loss is to be expected. The 911 is near Daniel's RSS work on that matter.

 

[Deleted member 197115]

911 is the worst.

 

[Deleted member 197115]

And this is information spread on CSP discord, that is mod cars no proper LODs = VR performance killer.
¯\_(ツ)_/¯

 

[AlleyViper]

Well, I can't tell you about VR, but the dip/lod importance surely true on a single screen with a greater cpu bottleneck than gpu like mine (OC'ed 8GB RX480 and FX8350@4.4), where the 911 is clearly the car with better fps on a single-make static 20 AI Barcelona grid starting from behind - model differences should prevail. Bear in mind that the 911 is also the one with more tris and objects on LOD A among the nfs class B cars, attesting to what good LOD work can do.
- 74 fps 911 C2
- 70 fps ZR1 (has no lods, but much lower tri and object count on lod a than any other cars, making it hard to compare directly; still looks top for a modest lod A on resources, would probably be on top with lods)
- 68 fps Viper (which accounts for the lods effectiveness, even if as not optimized as 911's, it still has 70% more objects on lod a and 35% more tris than the ZR1 for near performance)
- 63 fps Acura NSX (no lods, a bit more objs and tris than the ZR1)
- 61 fps 911 C2 (lods forcefully disabled)
- 60 fps Diablo VT*
- 52 fps 512TR (*both Diablo and 512 have higher Obj counts, 512's C/D lods are still in the 120s range, a possible reason for the huge drop on performance on this weak CPU; could be tied with material # too, I don't have that measure at hand)
For reference on Daniel's commissioned RSS GTs:
- 73fps 36 V8 NGT
- 71fps Lanzo v12
- 70fps Shadow v8
- 69fps 96 F6 NGT
- 69fps Vortex v10

911 is the worst.

As my experience differs greatly from what you report on VR, I'll stop assuming things for your setup. Good luck!

edit: added C2 with non-working lod files for reference

 

[Deleted member 197115]

TRI and DIP stats are not shown in VR, T-Main thread is under 85% (i9-9900k), must be GPU giving up.
Thanks for taking time to explain.

Still not sure, do LODs matter or not, I'd assume custom LOD can be better optimized than auto generated?

 

[AlleyViper]

LODs should always matter, as they alleviate both cpu and gpu:
I tried the carrera 2 without lods on same test conditions:
- 74 fps 911 C2
- 61 fps 911 C2 (no lods)
- 44 fps 911 C2 (no lods + no cockpit switch)
21% is a lot on my system for the same car. It matters most when LOD As have a good amount of required tris/objects/mats for their details to look good, which can be discarded at greater distances.

Auto generated lods tend to leave a very high object count, and only simplify tris in a ugly way (probably terrible for lod B, passable for D, not so okay on C, depending how bad it looks on mirror). On CPU bound scenarios, they don't help that much reducing draw calls.

Edit: Just to add, the numbers above are the reason users like me praise creators doing free content that go the extra mile for these optimizations, even after taking LOD As up to the highest standards. They come as terribly boring to do, delay the release of stuff, and some times must be redone after changes to LOD A, while new and more enticing projects are put on delay.

 

[Some1]

...they come as terribly boring to do...

Correction: time consuming

Also, I think that VR is still quite a new and niche thing (like ray-tracing), so that running VR on anything else than an absolute best GPU, is probably going to cause issues...no?

 

[Błażej Wianecki]

So here is what I was talking few posts back. Street 90s tires get hot surprisingly fast making the car borderline lethal. I compared the behaviour to Arch's CTR.
I took both cars around Spa in identical conditions with the same tire pressures (I kept them default on 993 and adjusted them on CTR).
I was obviously having some fun in CTR on the second lap. Despite that it keeps its balance till the end without ever gaining those knife-edge characteristics you can observe on 993. On my second lap in carrera I was basically only trying to survive and driving fairly conservative. The tire app also shows big dicsrepancies between the two.
Sorry for the framerate. My recording software turned this footage into a slideshow.

 

[gecco]

small bug report:
The speedometer seems to lag behind. At indicated 100mph, the actual speed is more like ~117mph

 

[Kyuubeey]

So here is what I was talking few posts back. Street 90s tires get hot surprisingly fast making the car borderline lethal. I compared the behaviour to Arch's CTR.
I took both cars around Spa in identical conditions with the same tire pressures (I kept them default on 993 and adjusted them on CTR).
I was obviously having some fun in CTR on the second lap. Despite that it keeps its balance till the end without ever gaining those knife-edge characteristics you can observe on 993. On my second lap in carrera I was basically only trying to survive and driving fairly conservative. The tire app also shows big dicsrepancies between the two.
Sorry for the framerate. My recording software turned this footage into a slideshow.

Hey,

You should really raise cold inflation pressures. The defaults are for semislicks. Try something around 30psi.

As to why the difference, I dunno. Probably just the power, damping etc. I looked at both SVs and they're not super different in heating.

EDIT: And 5 mins in, there is basically no real big difference anyway. Just raise pressures and it will be fine.

 

[aphidgod]

You want to start at ~33/35 at 20C for the 90s tires.

 

[Błażej Wianecki]

Hey,

You should really raise cold inflation pressures. The defaults are for semislicks. Try something around 30psi.


As to why the difference, I dunno. Probably just the power, damping etc. I looked at both SVs and they're not super different in heating.

EDIT: And 5 mins in, there is basically no real big difference anyway. Just raise pressures and it will be fine.

You want to start at ~33/35 at 20C for the 90s tires.

Ok, guys thanks. I’ll try it. I just wanted to point out that those cars even with the same temperature in the tires and the same pressure (admittedly wrong one), those cars act surprisingly different. And I am aware of the weight difference, different rear suspension design, and different wheelbase. I just can’t see how the ctr is considerably the easier one to drive (in terms of handling balance). But let’s see how things are with corrected pressures.

 

[aphidgod]

Don't worry about it. I've adjusted heating in this and every other car, among other things. Lots of little tweaks and tidying up coming from me in the next update.

 

[Błażej Wianecki]

Thanks! Appreciated.

 

[Deleted member 197115]

@Andrew_WOT if anything, it shouldn't be about mip maps (it's almost never about them, AC by itself generates them when loading textures missing them, taking a bit more time to load; many included textures on official content miss them for this reason)

@AlleyViper, can you please elaborate a bit on that automatic LODs generation, what exactly gets generated, is it real B, C, D mipmaps, as I see super bad performance on cars without LODs, and I'd guess if auto generation was a thing, LeBluem would not create this tool.

 

[Deleted member 197115]

Found answer on CSP discord. LODs are about mesh (car geometry) and mipmap is texture.
If no LODs are provided nothing is generated and LOD A is used all the times. Will try to play with LeBluem tool to see how much better no LOD cars can get.

 

[AlleyViper]

@Andrew_WOT, I'll answer as spoiler so not to derail this thread further:

Spoiler

OT: Sorry, I took long to answer as I got interrupted in the middle. Even cars without lods will make use of texture mip maps, even if textures don't include them these will be forcefully generated by acs.exe during session load (taking a bit more time), iirc ~4 mips by default last time I compared vram usage. It's just a matter of creating extra half per side resolution versions of textures of their previous mip-map in sucession, so it can be done on-the fly by AC while loading. Some texture creators might want to create their own mip-maps to improve looks/transitions or resize algorithms for their textures to look better at angles/distances, or control their alpha. Mip lod bias messes with these transitions, so you can make textures at a distance look sharper and possibly a bit heavier to process (higher resolution mips kept farther) or more blurred, for less aliasing. Then AF comes on top to complicate things.

Model LODs are increasingly simplified versions of the main LOD, which inherit stuff like textures, to be loaded at different distances (if existing, my AC mirrors on defaults will only show lods C and D, p.e.). Later lods rendered at greater distances will also show lower resolution mip maps of the inherited textures on LOD A applied.
For these to have maximum perf. results, they should simplify geometry and object/material count. Auto generated seem to fail in the last aspect other than on looks (geometry targets can be met, but the lower you go, the uglier they'll look). Hence 117 vs 78, 87 vs 35, and finally 74 vs 2 objects on auto-lods B/C/D vs manually created ones (from 139 on A) on LeBluem's example. So in that scenario you'll end with near the amount of objects on a auto-generated lod D for very distant use vs the same amount on a much better looking manually created lod B which has more reasons to require them as high for its detail up close.

Other than the significant tris reduction, they'll still help to reduce objects on scene/draw calls (greater impact on performance than tris in AC, usually), but not as much as lods more strictly following AC's pipeline rules as Gary's van. OTH, that generator could possibly give better results when applied to much heavier mods with a very high amount of tris on LOD A that really starts to hurt your GPU, or high object count where that percentage reduction becomes significant, and no lods. LeBluem also explains this on the generator lod page, it still has its uses.

 

[A3DR]

A3DR updated NFS Tournament Class B with a new update entry:

Fixes and stuff

View attachment 331116
v1.3 changelog:

Porsche 993 (C2 &C4)

- Physics updates (credits Jason and Arch)
- Fixed right rear wheel off center
- Corrected tread normals (thanks Gary!)
- Speedometer calibration
- Skin folder names
- Fixed rear wing speed and retract animation
- Added internal sunscreen
- Fixed LOD B brake light
- Updated dash cam position
- Updated crew with proper branding
- Functional oil press gauge

Thanks to @luathas fixed some...

Read the rest of this update entry...

 

[AlleyViper]

Thanks for the update,
A small issue to report, Viper is still missing the auto_cut_off time on gear up as in 1.2, oth it's now enabled on the ZR-1 .

 

[A3DR]

Thanks for the update,
A small issue to report, Viper is still missing the auto_cut_off time on gear up as in 1.2, oth it's now enabled on the ZR-1 .

100% Jason's fault