Forza 3 models (link inside)

Sadface.. kinda saw that coming though.

screenshot005vg.jpg


Yes, I cba with shadows.
 
It's interesting how the LOD1 model (one you see most of the time outside photomode I guess), the cars are still only about 20k polys...

I do wonder sometimes that when PD and so on start saying how many polys their cars have, they are just posturing the peak figures which you hardly ever see anyway.


I still think 40k all up for a car is ample to get the curvatures and quality in there in-game, only showing issue when the car is more than filling the screen (which is fairly rare unless you stop in a replay for a beauty shot, which is when you'd photomode anyway)


Makes me happy I'm still a supporter of lower density meshes for in-game driving use etc. I honestly can't see what high poly cars do for you when driving or watching a replay except drain resource away from better looking environments, or more cars etc :D

Dave
 
I'm pretty sure that what you say is true.
The 500k poly cars in GT5 are probably only used in photomode or whatever it's called ... it would be pretty amazing otherwise :)
The realism comes from good normal mapping (to make the reflections look smooth) and shaders.
40k polys should suffice to create a realistic looking car ingame.
 
It all depends on the topology of the car, some cars are originally more 3D complex....
Mitch is right, the UV mapping work is what makes the real look of a car, in Racer no problem at all, we deal easily with normal maps & there are enough good shaders to create a good looking car.

If Racer had a "damaging/collision" system, then we would deal directly with 2x more times more polys, but that's unfortunately not the case actually speaking...That's why I requested this, in Racer 0.8.22 thread which I'm sure is on your list for the next year, I do really hope so !

In all those cars (FM3, Shift, Dirt2, etc...), the real challenge is to get rid of those damaged 3D pieces to adapt/optimize it for Racer, which is one of the biggest work along with the texturing & shader creation process.
 
the game i edit for right now, has it's own damage and collision engine built in, which alleviates the need for creating "ones own damage" which is a plus because it saves alot of time and hassle creating it all, this is somethign that could work for racer,

edit/sorry i forgot to point something else out, i prefer higher poly cars, say 40-60,000 polygons as opposed to 20-40k, my reasons are thus

a car tends to be more accurate if it's higher poly and looks alot neater, this is just my opinion on this, sure i've used lower poly cars, in the game format, i'm using now, thats because i'm not smart enough to figure out to create my own heh
 
I guess there is always an argument for more polygons, but beyond a certain point do you really notice in a real-time driving/racing game?

Real time damage is a good method imo. Viper Racing did this a decade ago, and imo it worked well, with real deformed body meshes, scrapes to the texture, and physical suspension deformation. I'm not keen on these pre-made damage model systems. I'd rather no damage than that kind of damage to be honest, or simply add dints/scrapes ala CMR2 kind of thing...


I guess most cars can look fine at 20k polys, it's the panel edge technique, wheel poly count, and maybe the exterior seen interior quality that determine the extras. The FM3 models look like about 15k models really, with most of the extra count in the wheel and panel edging techniques.

Hmmm... :D
 
It also depends upon how long you want the content to stay 'current'. GPU power has increased enormously in the last five or ten years, and the screen resoluitons we use now are a lot higher than even five years ago. Is 20k poly's enough to look good in five years?
Guess it really depends on the car.
 
Real time damage is a good method imo. Viper Racing did this a decade ago, and imo it worked well, with real deformed body meshes, scrapes to the texture, and physical suspension deformation. I'm not keen on these pre-made damage model systems. I'd rather no damage than that kind of damage to be honest, or simply add dints/scrapes ala CMR2 kind of thing...

I agree "canned" damage is worthless. I'm glad to see someone else remembers Viper Racing, that was an amazing feat back then. I've been hounding poor Ruud for YEARS for a good damage system, I'm glad to see some support.:)
I would think that with the skinning working now we should be able to do something soon.

Mitch: This link might help...
http://newtondynamics.com/forum/viewtopic.php?f=9&t=5115&p=36621&hilit=damage#p36621
Is this not basically what we need for damage?

Alex Forbin
 
It also depends upon how long you want the content to stay 'current'. GPU power has increased enormously in the last five or ten years, and the screen resoluitons we use now are a lot higher than even five years ago. Is 20k poly's enough to look good in five years?
Guess it really depends on the car.

Well, as said, most of the extra polys on the FM3 cars is in the chamfers for poly panels, and smaller details are now modelled.

The overall topography mesh density is similar to how it's always been for the last decade in my view. Maybe it's doubled (quadruple mesh density), but that is a tiny increase considering the GPU power increases.

That isn't to say more density is useful, but it should be a 'special' model.

My current Z4 looks good and is similar to the FM3 LOD1 meshes. It is designed to be able to have a sub-D run on it without interference (ie, sub-D, then export), and that will be the higher quality model for the future, or for photomode etc.
Basically I'm modelling faithfully enough for real-time, but with sub-D in mind, so the mesh can scale up in future. It's fairly hard, but it means the work isn't too lossy.


In game/sim it really does look smooth even at the LOD1 level. Adding the sub-D really is excessive imo. You only need it for when the car is huge on screen and you stop and start to look. Something we do often in Racer right now because we have few tracks and so focus on the cars.
Get in a race, in the rain, with slippy roads, and suddenly my old M3 looks amply detailed enough to immerse you :D


I think we need to see a move towards quality of content, rather than just detailed high spec content. Endo's Nissan GTR is a perfect case. It looks great, as good as maybe you would ever want, but what it needs most longer term is functions to let it get dirty where the real car does (so a map to control that), then damage control meshes so it deforms nicely and in the right places), options, so we can change colours and add liveries nicely. Swap wheels out with other ones easily.

It's all those things, imo, that we should focus on with Racer. Making high quality meshes and then simply sitting looking at them in-game is boring. I want to interact with these cars and enjoy their reaction to the environment more :D

Polygon counts are not a pre-requisite for that. You could argue it just hogs resource that would be better spent in other areas, which right now, are severely lacking in Racer (ie, environments!)

Dave
 
i use alot of lower end models, at present because they are the only ones that are available to me, for the series i'm creating, but in saying that, they arent inaccurate, quite to the contrary, they are positively accurate, i would have liked them to be higher poly, but thats ok, i find if i user high resolution graphics they look alot better,

and my download stats speak for themselves - mind you these arent for racer, the game i'm currently modding for is 5 years old or so now, i'm still considering moving to racer, only because i can add stuff not replace, as in some games...just not sure if i will...becuase it keeps changing the way things are done, and drives me nuts (no pun intended)
 
Polygon counts are not a pre-requisite for that. You could argue it just hogs resource that would be better spent in other areas, which right now, are severely lacking in Racer (ie, environments!)

That and multiplayer (or vs. AI) where you'll be multiplying poly-counts by 10 just by virtue of having more cars on the field. I drop my fps to about 1/3 if I'm up against 9 AI Lamborghinis. And doing hotlaps is only so much. It's more interesting to have an AI around, or play online.

The track on the other hand, might as well push quality high cause you can pretty well control how much resources it consumes.
 
Yep, you only ever have one track loaded at a time :D

Worth pushing track details a bit more at least to match cars quality. No Racer track has ever really come close to pushing the "visuals of the day" in cars apart from Speedest 3 I think (the one with the wind turbines and vert baking ambient lighting)

I hate to say it but there is a perverse obsession with making cars that are only good as showroom queens. They need to be fast and efficient in-game. Cosmo and I play online on relatively bare tracks quite often and the high poly cars kill the fps a fair old bit. Two cars on a relatively bare track and you wouldn't want another.
One day soon Racer is going to be worth playing properly, online, with lots of friends, on nice looking tracks. High poly cars will simply not get used if they are not efficient at running down their LOD's elegantly.

I feel my system is fast enough, Endo's GTR on a good track looks really good, GT5 good some times, so I know we don't NEED these silly LOD's to get a good look. We just need to calm down a bit with polys :D


Dave
 
Worth pushing track details a bit more at least to match cars quality. No Racer track has ever really come close to pushing the "visuals of the day" in cars apart from Speedest 3 I think (the one with the wind turbines and vert baking ambient lighting)

Well, now that Racer has CSM and TOD, the static lighting in Speedest3 looks pretty bad. :)

I feel my system is fast enough, Endo's GTR on a good track looks really good, GT5 good some times, so I know we don't NEED these silly LOD's to get a good look. We just need to calm down a bit with polys :D

Well, yes and no. I, for one, cannot imagine having bumpmapped bodylines or hard edges (vertices split along an edge) anymore, I want 3d bodylines and chamfering! You can still make pretty optimized mesh with these features, just don't add polies, where they are not needed.

Need For Speed meshes are a very good example, of how to make optimal good looking models, me thinks...

Racer engine also needs to be optimized the hell out of it!

It would be very helpful, if in Racer one could see, how much memory (ram and vram) does the car or track use, how much time is spent rendering it (per frame or something) and so on... sort of profiling of the graphics assets.
 
Yeah, Speedest 3 isn't so nice now. But when it was released, around the time of the DB9 and F430, the tracks and cars felt 'together' in details.

Right now I see huge poly count cars with no feel to them, as if throwing polygons at a problem is a solution, on relatively bare tracks. That is just silly.

I agree, poly edges are good now and I'm planning on using them, and can be done cheaply. However, I feel many details can still be normal map decals, or normal maps can be used more on grilles to soften poly edges etc, but instead we see tiny details done with polygons, and they DO appear to cost right now. Even a slight shadow on the diffuse map, or ambient occlusion map, can hint at the details enough.

Also, general surface poly density need not be too high either... the LOD0 models from FM3 for example. They are overkill for realtime use imo. No matter how fast the system, if the details are not appreciable vs a lower quality asset, then it's wasted performance that could be used to improve details that WILL be visible :)

Racer engine no doubt does need optimising :D


I'm just going on my old M3, my new Z4, Endo's GTR, or Cams F458.

FPS with the former 3 is over 100fps on my test track. Load up the F458 which ultimately doesn't look any better 95%+ of the time, and the FPS are down at 50fps... That is a BIG hit simply down to polygon count, because there are no significant textures on it.
Just my 2p, but that isn't good on tracks that are empty. If we bring the LOD of the tracks up to the levels we are in these mega poly cars, we'll be looking at 5fps with 6 cars on track hehe :D

Simply removing the cars dof files is a good way to do a before/after check of impact on performance :D OK, it's a bit of a weird way to test things, but it works. High poly cars ARE hogging rendering performance a huge amount vs lower poly ones (not LOW poly ones, just lower poly ones that still look fine)! Not sure why!?


NFS meshes are a good target, the ones in Shift were only around 40k with all details iirc (wheels, brakes, interiors seen from outside etc)

Hmmmm

I feel it's nice that FM3 are using mega detailed cars for photomode, but their LOD1 models show what the technical guys and artists are really thinking is right for a realtime game use still. And it's not that much further on from what PGR3/4 were using, about 40k or so all up.



Lastly, a little side note. Poly lines are nice for panels, BUT, right now in Racer the envmap is projected with the high intensity sun spot on it, and that is visible even in occluded areas of a car. So you can end up with highlights of the sun in the envmap, in shaded areas. That isn't really very nice.

I discussed using a second envmap (render every other envmap render without the sun spot), and then use the sunny one on the lit side, and the non sunny one on the occluded pass. Not sure how hard this is, but it'd add some realism. The Lambo's huge side vents are a perfect example of this issue.
Texture panel lines on the other hand, don't suffer from this and so generally in a replay it looks more real, more solid, and is less resource hogging (possibly meaning you can run some DOF and stuff instead) :D


Dave
 
I agree! Although I think it looks better with the lod0 interior. I want to do that, but I need to get rid of the steering wheel, which isn't going to happen as every poly delete takes about 15 seconds - it freezes up for that long each time.
 

Latest News

Shifting method

  • I use whatever the car has in real life*

  • I always use paddleshift

  • I always use sequential

  • I always use H-shifter

  • Something else, please explain


Results are only viewable after voting.
Back
Top