Latest Roadmap Update - March 2017!

[patchedupdemon]

I've learned about iRacing tracks being off in elevations (not camber or radius, just height varying) so there is room for error on Laser Scan, since the track still has to be built and there's a human in the process. And you might want to consider that some tracks do provide either laser, survey, or CAD data to software developers. As long as S397 gets data - survey, scan, CAD, or good LiDAR - and build it right, then we can have accurate tracks. I have brought it up before and will do so again, but look at Atlanta Motorsports Park. Don't focus on the bush/tree art, just the quality of the layout itself.

Mind you, you are not alone in desiring accuracy. The argument of "you'll never drive an F1 or GT car there" is weak since it's specific personal preference and the point is simulating the feeling as best as you can. These types of games are not Sega Rally, Daytona USA, The Crew, or Super Monaco GP 25 years ago. If you watch racing or even visited a track, you can spot the differences. Something that could be a reference point in real life may be completely missing, a corner profile be very different - and thus possibly kill the corner -, and an awesome track may become a shadow of itself.

Do you honestly think our consumer grade pc's are capable of handling the models needed to deliver a realistic true to life experience both for physics and tyres,no,so until we can run banks of cpus like f1 teams do,the driving aspect is just a very vague approximation of rl,that's the way I see it,and hence why I'm not bothered about laser scanned tracks.
It's not real,and will never be real,it's a game we can play on a laptop for god sake

 

[vittorio]

Do you honestly think our consumer grade pc's are capable of handling the models needed to deliver a realistic true to life experience both for physics and tyres,no,so until we can run banks of cpus like f1 teams do,the driving aspect is just a very vague approximation of rl,that's the way I see it,and hence why I'm not bothered about laser scanned tracks.
It's not real,and will never be real,it's a game we can play on a laptop for god sake

As I wrote in post #281.
Assetto Corsa for instance uses irregular meshes instead of regular meshes like rF2 to handle this level of more details. So parts of the tarmac that needs more details get more polygons (like 10cm triangles), parts that need less detail get less polygons. You think rF2 with its more complex physics and tyre model can't handle this? Wrong! rF2 can handle this level of more details without problems. There are convertions of ACs Nordschleife for rFactor2 using the same mesh. rF2 handles these without any problem.
(BTW I use ACs Nordschleife in rF2. In my country reverse engineering to solve incompatibility between plattforms for software and content is allowed, as long as software/content is owned)

EDIT: Reading your post again I think you totally question the possibility to simulate "car racing" on consumer PCs. Well, like Moors law says: processing power grows exponential with time. Exponential grow is huge. E.g. smartphones have more processing power now then the best professional racing simulators had 5-10 years ago.

 

[stonec]

As I wrote in post #281.
Assetto Corsa for instance uses irregular meshes instead of regular meshes like rF2 to handle this level of more details. So parts of the tarmac that needs more details get more polygons (like 10cm triangles), parts that need less detail get less polygons. You think rF2 with its more complex physics and tyre model can't handle this? Wrong! rF2 can handle this level of more details without problems. There are convertions of ACs Nordschleife for rFactor2 using the same mesh. rF2 handles these without any problem.

The AC Nords conversion I know of doesn't use the physical mesh for rF2, it uses the visual mesh, which is much less dense. rF2 with its 2400 Hz physical tire model demands a lot more CPU power once you start increasing the mesh density compared to something like rF1/AC, which runs a comparably simple mathematical model. Also Moore's law hasn't applied for consumer CPU's in years, these days we are lucky to get 5% increase in performance per each new Intel generation. Basically a six year old Intel i5 can run today's games just fine.

 

[vittorio]

The AC Nords conversion I know of doesn't use the physical mesh for rF2, it uses the visual mesh, which is much less dense. rF2 with its 2400 Hz physical tire model demands a lot more CPU power once you start increasing the mesh density compared to something like rF1/AC, which runs a comparably simple mathematical model. Also Moore's law hasn't applied for consumer CPU's in years, these days we are lucky to get 5% increase in performance per each new Intel generation. Basically a six year old Intel i5 can run today's games just fine.

This post has the potential to ruin my day I didn't know there is a physical and visual mesh in AC. But IIRC the AC mesh i was exploring in 3DsimED, and which details impressed me so much, was the same as i have seen in rfactor2. This level of detail (10cm triangles for detailed parts) i thought is great.

You are right, PCs CPU processing power didn't follow Moore's law the last years. GPUs, smartphones and embedded CPUs probably did. But I think it's still true. Lets see how new technologies like VR changes this.

 

[patchedupdemon]

As I wrote in post #281.
Assetto Corsa for instance uses irregular meshes instead of regular meshes like rF2 to handle this level of more details. So parts of the tarmac that needs more details get more polygons (like 10cm triangles), parts that need less detail get less polygons. You think rF2 with its more complex physics and tyre model can't handle this? Wrong! rF2 can handle this level of more details without problems. There are convertions of ACs Nordschleife for rFactor2 using the same mesh. rF2 handles these without any problem.
(BTW I use ACs Nordschleife in rF2. In my country reverse engineering to solve incompatibility between plattforms for software and content is allowed, as long as software/content is owned)

EDIT: Reading your post again I think you totally question the possibility to simulate "car racing" on consumer PCs. Well, like Moors law says: processing power grows exponential with time. Exponential grow is huge. E.g. smartphones have more processing power now then the best professional racing simulators had 5-10 years ago.

I was trying to convey my opinion,in a very poor fashion,that laser scanning shouldn't be the be all and end all of the equation.
If the physics and tyres are good,then he'll,I'd run on any track as long as it's well made.
I just don't understand why people believe the hype of realistic this real data that,when the software is made to run on potatoes.
I'm not bashing any particular game,or undermining any effort put in by the devs,I just the community had taken this "its real" way too far
I'm not questioning the ability of today's cpu's to run the models needed to simulate racing,just that the depth of the models is limited by those cpu's.
Why would f1 teams need banks of cpu's to run rfpro if my i5 could do it just fine,hence why I say all consumer grade sims are vague representations of the real deal,because the models are resource limited.
i may be completely wrong,probably am,so if anyone can educate me on this I'm all for it,ignorance is my strongest trait regrettably,and grammar my weakest lol

 

[vittorio]

Why would f1 teams need banks of cpu's to run rfpro if my i5 could do it just fine,hence why I say all consumer grade sims are vague representations of the real deal,because the models are resource limited.

This is indeed an interesting question i often asked myself too. How realistic are consumer grade simulations. I don't know. But I find this article written by a formula 3 racing engineer comparing simulation data (using rfactor) with real meassured data very interesting:

https://drracing.wordpress.com/2014/01/03/how-close-is-close-enough/

His conclusion:
"To sum up, all what we saw proves that, when real and accurate input are used to model vehicles in rFactor, a very good correlation between real car vs. simulated car output data can be achieved.
There are of course some small differences, as we have seen, that are difficult to eliminate without artificially “tricking” the data, but they are in general very small and sometimes not directly connected to the vehicle model, but to the track or to the driver.
Some more investments in track modeling could easily reduce these effects, but it was out of the scope of this for me! ..."

 

[T1specialist]

It is a misconception to think that if you can model something only at 1% in scale you can only get 1% accuracy. Simulations can provide very good accuracy at running at 333Hz while driving on polygon faces that are at minimum tens of centimeters wide. While increasing the frequency probably helps with iteration there is not some magical thing missing that makes modern sims just a vague approximation... After all at 333Hz you are only modelling the tire universe 333 times per second. One planck time is 5.39 × 10e−44 seconds. So technically even the best universe model running at 333Hz can only be 5,566× 10e-41 percent accurate. But even then we can reach the right cornering Gs and lap times very accurately... the correlation between how much you simulate and what is your accuracy is not even close to linear. Not even by a long shot.

Computer code is all about simplifying things. To have gravity in your game you don't need to model string theory. In fact doing so won't give you any benefits at all unless your game is about string theory and gravity. Even the simple newtonian gravity is already overkill. A simple equation of vertical acceleration some basic math about how positions change over time is already good enough. But not simple.

Same thing with tires, dampers, aero... Going from relatively simple drag and lift numbers to full blown wind tunnel and cfd doesn't really bring anything new to the table. You are simply consuming tons of processing power where simple variables give identical result to you when you drive the car. Calculating the vortices and flow structures adds absolutely nothing when in the end all you have is drag and lift and other variables about how those affect your car. All of which can be simulated with simpler models. After all cfd is not the universe either but a simplification as well. Cfd can be useful for figuring out those lift and drag numbers but it is simply stupid to run cfd real time in a game because it adds nothing and slows down everything. Even if you had unlimited processing power nobody sane would add real time wind tunnel into their game. Not just because coding a real timewind tunnel is not only bigger project than coding a game but puts insane stress on making sure the meshes are accurate enough so you get the right result...

Getting just 1% (or 0.001%!) of all the variables that control the tire behaviour is already a lot and already cover 107% of the everyday situations tire goes through. Adding quantum physics and modelling all atoms and particles individually every planck second may make your 1% model into 30% model or 50% model but in the end you will not notice any difference except in the electricity bills. More complex is not always better.

If 0.01% already get you close to 99.9% and going from 0.01% to 0.02% only gives you an increase of 99.901 then what's the point? It is diminishing results. While 20 years ago tire models had to leave out important features and simplify heavily nowadays there is no such requirement. Today the issue is that the code we have doesn't really model what it tries to do accurately. It is not a question of lack of resources but question of quality of the code and how accurate are the numbers going into the math. No super computer will help you if your math is wrong, your data is wrong. And the key is getting the right data. Even the crappiest simulation comes alive when you feed real data into it. Just like the bestest simulation ever falls into pieces if your numbers are considerably off.

This doesn't mean stuff like laser scanning is useless. Even with relatively low poly result (you don't drive on the point cloud) laser scanning adds lots of bumps into the surface that can be felt even at 333Hz and 50cm minimum polygon size.The strength of laser scanning is that the resulting mesh has lots of detail in it which is missing if you try to create the mesh in any other way (like photo mapping which I think is how rf2 tracks are done). You can of course manually add bumps where you want and if you do it well then the surface can be as lively as the laser scan. Not as accurate but imho good enough.

 

[Louis]

http://www.computerweekly.com/feature/Apollo-11-The-computers-that-put-man-on-the-moon

 

[patchedupdemon]

It is a misconception to think that if you can model something only at 1% in scale you can only get 1% accuracy. Simulations can provide very good accuracy at running at 333Hz while driving on polygon faces that are at minimum tens of centimeters wide. While increasing the frequency probably helps with iteration there is not some magical thing missing that makes modern sims just a vague approximation... After all at 333Hz you are only modelling the tire universe 333 times per second. One planck time is 5.39 × 10e−44 seconds. So technically even the best universe model running at 333Hz can only be 5,566× 10e-41 percent accurate. But even then we can reach the right cornering Gs and lap times very accurately... the correlation between how much you simulate and what is your accuracy is not even close to linear. Not even by a long shot.

Computer code is all about simplifying things. To have gravity in your game you don't need to model string theory. In fact doing so won't give you any benefits at all unless your game is about string theory and gravity. Even the simple newtonian gravity is already overkill. A simple equation of vertical acceleration some basic math about how positions change over time is already good enough. But not simple.

Same thing with tires, dampers, aero... Going from relatively simple drag and lift numbers to full blown wind tunnel and cfd doesn't really bring anything new to the table. You are simply consuming tons of processing power where simple variables give identical result to you when you drive the car. Calculating the vortices and flow structures adds absolutely nothing when in the end all you have is drag and lift and other variables about how those affect your car. All of which can be simulated with simpler models. After all cfd is not the universe either but a simplification as well. Cfd can be useful for figuring out those lift and drag numbers but it is simply stupid to run cfd real time in a game because it adds nothing and slows down everything. Even if you had unlimited processing power nobody sane would add real time wind tunnel into their game. Not just because coding a real timewind tunnel is not only bigger project than coding a game but puts insane stress on making sure the meshes are accurate enough so you get the right result...

Getting just 1% (or 0.001%!) of all the variables that control the tire behaviour is already a lot and already cover 107% of the everyday situations tire goes through. Adding quantum physics and modelling all atoms and particles individually every planck second may make your 1% model into 30% model or 50% model but in the end you will not notice any difference except in the electricity bills. More complex is not always better.

If 0.01% already get you close to 99.9% and going from 0.01% to 0.02% only gives you an increase of 99.901 then what's the point? It is diminishing results. While 20 years ago tire models had to leave out important features and simplify heavily nowadays there is no such requirement. Today the issue is that the code we have doesn't really model what it tries to do accurately. It is not a question of lack of resources but question of quality of the code and how accurate are the numbers going into the math. No super computer will help you if your math is wrong, your data is wrong. And the key is getting the right data. Even the crappiest simulation comes alive when you feed real data into it. Just like the bestest simulation ever falls into pieces if your numbers are considerably off.

This doesn't mean stuff like laser scanning is useless. Even with relatively low poly result (you don't drive on the point cloud) laser scanning adds lots of bumps into the surface that can be felt even at 333Hz and 50cm minimum polygon size.The strength of laser scanning is that the resulting mesh has lots of detail in it which is missing if you try to create the mesh in any other way (like photo mapping which I think is how rf2 tracks are done). You can of course manually add bumps where you want and if you do it well then the surface can be as lively as the laser scan. Not as accurate but imho good enough.

While your post in some ways agrees with what I'm saying,I'm predominantly use iracing,where they all believe that they will add aero effects when they have down cfd wind tunnel tests,this attitude was what I was on about in my post,so I was in a way thinking that the community here was on the same "its real" bs as the iracing community.
But your post was a joy to read,and proved that's no the case,rf2 community seems like a more down to earth common sense bunch of guys.
I must remember that in future