PDA

View Full Version : How much performance does the simulation really cost?



B1rdy
12-03-2018, 11:51
About two months ago PC Games Hardware tested a few games regarding detail settings and measured minimum and average FPS for each preset. The results for Project Cars 2 were very different to other games. One test was done in 1280x720 to take a GPU limit out of the equation. While most games usually don't gain much FPS by reducing details in this scenario, PCars 2 gains a very significant FPS boost when switching from high details to low. At low the game engine produced more than 300 FPS without any impact on the physic simulation according to the author. That raises two questions: Why is there such a huge impact on frame rate? How much performance does the simulation actually need or is it also being dialled back by the graphic detail settings?
Also interesting is that Ian Bell arguments other racing sims perform better (in VR) because they don't have such an elaborate simulation. Pretty much the opposite of what that test is showing.

PS: If desired, I could provide precise numbers later from that test later.

Invincible
12-03-2018, 12:06
That raises two questions: Why is there such a huge impact on frame rate? How much performance does the simulation actually need or is it also being dialled back by the graphic detail settings?
Also interesting is that Ian Bell arguments other racing sims perform better (in VR) because they don't have such an elaborate simulation. Pretty much the opposite of what that test is showing.


To answer both of your questions: The physics aren't dialed back with the graphic setting, but the big impact from higher settings also comes because of the physics. Why is that? Most of the performance hungry stuff comes from Livetrack 3.0 which at low settings mostly taxes your CPU only. The CPU load doesn't really change with higher settings. But the graphical representation of Livetrack does. Puddles, mud, dust, flags blowing in the wind. Each of these are directly linked to the sophisticated physics engine. And rendering these into something you can see is what hits your system.

Sankyo
12-03-2018, 12:11
FPS is the refresh rate of the graphics. In any modern race sim, the rate at which the physics calculations are done is independent of the graphics refresh rate, so also in pC2.

When reducing visual details, the CPU and especially the GPU have less calculations to do per rendering frame and hence can produce more frames per second. As said, this will not affect physics accuracy. On the other hand, if a CPU has to spend lots of power on physics calculations, it has less resources for the rest, including providing the GPU with data. Especially in VR this may result in decreased performance, which is was Ian Bell was referring to.

B1rdy
12-03-2018, 12:42
To answer both of your questions: The physics aren't dialed back with the graphic setting, but the big impact from higher settings also comes because of the physics. Why is that? Most of the performance hungry stuff comes from Livetrack 3.0 which at low settings mostly taxes your CPU only. The CPU load doesn't really change with higher settings. But the graphical representation of Livetrack does. Puddles, mud, dust, flags blowing in the wind. Each of these are directly linked to the sophisticated physics engine. And rendering these into something you can see is what hits your system.But that is also true to most other games and it doesn't explain why the car physics should be very taxing. Also the test showed something like four times better FPS on low details than on high.


FPS is the refresh rate of the graphics. In any modern race sim, the rate at which the physics calculations are done is independent of the graphics refresh rate, so also in pC2.It would be ridiculous and unplayable otherwise.;)


When reducing visual details, the CPU and especially the GPU have less calculations to do per rendering frame and hence can produce more frames per second. As said, this will not affect physics accuracy. On the other hand, if a CPU has to spend lots of power on physics calculations, it has less resources for the rest, including providing the GPU with data. Especially in VR this may result in decreased performance, which is was Ian Bell was referring to.Ian Bell argued visual quality in VR might be worse compared to other sims due to the tire physics - I just found it here (http://forum.projectcarsgame.com/showthread.php?53851-PCars-2-Poor-VR-performance&p=1395773&viewfull=1#post1395773). But this game runs at more that 300 FPS with low details, so the CPU can't be taxed that hard due to AI and the tire simulation.

Invincible
12-03-2018, 13:24
But that is also true to most other games and it doesn't explain why the car physics should be very taxing. Also the test showed something like four times better FPS on low details than on high.

It would be ridiculous and unplayable otherwise.;)

Ian Bell argued visual quality in VR might be worse compared to other sims due to the tire physics - I just found it here (http://forum.projectcarsgame.com/showthread.php?53851-PCars-2-Poor-VR-performance&p=1395773&viewfull=1#post1395773). But this game runs at more that 300 FPS with low details, so the CPU can't be taxed that hard due to AI and the tire simulation.

Could you provide a link to said article? I only found this: http://www.pcgameshardware.de/Project-Cars-2-Spiel-55773/Specials/Benchmark-Test-Grafikkarten-1238952/ and here even with 1280x720 they barely reach 100 FPS.

In their verdict they said:


Generell erscheinen die erzielten Frameraten in Anbetracht der Optik sowie der recht anspruchsvollen Szene ordentlich - für Full HD und 60 Fps bei maximierten Details reicht im Grunde eine flotte Mittelklasse-Grafikkarte. Jedoch ist kaum zu übersehen, dass insbesondere die AMD-GPUs beim Wechsel auf die nächst höhere WQHD-Auflösung kaum Frames einbüßen, was für eine Limitierung durch unseren Prozessor spricht.


In general, the achieved frame rates seem good in relatance to the the optics as well as the quite demanding scene - for full HD and 60 Fps with maximized details a middle class GPU is sufficient. However, it is hard to overlook the fact that especially the AMD GPUs lose hardly any frames when changing to the next higher WQHD resolution, which speaks for a limitation by our processor.

Bealdor
12-03-2018, 13:33
Could you provide a link to said article? I only found this: http://www.pcgameshardware.de/Project-Cars-2-Spiel-55773/Specials/Benchmark-Test-Grafikkarten-1238952/ and here even with 1280x720 they barely reach 100 FPS.

In their verdict they said:

I guess it's this one: http://www.pcgameshardware.de/Grafikkarten-Grafikkarte-97980/Specials/Spiele-Detailskala-1247839/

Unfortunately it's a premium article.

Although I don't know the details of the test I have to say that using an overclocked 10 core CPU is not quite, err... representative.

Invincible
12-03-2018, 13:46
I guess it's this one: http://www.pcgameshardware.de/Grafikkarten-Grafikkarte-97980/Specials/Spiele-Detailskala-1247839/

Unfortunately it's a premium article.

Although I don't know the details of the test I have to say that using an overclocked 10 core CPU is not quite, err... representative.

Also in their non-premium article, they confirmed what Ian said - that in lower resolutions, the game is mostly CPU bound.

Edit: I just read the article - they used rain, but it wasn't said if they had AI on the track for their benchmark. Seeing the numbers, I dare to say they did not.

Mike1304
12-03-2018, 14:22
Just because the physics are not so CPU demanding as in other sims this doesn’t have to mean that the physics are worse. They can be programmed in a more efficient way...
At least what I feel when driving tells me that the physics are spot on in most cars...

zedeeyen
12-03-2018, 14:30
There are far too many variables even to speculate on these questions. No part of a game engine exists in isolation and there are always multiple ways to implement and optimise the same features. Even just unrealistically restricting the conversation to the sim itself, the CPU hit tells you literally nothing about its complexity or accuracy. A simplistic simulation can be written badly and run more slowly than a well-written but far more complex simulation.

B1rdy
12-03-2018, 14:30
Could you provide a link to said article? I only found this: http://www.pcgameshardware.de/Project-Cars-2-Spiel-55773/Specials/Benchmark-Test-Grafikkarten-1238952/ and here even with 1280x720 they barely reach 100 FPS.Right, true for max/ ultra details.


I guess it's this one: http://www.pcgameshardware.de/Grafikkarten-Grafikkarte-97980/Specials/Spiele-Detailskala-1247839/

Unfortunately it's a premium article.

Although I don't know the details of the test I have to say that using an overclocked 10 core CPU is not quite, err... representative.Thanks, that is the article. The intention of the tests is to show the performance scaling of the graphic options, not if average Joe can run it on his potato.


Also in their non-premium article, they confirmed what Ian said - that in lower resolutions, the game is mostly CPU bound.Right, but the questions still remain: 1. Why is the FPS gap between low and high details so huge in a non-GPU bound scenario? 2. How much performance does the driving model really take?


Edit: I just read the article - they used rain, but it wasn't said if they had AI on the track for their benchmark. Seeing the numbers, I dare to say they did not.I can be more specific later, but I am pretty sure they used AI and started on last position as they always do (there should be screenshots for comparison in the article).

Bealdor
12-03-2018, 14:36
Right, but the questions still remain: 1. Why is the FPS gap between low and high details so huge in a non-GPU bound scenario? 2. How much performance does the driving model really take?

1. Because they used an overclocked high performance 10 core CPU that doesn't have an issue with the physics system. If they had compared PCARS 2 with similar games (AC, R3E) they most likely would've found similar results.

2. It eats up a whole CPU core on consoles for the player car alone, which is indeed quite demanding compared to other sims.

Mahjik
12-03-2018, 14:46
Real-time shadows and other graphical effects are costly. If PC2 didn't have day/night transitions and dynamic weather, they could make everything static (since the environment wouldn't change during a race) which would free up a ton of graphical processing for the rest of the system which is exactly what the test showed that is mentioned. In short, it has to do with the dynamic nature of the environment/scene.

B1rdy
12-03-2018, 15:19
1. Because they used an overclocked high performance 10 core CPU that doesn't have an issue with the physics system.It doesn't really matter, does it? A weaker CPU would still show a similar percentage drop in FPS.


2. It eats up a whole CPU core on consoles for the player car alone, which is indeed quite demanding compared to other sims.Interesting!


Real-time shadows and other graphical effects are costly. If PC2 didn't have day/night transitions and dynamic weather, they could make everything static (since the environment wouldn't change during a race) which would free up a ton of graphical processing for the rest of the system which is exactly what the test showed that is mentioned. In short, it has to do with the dynamic nature of the environment/scene.So the dynamic environment actually becomes static on low details, hence the huge gain in FPS?

Bealdor
12-03-2018, 15:25
It doesn't really matter, does it? A weaker CPU would still show a similar percentage drop in FPS.

Yeah, but what's your point?
If anything this shows how well the detail scaling works ingame, which is what PCGH mentions too btw.


So the dynamic environment actually becomes static on low details, hence the huge gain in FPS?

Well, at least all dynamic shadows and most environmental reflections are gone on the lowest setting.

Edit: As I said, if you want representative results you need to compare similar games. Comparing PCARS 2 to AC:Orgins and Overwatch doesn't tell you anything about physics fidelity.

Alex_76
12-03-2018, 15:30
I find that PC2 is noticeably less CPU dependent than other sims, probably due to the decision to run simplified AI physics. On my CPU bottlenecked system (i5 4690k 4.2Ghz, GTX 1080) I can start at the back of 20+ car grids in rain, at night, in VR and be at a smooth 45fps ASW, then after a few corners be up to a sold 90fps. In Assetto Corsa or Raceroom a grid of 20 cars would be permanent 45fps ASW irrespective of graphics settings, with mod cars in AC that can drop to as few as 10 cars.

I can imagine that the transition points would be difficult for developers to implement but it would be good to have a best of both worlds compromise, similar to 3d model LODs, where the complexity of the AI physics are reduced at a certain distance from the player.

Mad Al
12-03-2018, 16:05
I find that PC2 is noticeably less CPU dependent than other sims, probably due to the decision to run simplified AI physics. On my CPU bottlenecked system (i5 4690k 4.2Ghz, GTX 1080) I can start at the back of 20+ car grids in rain, at night, in VR and be at a smooth 45fps ASW, then after a few corners be up to a sold 90fps. In Assetto Corsa or Raceroom a grid of 20 cars would be permanent 45fps ASW irrespective of graphics settings, with mod cars in AC that can drop to as few as 10 cars.

I can imagine that the transition points would be difficult for developers to implement but it would be good to have a best of both worlds compromise, similar to 3d model LODs, where the complexity of the AI physics are reduced at a certain distance from the player.

How would that help.. on the starting grid, with more cars visible, your CPU overhead for physics would be at it's highest.. just at the point when you need most CPU cycles to deal with more visual info too?

Alex_76
12-03-2018, 16:31
It depends how it's be implemented, for example there could be a limit on the number of AI cars running full physics at any one time, with only those nearest to the player on full physics. I can see how there could be many issues with switching AI physics on the fly but if it were possible to implement something like that it would be a good compromise between the costs and benefits of full physics AI.

Anyway, most people are willing to live with lower framerates at race start provided that they increase to a reasonable level after the first couple of corners. In AC, which has the same physics for the player and AI, that doesn't happen past a certain point.

Mad Al
12-03-2018, 16:54
It depends how it's be implemented, for example there could be a limit on the number of AI cars running full physics at any one time, with only those nearest to the player on full physics. I can see how there could be many issues with switching AI physics on the fly but if it were possible to implement something like that it would be a good compromise between the costs and benefits of full physics AI.

Anyway, most people are willing to live with lower framerates at race start provided that they increase to a reasonable level after the first couple of corners. In AC, which has the same physics for the player and AI, that doesn't happen past a certain point.

You would struggle to get the majority of CPUs to support more than one instance of the full physics... the current AI physics are more complex than PC1 already (one of the reasons why the max number was dropped).

B1rdy
12-03-2018, 17:14
Yeah, but what's your point?
If anything this shows how well the detail scaling works ingame, which is what PCGH mentions too btw.As I am reading the article once more, they don't. They are wondering where this huge performance impact with added details (without being GPU limited) comes from, because other games don't behave like that. They assume particles and reflections create a large overhead on higher details.

Regarding the numbers: Their CPU scaling test (no GPU limit) with a i7-6950@4Ghz (with lowered cores/ threads) in the rain at Le Mans with AI showed the following numbers:

low details
8C/16T 350 FPS
4C/4T 310 FPS

High details
8C/16T 130 FPS
4C/4T 115 FPS

This leads to the following conclusions:

- any physic regarding the cars, track and basically everything that is affecting driving isn't particular demanding*
- AI isn't either
- some graphic settings create a very high CPU overhead

* This leads me back to Ian Bells´s statement: The elaborate physic simulation of PCars 2 cannot be a reason of performance bottlenecks (in VR).

Bealdor
12-03-2018, 17:47
As I am reading the article once more, they don't. They are wondering where this huge performance impact with added details (without being GPU limited) comes from, because other games don't behave like that. They assume particles and reflections create a large overhead on higher details.

They do, right below the first graph...


Regarding the numbers: Their CPU scaling test (no GPU limit) with a i7-6950@4Ghz (with lowered cores/ threads) in the rain at Le Mans with AI showed the following numbers:

low details
8C/16T 350 FPS
4C/4T 310 FPS

High details
8C/16T 130 FPS
4C/4T 115 FPS

This leads to the following conclusions:

- any physic regarding the cars, track and basically everything that is affecting driving isn't particular demanding*
- AI isn't either
- some graphic settings create a very high CPU overhead

* This leads me back to Ian Bells´s statement: The elaborate physic simulation of PCars 2 cannot be a reason of performance bottlenecks (in VR).

It can, in situations where graphics are not set to minimum with a resolution that would make your eyes bleed.
That's also a reason why I think this kind of benchmark is nonsense (in general, not only for this scenario).
Literally nobody is playing in those settings and although you can get an indication about the CPU load under certain circumstances, the result is not representative or useful for the regular player.

I'll say it again, just because a 1500€ CPU doesn't have performance issues with PCARS 2's physics engine doesn't mean that it doesn't have a negative impact on the overall graphic fidelity under normal gaming circumstances.

Sankyo
12-03-2018, 17:50
How do you know that if pC2 would have had a less complex physics model, at High details the FPS wouldn't have been 200+? Things often work very non-linearly in computing.

pC2's core physics engine runs at 600Hz irrespective of graphics frame rate, so at low settings there is already much more going on so that the GPU 'only' manages ~300 FPS. I don't know what Ian was referring to specifically, but as mentioned above it could very well be LiveTrack3.0 which is basically the environmental physics simulation part of the game that is limiting VR performance w.r.t. other games.

B1rdy
13-03-2018, 09:36
It can, in situations where graphics are not set to minimum with a resolution that would make your eyes bleed.Maybe, but it is still a very minor factor.


That's also a reason why I think this kind of benchmark is nonsense (in general, not only for this scenario).
Literally nobody is playing in those settings and although you can get an indication about the CPU load under certain circumstances, the result is not representative or useful for the regular player.I already described the intention of that benchmark. Would you, please, stay on topic.


I'll say it again, just because a 1500€ CPU doesn't have performance issues with PCARS 2's physics engine doesn't mean that it doesn't have a negative impact on the overall graphic fidelity under normal gaming circumstances.Please, read my posting again. They also did that test with 4C/4T, that is the equivalent of an i5 @4Ghz+. And even if you chose a CPU with half that clock rate, you'd still be at around 150 FPS on low details.


How do you know that if pC2 would have had a less complex physics model, at High details the FPS wouldn't have been 200+? Things often work very non-linearly in computing.That would mean, the physic model were dependent on graphic settings. According to the author that doesn't seem to be the case.


pC2's core physics engine runs at 600Hz irrespective of graphics frame rate, so at low settings there is already much more going on so that the GPU 'only' manages ~300 FPS.Yes, maybe a simpler physic would lead to even more FPS. Anyway the point still stands, the physic isn't that taxing, something else with added graphic details is.


I don't know what Ian was referring to specifically, but as mentioned above it could very well be LiveTrack3.0 which is basically the environmental physics simulation part of the game that is limiting VR performance w.r.t. other games.That would mean less graphic details would highly impact LiveTrack 3.0. I still fail to see a connection between LiveTrack 3.0 and performance in VR. Why would it be specific to VR? We already ruled out CPU performance to an extend.

Bealdor
13-03-2018, 10:19
Yes, maybe a simpler physic would lead to even more FPS. Anyway the point still stands, the physic isn't that taxing, something else with added graphic details is.

That would mean less graphic details would highly impact LiveTrack 3.0. I still fail to see a connection between LiveTrack 3.0 and performance in VR. Why would it be specific to VR? We already ruled out CPU performance to an extend.

Firstly dynamic reflections, shadows and particles tend to create a high overhead on the CPU in many games. How is that news to you/anyone?
Just because they're graphics settings doesn't mean that they're exclusively calculated by the GPU.
As a result, performance will decrease especially in rain and/or at night.

Secondly, nobody said that this is a VR exclusive issue.
Which leads me back to the point I made earlier that those CPU "benchmarks" aren't representative because nobody's going to play under those conditions anyway.
If the tire physics are more demanding than in other sims it leaves less headroom for said reflections/shadows etc. therefore leading to worse performance compared to games with less sophisticated physics.

Sankyo
13-03-2018, 11:15
That would mean, the physic model were dependent on graphic settings. According to the author that doesn't seem to be the case.
No it doesn't mean that. It means that, as Bealdor stated, there is no 100% separation between graphics and the rest of the game engine. The CPU still has to do the preparations to feed the GPU before rendering is done. The more complex you want the graphics to be, the more work the CPU has to do. Furthermore, not all graphics stuff is GPU-related. And finally, pC2's deferred rendering engine has some great visual advantages, but in general is 'heavier' than a traditional rendering engine, requiring more CPU work (I think, not into the details myself).


Yes, maybe a simpler physic would lead to even more FPS. Anyway the point still stands, the physic isn't that taxing, something else with added graphic details is.
W.r.t. graphics output the physics calculations aren't the biggest bottleneck, but that's something different than 'the physics aren't that taxing'. As I stated, the core physics engine of the player's car (vehicle dynamics + tyres) runs at 600 Hz. This means that every update of the car's state needs to happen within 1.67 ms. Given the complexity of especially the tyre model, it takes a significant amount of CPU resources and heavily optimized threaded coding to get that done. Older CPUs simply cannot cope with pC2's physics model and are not able to keep up the 600 Hz physics refresh rate even under 'simple' conditions. The fact that a modern CPU is able to do it and produce good FPS is not a good indication of how taxing the physics model of pC2 is.

B1rdy
13-03-2018, 17:18
Firstly dynamic reflections, shadows and particles tend to create a high overhead on the CPU in many games. How is that news to you/anyone?See post #19. Anyone? I am sure your mom knows about that.


Just because they're graphics settings doesn't mean that they're exclusively calculated by the GPU.
As a result, performance will decrease especially in rain and/or at night.Yes, right. A 66% loss in FPS in a CPU limiting scenario, is still unusual, thus interesting.


Secondly, nobody said that this is a VR exclusive issue.Right. People blamed the physic simulation in PCars 1 (especially in conjunction with AMD GPUs) and now PCars 2 (VR) for higher CPU load. This test shows, it is obviously not the case.


Which leads me back to the point I made earlier that those CPU "benchmarks" aren't representative because nobody's going to play under those conditions anyway.
If the tire physics are more demanding than in other sims it leaves less headroom for said reflections/shadows etc. therefore leading to worse performance compared to games with less sophisticated physics.Again, this test is representative. It doesn't need to meet usual play conditions to bring a valid result. It depends on what aspect you want to investigate.


No it doesn't mean that. It means that, as Bealdor stated, there is no 100% separation between graphics and the rest of the game engine. The CPU still has to do the preparations to feed the GPU before rendering is done. The more complex you want the graphics to be, the more work the CPU has to do. Furthermore, not all graphics stuff is GPU-related. And finally, pC2's deferred rendering engine has some great visual advantages, but in general is 'heavier' than a traditional rendering engine, requiring more CPU work (I think, not into the details myself).Yes, I'd like to know what it is.


W.r.t. graphics output the physics calculations aren't the biggest bottleneck, but that's something different than 'the physics aren't that taxing'. As I stated, the core physics engine of the player's car (vehicle dynamics + tyres) runs at 600 Hz. This means that every update of the car's state needs to happen within 16.7 ms. You probably mean 1.67ms. The article aimed at the car handling physic being much less taxing than expected - maybe I was too unspecific about that.


Given the complexity of especially the tyre model, it takes a significant amount of CPU resources and heavily optimized threaded coding to get that done.As far as that benchmark shows us, there is no significant usage of CPU resources compared to the added load with more graphic details.


Older CPUs simply cannot cope with pC2's physics model and are not able to keep up the 600 Hz physics refresh rate even under 'simple' conditions. The fact that a modern CPU is able to do it and produce good FPS is not a good indication of how taxing the physics model of pC2 is.What do you mean by "older CPUs"? Even the slow Jaguar cores of consoles and seven year old i5s with less then 3Ghz can run the game/ simulation fine (with lowish graphics details!).

Bealdor
13-03-2018, 17:46
See post #19. Anyone? I am sure your mom knows about that.

You should seriously consider dropping that Mr. Wise Guy attitude because it's getting you nowhere here...


Yes, right. A 66% loss in FPS in a CPU limiting scenario, is still unusual, thus interesting.

If you want to know if this is unusual you should compare similar games because the results for AC:O or Overwatch aren't really comparable here.
How does AC's or R3E's engine scale with dynamic reflections during rain and night? Oh wait...


Right. People blamed the physic simulation in PCars 1 (especially in conjunction with AMD GPUs) and now PCars 2 (VR) for higher CPU load. This test shows, it is obviously not the case.

Again, this test is representative. It doesn't need to meet usual play conditions to bring a valid result. It depends on what aspect you want to investigate.

Yes, I'd like to know what it is.

You probably mean 1.67ms. The article aimed at the car handling physic being much less taxing than expected - maybe I was too unspecific about that.

As far as that benchmark shows us, there is no significant usage of CPU resources compared to the added load with more graphic details.

What do you mean by "older CPUs"? Even the slow Jaguar cores of consoles and seven year old i5s with less then 3Ghz can run the game/ simulation fine (with lowish graphics details!).

How can you say that the physics aren't demanding and the consoles are doing just fine when they're struggling to handle higher AI counts (especially XB1)?

I'm still not sure what you're trying to achieve here.
Do you want to imply that Ian is lying? Or that he doesn't know how his own game's engine works?

Edit: Forgot to mention that there's actually proof out there that PCARS 2's physics are more demanding than other sims, because AC can run full AI grids with player physics while PCARS 2 can't.

Sankyo
13-03-2018, 19:44
You probably mean 1.67ms.
Yep, typo, thanks.


The article aimed at the car handling physic being much less taxing than expected - maybe I was too unspecific about that.
Not sure what they were expecting, though? A 'taxing' physics engine meaning that only 3 physics calculations per second are done? When simulating cars at high speed, you need a high physics tick rate or what comes out of the physics engine is not useful at all. The physics engine limiting FPS to less than 100 or so with graphics on Low means it's not usable.


As far as that benchmark shows us, there is no significant usage of CPU resources compared to the added load with more graphic details.
Which is not the same as saying that the physics model isn't taxing. Just wrong expectations, see above. Anyway, as stated already, LiveTrack3.0 is part of the physics part of the game as well and unless I remember incorrectly that does not and cannot run at 600 Hz.


What do you mean by "older CPUs"? Even the slow Jaguar cores of consoles and seven year old i5s with less then 3Ghz can run the game/ simulation fine (with lowish graphics details!).
They can, but they're really struggling as you showed yourself. The CPU hardly has any grunt left for anything but low-ish graphics, and an amazing amount of optimization was pulled out of the hat to make it work. So yeah it runs, but compare to other race games/sims and you see that pC2 is a lot more taxing than anything else out there.

CSL-Drive
14-03-2018, 01:52
To answer both of your questions: The physics aren't dialed back with the graphic setting, but the big impact from higher settings also comes because of the physics. Why is that? Most of the performance hungry stuff comes from Livetrack 3.0 which at low settings mostly taxes your CPU only. The CPU load doesn't really change with higher settings. But the graphical representation of Livetrack does. Puddles, mud, dust, flags blowing in the wind. Each of these are directly linked to the sophisticated physics engine. And rendering these into something you can see is what hits your system.
Oh thanks. I was wondering, why I coulden't see some puddles. I thought it was a bug. I should try max graphics. The same for iRacing dirt ovals. Low graphics doesn't allow one to see the important dynamic evolution of the track. I did an ireland greenwood kart race on career, started raining around 3/5 lap, last corner was full aqua planing, even tho I saw no puddle. If the the max graphics can reveal those, that'd be pretty cool to finally see all those puddles on that track I had never seen before.

morpwr
14-03-2018, 02:05
Oh thanks. I was wondering, why I coulden't see some puddles. I thought it was a bug. I should try max graphics. The same for iRacing dirt ovals. Low graphics doesn't allow one to see the important dynamic evolution of the track. I did an ireland greenwood kart race on career, started raining around 3/5 lap, last corner was full aqua planing, even tho I saw no puddle. If the the max graphics can reveal those, that'd be pretty cool to finally see all those puddles on that track I had never seen before.


I was never a fan of racing in the rain before the sharpening tool but since that and raising a couple settings I'm having a blast. They did a really good job with the rain and puddles imo. Pretty much reacts and looks just like I would expect and being able to clearly see where the puddles are and how deep they are really helps.

BigDad
14-03-2018, 06:45
To answer both of your questions: The physics aren't dialed back with the graphic setting, but the big impact from higher settings also comes because of the physics. Why is that? Most of the performance hungry stuff comes from Livetrack 3.0 which at low settings mostly taxes your CPU only. The CPU load doesn't really change with higher settings. But the graphical representation of Livetrack does. Puddles, mud, dust, flags blowing in the wind. Each of these are directly linked to the sophisticated physics engine. And rendering these into something you can see is what hits your system.

Could we please have a detailed list of what settings disable what? For instance, if we run medium track detail what is missing and just what the levels are per setting, not just track detail though, a list for all settings would be awesome.
I can't see a difference between High and Ultra track detail and going from Medium to High i only see more trees and spectators but to know what is actually changing would be great so we know what more to optimize on our systems for max effect in VR.

morpwr
14-03-2018, 10:45
Could we please have a detailed list of what settings disable what? For instance, if we run medium track detail what is missing and just what the levels are per setting, not just track detail though, a list for all settings would be awesome.
I can't see a difference between High and Ultra track detail and going from Medium to High i only see more trees and spectators but to know what is actually changing would be great so we know what more to optimize on our systems for max effect in VR.

I noticed this last night after changing particle settings from low to medium now if you go off roading it throws up very visible chunks of grass and dirt which was a neat effect. :D So I'm sort of wondering the same thing .

BigDad
14-03-2018, 10:54
I noticed this last night after changing particle settings from low to medium now if you go off roading it throws up very visible chunks of grass and dirt which was a neat effect. :D So I'm sort of wondering the same thing .

It would be very cool but i don't think we will get it :(

bmanic
14-03-2018, 11:25
I find that PC2 is noticeably less CPU dependent than other sims, probably due to the decision to run simplified AI physics. On my CPU bottlenecked system (i5 4690k 4.2Ghz, GTX 1080) I can start at the back of 20+ car grids in rain, at night, in VR and be at a smooth 45fps ASW, then after a few corners be up to a sold 90fps. In Assetto Corsa or Raceroom a grid of 20 cars would be permanent 45fps ASW irrespective of graphics settings, with mod cars in AC that can drop to as few as 10 cars.

I can imagine that the transition points would be difficult for developers to implement but it would be good to have a best of both worlds compromise, similar to 3d model LODs, where the complexity of the AI physics are reduced at a certain distance from the player.

AC is actually the exception to the norm. It's the only simulator out there that has the AI cars running the player physics. All the other sims have simplified physics for AI.

morpwr
14-03-2018, 11:27
It would be very cool but i don't think we will get it :(

At least give us a hint.lol Id really like to know if track detail is just tied to people and objects or is tied to live track also.

B1rdy
14-03-2018, 11:48
To answer both of your questions: The physics aren't dialed back with the graphic setting, but the big impact from higher settings also comes because of the physics. Why is that? Most of the performance hungry stuff comes from Livetrack 3.0 which at low settings mostly taxes your CPU only. The CPU load doesn't really change with higher settings. The article about the detail levels proves the opposite.


Also in their non-premium article, they confirmed what Ian said - that in lower resolutions, the game is mostly CPU bound.Mostly GPU bound yes, at more than 110 FPS (with 4C/4T), however. In VR the frame rate is capped at 90 FPS. Sure the compositor takes some CPU time, but anything more than an i5 should have enough headroom to hit 90 FPS on low details (with regards to that benchmark). That is why I don't understand how car physics should impact VR performance in a relevant way?

Bealdor
14-03-2018, 12:25
That is why I don't understand how car physics should impact VR performance in a relevant way?

It doesn't, as long as you're using minimum settings. But nobody is going to play that way...
But as soon as you're increasing those details that are adding a lot of CPU overhead it does, simply because you have less spare budget left for eye candy than other sims have.

BigDad
14-03-2018, 14:14
At least give us a hint.lol Id really like to know if track detail is just tied to people and objects or is tied to live track also.

i just did some testing from Low to Ultra Track detail and with heavy rain at Long Beach i couldn't notice any difference in puddles .
I sat there watching a specific puddle with x20 weather so puddles formed quickly and it seemed the same from Low right through to Ultra .
The thing that made the puddles look different was because the building right behind the puddle disappeared changing the reflected image with Low detail but puddles looks identical after the same amount of time in all tests .

gregc
14-03-2018, 15:03
From memory the Reflections setting (might be track reflections or similar) is what controls the rendering of water - at work so don't have access to the game to check, but the hint text on the settings screen (top right) gives a guide to what each setting does.

BigDad
14-03-2018, 21:25
From memory the Reflections setting (might be track reflections or similar) is what controls the rendering of water - at work so don't have access to the game to check, but the hint text on the settings screen (top right) gives a guide to what each setting does.

I started testing this also but a Ultra there is toooo much a performance hit so even if Ultra adds everything I can't use it so just sticking with High.

BigDad
14-03-2018, 22:30
I just wish pC2 was popular enough for a channel like Gamers Nexus or Digital Foundry to do a performance summary like they do for other big popular titles. GN have one atm for FFXV and those games suck to me but must have a following :(
I just asked GN on their latest FFXV vid if they would do it, but I'm only one person, they would need more interest before doing it I guess. Get over there and show interest so maybe they'll do it.

B1rdy
15-03-2018, 08:36
It doesn't, as long as you're using minimum settings. But nobody is going to play that way...Look at the settings people use when playing in VR. If you want to archive 90 FPS in VR, you are stuck at a mix of low and med settings even with a 6 core CPU and a 1080Ti.

Sankyo
15-03-2018, 08:46
I'm getting lost. What question are we now trying to answer?

Shepard2603
15-03-2018, 08:49
I'm getting lost. What question are we now trying to answer?
There is no question. Just the hidden statement that the physics engine and tire model are not the culprit, but the bling-bling on high to ultra settings are.

Bealdor
15-03-2018, 08:59
There is no question. Just the hidden statement that the physics engine and tire model are not the culprit, but the bling-bling on high to ultra settings are.

Yeah and it's been stated multiple times already that (and why) it's a mix of both...

Shepard2603
15-03-2018, 09:01
Can you please repeat it once more? I am not sure I understand it.... :onthego:

morpwr
15-03-2018, 11:44
Look at the settings people use when playing in VR. If you want to archive 90 FPS in VR, you are stuck at a mix of low and med settings even with a 6 core CPU and a 1080Ti.

Not even close. Mostly maxed settings. The only setting I don't have maxed is grass,track and shadows on high.

BigDad
15-03-2018, 14:45
Look at the settings people use when playing in VR. If you want to archive 90 FPS in VR, you are stuck at a mix of low and med settings even with a 6 core CPU and a 1080Ti.

Basinga!

BigDad
15-03-2018, 14:48
Not even close. Mostly maxed settings. The only setting I don't have maxed is grass,track and shadows on high.

Not everyone is as fortunate to be running 1080Ti's like us and must be using Low on from what I've seen people are playing VR on anywhere from 970's and 1060's.
They must be on Low and in ASW constantly.
Poor saps.

Keena
15-03-2018, 14:56
^ But rich in other areas not necessarily aligned with ownership of the latest gpu bling I'm sure ;)

pferreirag60
15-03-2018, 15:33
Not everyone is as fortunate to be running 1080Ti's like us and must be using Low on from what I've seen people are playing VR on anywhere from 970's and 1060's.
They must be on Low and in ASW constantly.
Poor saps.

I concur with you, but tell me, why such a weak CPU when you have bought an 1080ti ?! When I bought my PC, the best gpu unfortunelly was a GTX980( i bought 2), not even the gtx 980TI. I bought the third best cpu of the time, the diference in price/perfomance for 8 cores at the time, was not the best option( 3 times more expensive) my Motherboard was just 485 Euros!

And even if everyone says an I5 is ok, it is not! I´m an MS admin for maybe 25 years now, my present workstation at work is an 7th gen i5, and if I have more than 3 aplications open(Outlook, SQL server interface,Browser with 2 tabs) it just crawl, becomes to slow.

I´m just so against 2 or 4 cores cpu´s, today to slow to multitasking.

morpwr
15-03-2018, 15:55
Not everyone is as fortunate to be running 1080Ti's like us and must be using Low on from what I've seen people are playing VR on anywhere from 970's and 1060's.
They must be on Low and in ASW constantly.
Poor saps.

I know but he specifically said you cant even do it with that and you can.

BigDad
15-03-2018, 21:58
I concur with you, but tell me, why such a weak CPU when you have bought an 1080ti ?! When I bought my PC, the best gpu unfortunelly was a GTX980( i bought 2), not even the gtx 980TI. I bought the third best cpu of the time, the diference in price/perfomance for 8 cores at the time, was not the best option( 3 times more expensive) my Motherboard was just 485 Euros!

And even if everyone says an I5 is ok, it is not! I´m an MS admin for maybe 25 years now, my present workstation at work is an 7th gen i5, and if I have more than 3 aplications open(Outlook, SQL server interface,Browser with 2 tabs) it just crawl, becomes to slow.

I´m just so against 2 or 4 cores cpu´s, today to slow to multitasking.

When I bought my system I bought the 6600k and a 1060, so was a perfectly matches system.

I then bought a 1070 then 1080TI and tried a 7700k but was playing in 4k at the time and had NO change in fps so took the 7700k back.

At 4.7GHz my 6600k has up there with the fastest single core performance by even today's latest CPU's so not really a weak gaming CPU.

When real gaming performance comes i will upgrade both CPU and GPU again.

And my system has nothing open when gaming and never will even if I ran 20 cores, it's a gaming rig only.

Why would your "workstation" at MS workplace have a i5 in it, lol, speaks volumes!

B1rdy
16-03-2018, 12:10
Not even close. Mostly maxed settings. The only setting I don't have maxed is grass,track and shadows on high.Ok, maybe your PC is somewhat of an exception having the best CPU for gaming and oc'ed to the max. Are you running super sampling and/ or MSAA? Are you having 90 FPS most of the time (even in Rain with AI)?

BigDad
16-03-2018, 13:36
Ok, maybe your PC is somewhat of an exception having the best CPU for gaming and oc'ed to the max. Are you running super sampling and/ or MSAA? Are you having 90 FPS most of the time (even in Rain with AI)?

Even with my "weak" CPU i can SS to 1.3 and have MSAA High with most other on Medium and High and yes i get 90 FPS just about everywhere and in all situations . ASW looks bad in my opinion and have done everything in my power to eliminate it with current system .
It really is a matter of trial and error and having found my sweet prop i've been racing more and more happily than ever .

B1rdy
20-03-2018, 12:09
Even with my "weak" CPU i can SS to 1.3 and have MSAA High with most other on Medium and High and yes i get 90 FPS just about everywhere and in all situations . ASW looks bad in my opinion and have done everything in my power to eliminate it with current system.This is interesting: Is performance on Rift by far better than on Vive? In general: SS and MSAA are calculated on GPU not CPU. Would you try your settings on Cota in rain, 12 AI cars and player car at last position? Do you get 90FPS in that situation? Where did you apply SS?

BigDad
20-03-2018, 13:27
This is interesting: Is performance on Rift by far better than on Vive? In general: SS and MSAA are calculated on GPU not CPU. Would you try your settings on Cota in rain, 12 AI cars and player car at last position? Do you get 90FPS in that situation? Where did you apply SS?

I just left it as i usually have it , so 19 Ai and me and sat behind the pack for 1 lap and checked MSI AB to see the graph and it was a flat line except for a drop to 47 ( i guess it was 45 but was only brief so only showed 47) about half way , and from a player point of view that stutter seemed to be on the back straight otherwise it felt quite smooth .
I had a mixed class lap GT3 and GTE , i was in the GT3 M6 . It was Spring at 14:00 , rain .
SS in applied ingame @ 1.3.
My 1080Ti is undervolted to 1911MHz @ 0.900mv and my 6600k is @ 4.7GHz and my new RAM @ 3200MHz and pC2 is on a full 1TB HDD .