[computerbase] Project CARS benchmarks

[RussianSensation]

It would be pretty dumb if the Project CARS devs tied GPU physx to a bunch of effect settings instead of having one big on-off switch/slider ala Borderlands and Metro.

But it looks like that's exactly what they did. I think you need to read the thread as some users already provided benchmarks on NV cards that show that the game starts to perform extremely poorly when GPU PhysX is forced OFF in the NV's control panel.

http://forums.anandtech.com/showpost.php?p=37389081&postcount=101

Also, the various members who contributed funds during early development of the game voiced information to the developer early on that because PhysX was underlying the game's main physics effects (i.e., tire and car handling) that it would inevitably be hurting performance on all AMD GPU systems, but the developer refused to change the development process of the game knowing and publicly acknowledging this.

 

[DarkKnightDude]

I was actually one of those people who crowdfunded the game back in 2012 (Senior Manager member), and the alpha ran fine on my 7970 and a borrowed 7850. Though never played it much till this year.

 

[Gikaseixas]

I noticed while playing this game an abnormally high CPU load and temperature for my i5, I guess using a Geforce would lower this load...

WOW, and to think that some people defend Nvidia's actions religiously
So PhysX is a standard feauture that a user cannot turn off correct? Forcing AMD users to take a hit since the CPU has to run PhysX. Just bad business practices by devs and Nvidia

 

[Silverforce11]

I was actually one of those people who crowdfunded the game back in 2012 (Senior Manager member), and the alpha ran fine on my 7970 and a borrowed 7850. Though never played it much till this year.

Can you do another test with PhysX GPU vs CPU on a track with rain? We can settle this GameWorks weather once and for all.

 

[DarkKnightDude]

WOW, and to think that some people defend Nvidia's actions religiously
So PhysX is a standard feauture that a user cannot turn off correct? Forcing AMD users to take a hit since the CPU has to run PhysX. Just bad business practices by devs and Nvidia

There is no option to turn off PhysX in the visual options as far as I can see. It might be tied with particles option or density, but not sure.

EDIT:

@Silverforce11 I'll give that a shot.

 

[3DVagabond]

That won't work, cos if Intel doesn't support it natively, Motherboard makers will just add a chip to enable it. The same motherboard makers as the ones selling dGPUs. ;p

Why Intel has no reason to do that, because they got nothing to gain. NV has reasons to destroy AMD GPU performance.

Nothing to gain? You're joking, right? Destroy the number 1 vendor in HPC. Intel surely doesn't need nVidia. And if the mobo manufacturers want to go back to pci-e support on the mobo but without Intel's chipsets supporting it, I suppose they can try.

 

[Vaporizer]

Ultimatly this is now the situation NV is working on the past few years. Surprisingly most of the people are apolausing and cheering to all the steps NV did to achieve this and throwing money on them.

 

[Vesku]

Typical post -- as always jumping to baseless conclusions, without vetting any of the information beforehand. It's already been stated in this and other threads that the game's engine uses PhysX as the underlying physics engine. This code is proprietary to NV and is being offloaded to the GPU side for NV, but it's being run on the CPU for AMD GPUs. AMD has nothing to do with PhysX, and AMD can't optimize PhysX source code or offload it to the GPU. :whiste:

Also the devs had to correct themselves, AMD hadn't stopped trying to work with them. Communication was open at least into March this year.

 

[mikk]

Ian Bell said that's the problem. PhysX calculations run on the GPU for nVidia and CPU for AMD.

Here we have another blatant lie. He never told this and I think you know it but prefer to spread misinformations in order to defend AMD.

This thread is full of lies and bullshit. Any serious player could enable the Physx Indicator and realize that whatever he sets in the control panel (GPU or CPU) only Physx + CPU is running in the game. It also makes absolutely no difference in performance. In case of a GPU Physx support we would have known it from serious testers already and AMD itself. This is not the case.

 

[Silverforce11]

Here we have another blatant lie. He never told this and I think you know it but prefer to spread misinformations in order to defend AMD.

This thread is full of lies and bullshit. Any serious player could enable the Physx Indicator and realize that whatever he sets in the control panel (GPU or CPU) only Physx + CPU is running in the game. It also makes absolutely no difference in performance. In case of a GPU Physx support we would have known it from serious testers already and AMD itself. This is not the case.

Users on forums have already found, forcing PhysX to CPU only hurts performance on NV setups.

http://steamcommunity.com/app/234630/discussions/0/613957600537550716/

Also @DarkKnightDude

You calling them liars?

Edit: When the dust settles and the devs patch the game up to fix AMD performance, you should be here to apologize.

 

[Gikaseixas]

Here we have another blatant lie. He never told this and I think you know it but prefer to spread misinformations in order to defend AMD.

This thread is full of lies and bullshit. Any serious player could enable the Physx Indicator and realize that whatever he sets in the control panel (GPU or CPU) only Physx + CPU is running in the game. It also makes absolutely no difference in performance. In case of a GPU Physx support we would have known it from serious testers already and AMD itself. This is not the case.

mikk i suggest you take a chill pill and read the thread before name calling others. I wasn't so sure but now it's very clear to me that Nvidia is in bed with the dev here.

 

[Flapdrol1337]

because PhysX was underlying the game's main physics effects (i.e., tire and car handling) that it would inevitably be hurting performance on all AMD GPU systems,

I don't think this is the case, these type of calculations are generally always done on the cpu, and use very little cpu time. It runs fine on the console cpu's, so this can't be the problem.

 

[3DVagabond]

Here we have another blatant lie. He never told this and I think you know it but prefer to spread misinformations in order to defend AMD.

This thread is full of lies and bullshit. Any serious player could enable the Physx Indicator and realize that whatever he sets in the control panel (GPU or CPU) only Physx + CPU is running in the game. It also makes absolutely no difference in performance. In case of a GPU Physx support we would have known it from serious testers already and AMD itself. This is not the case.

Calm down there. Maybe you need to read up a bit before you get so carried away.

 

[Erenhardt]

Just tested it myself

HUD on: 55 FPS
HUD off: 65 FPS

7950@1.1Ghz and FX6300@4.4
Custom settings

I like how every PC gaming site in my country is going wild about amd drivers in Projectcars because of Roy's Tweet about drivers for this game ^^. Bad performance on amd cards is only a driver thing on amd site. Nvidia didn't put a finger into it.

Again, month or two from now they will fix their game. But not a single site will cover how suddenly radeons are competing with their nv equivalents.

Every single Gameworks release reviewed on pclab had in their conclusion that amd drivers are crap and are killing cpu. We at least know that it is gameworks fault.

Well played nvidia

 

[jj109]

Users on forums have already found, forcing PhysX to CPU only hurts performance on NV setups.

http://steamcommunity.com/app/234630/discussions/0/613957600537550716/

Also @DarkKnightDude

You calling them liars?

Edit: When the dust settles and the devs patch the game up to fix AMD performance, you should be here to apologize.

mikk i suggest you take a chill pill and read the thread before name calling others. I wasn't so sure but now it's very clear to me that Nvidia is in bed with the dev here.

Calm down there. Maybe you need to read up a bit before you get so carried away.

I don't even know who's trolling who at this point, but there are blatant liars in this thread and on the steam forums just looking to make trouble. The fact is mikk was right and is owed some apologies.

As you can see, there's negligible GPU usage drop when I force PhysX onto CPU because there are no GPU accelerated Physx happening... Well, at this point I know SOMEONE is blatantly lying.

This is what it looks like when there are GPU accelerated physx features turned on:

 

[railven]

Conditions:
24 AI cars
Weather set to Clear
Map: Nordschleife
Pagani Zonda R
3:00PM

FRAPS benchmark log for one lap each. Though the max seems off.

My settings:
http://i.imgur.com/58QGrhy.png

Everything maxed besides motion blur off and SMAA was left on low. Honestly the game felt smooth both ways, but doesn't excuse the fps loss.

Thanks for testing. I look forward to your test with rain on.

Really interesting, as of now it seems more people are reported performance penalty when switching to CPU PhysX. Of course there are outliners, but they aren't providing as much info to the overall convo.

@96Firebird
Just doing my part to warn potential AMD owners who didn't know about this GameWorks title and buy it on Steam, wasting their $. Wouldn't want even more AMD gamers raging about it, the AMD threads are getting big as it is. Feel free to do your part, chime in and defend the developers & NV.

@RS Thanks for the PhysX CARS & Project nVidia name (http://steamcommunity.com/app/234630/discussions/0/613957600538780982/). It truly represents this disgraceful game.

Wow, I got mentioned on Steam forums? Crap, hope they don't visit my profile (I haven't been as active lately :/ damn you WoW addiction!)

 

[ShintaiDK]

I don't even know who's trolling who at this point, but there are blatant liars in this thread and on the steam forums just looking to make trouble. The fact is mikk was right and is owed some apologies.

As you can see, there's negligible GPU usage drop when I force PhysX onto CPU because there are no GPU accelerated Physx happening... Well, at this point I know SOMEONE is blatantly lying.

This is what it looks like when there are GPU accelerated physx features turned on:

Nice with facts rather than peoples homebrew stories or trolling. :thumbsup:

Added with the large difference in AMDs own drivers. Then you just have to be unserious if you still pretend its not an AMD side issue.

 

[railven]

I don't even know who's trolling who at this point, but there are blatant liars in this thread and on the steam forums just looking to make trouble. The fact is mikk was right and is owed some apologies.

As you can see, there's negligible GPU usage drop when I force PhysX onto CPU because there are no GPU accelerated Physx happening... Well, at this point I know SOMEONE is blatantly lying.

This is what it looks like when there are GPU accelerated physx features turned on:

In defense of mikk, it's true Ian never said that Nvidia was offloading PhysX (not from my understanding of his transcribed posts) but he did mention on AMD side it was being done on CPU. Which to me alludes to Nvidia is offloading or he would have added a modifier such as "and Nvidia too."

But as someone with a GTX 780 and the game, add to the convo and run some benches. You can help clear up this snafu.

 

[Erenhardt]

Am I missing something or it says physx>cpu on both screenshots. Way to fail.

 

[railven]

Nice with facts rather than peoples homebrew stories or trolling. :thumbsup:

Added with the large difference in AMDs own drivers. Then you just have to be unserious if you still pretend its not an AMD side issue.

Only fact there is you can display what PhysX is running on. I didn't see any numbers to support "there is no change."

I'd love to eat crow though. Ball is in his court.

EDIT:

In defense of mikk, it's true Ian never said that Nvidia was offloading PhysX (not from my understanding of his transcribed posts) but he did mention on AMD side it was being done on CPU. Which to me alludes to Nvidia is offloading or he would have added a modifier such as "and Nvidia too."

But as someone with a GTX 780 and the game, add to the convo and run some benches. You can help clear up this snafu.

Ouch, didn't even notice that. Hope he provides better benches now to clear his own name cuz that's dastardly!

 

[jj109]

Am I missing something or it says physx>cpu on both screenshots. Way to fail.

That just means there are no GPU physx enabled in the game, period.

 

[Silverforce11]

Am I missing something or it says physx>cpu on both screenshots. Way to fail.

Lets wait for reliable reviewers to cover this.

Back & forth from forum user testing isn't gonna be believed with people insulting others as liars.

Edit: I want to be proven wrong, it means then AMD can optimize their drivers or the devs can patch up and fix the issue. If it IS GPU-offloading of particles, rain, or PhysX calculations, then I don't think AMD has a chance to fix it, besides waiting for the DX12 version to remove CPU bottlenecks as a factor.

 

[Spanners]

That just means there are no GPU physx enabled in the game, period.

You can't say that the reporting tool is accurate or even if forcing Physx via the control panel does override things in the game in the way you'd expect. I agree with the lets see how this plays out approach. I know there is a PhysX3Gpu_x64.dll in the game folder. Whether it's naive to assume that means some GPU based Physx is going on I don't know.

People seem quick to jump on anything that confirms their pre-conceived notions and call it "proof". I'd be hesitant to call anyone a liar or anything of the sort until more concrete info comes out.

 

[psolord]

I was intrigued with what a poster said about the game running better on radeons in previous versions of the game, but Gameworks was updated to make the game run worse on the final version.

So I did the following test (spicy wallpaper alert on vids ).

Project Cars (0.831 vs 1.2) 1920X1080 maxed (SMAA) 7950 @1.1Ghz CORE i7-860@4Ghz

What this is, is an automated run in the Azure coast, which I benchmarked with fraps and recorded with a digital recorder.

Of course I did the same run on both game versions, 0.831 and 1.2.

I have used the same maxed settings on both runs (smaa only), but there is one discrepancy. I missed the streched headlights reflections setting on the 1.2 run and left it to No, while it was Yes at 0.831. Still this has no effect on the performance. I redid the runs after I recorded the video and was getting the same results within a margin of error. I am using a morning setting, so I guess the headlights should be turned off anyway.

Also the 1.2 version has a track detail setting at Ultra where it was only High at 0.831. That is not of my own doing. This is what these versions are offering.


Now here comes the interesting part. The old version scored 20fps while the new version scored 29fps. That's a 45% performance increase!

I cannot vouch for what happens on other Radeons though. Maybe Hawais do get a performance hit, I don't have one to test.

Also what I would like to point out, is that the GPU load was at maximum in the 0.831 run, while it had some minor cpu limits on the second run (gpu load droped from maximum). So with a better cpu, I would have more than 45% better performance, maybe 50%.

Also I cannot be sure about PhysX's role in this, but normally when you have hardware accelerated PhysX at play, you get negative load spikes even on Nvidia's cards themselves. On the 7950 I had mostly max gpu load.

For example, look at the following Mafia 2 MSI Afterburner gpu load screenshot. That's from a 970@1.5Ghz.



This screenshot came from this video if you want to see the whole runs.

Mafia II 1920x1080 Maxed (physx on & off) GTX 970 @1.5Ghz Core i5 2500k @4.

Here is the Project Cars gpu load on both runs, with the 7950. Not consistent with hardware Physx being present.



Also regarding the cpu heaviness of this game, these are my runs of the 970 with 3 different cpus. These are from the 0.831 version. I guess I will have to redo them with the 1.2.

Project Cars 1920x1080 Maxed (SMAA) GTX 970 @1.5Ghz Core i5 2500k @4.8G - 62 fps

Project Cars 1920X1080 maxed (smaa) GTX 970 @1.5Ghz CORE i7-860 @4GHz - 53 fps

Project Cars 1920x1080 Maxed (SMAA) GTX 970 @1.5Ghz Q9550 @4GHz - 34 fps

I could be wrong, but what I see is not consistent with hidden hardware accelerated PhysX, unless this is something totally new.

I believe we will see a HUGE performance increase with future drivers and/or patches, on the Radeons. Remember Assassin's Creed Unity?

 

[Tapoer]

I don't even know who's trolling who at this point, but there are blatant liars in this thread and on the steam forums just looking to make trouble. The fact is mikk was right and is owed some apologies.

As you can see, there's negligible GPU usage drop when I force PhysX onto CPU because there are no GPU accelerated Physx happening... Well, at this point I know SOMEONE is blatantly lying.

This is what it looks like when there are GPU accelerated physx features turned on:

Some problems with your test:

1 - Unknown CPU, you can have an intel 6 core at 4.5GHz for all we know, which will have little troubles with CPU physx.

2 - No rain/thunderstorm which is said to have a bigger impact on physx.

3 - Unknown AI cars if any, which will eat away CPU resources

So your test can be very CPU light and GPU heavy, which when you offload the physx to the CPU you actually have more fps with CPU physx.

Your test can actually proof that there is physx in this game that can be GPU accelerated.

@ShintaiDK, you cannot pick a test that suits you liking and ignore all others that have been done on this thread, it doesn't work like that, we people don't have just 2 neurons...