Project Cars: CPU benchmarks (well threaded!)

[TurtleCrusher]

I would rather see you say: I think it doesn't use it. Or you have a source from nvidia saying project cars has no gpu physx?

What he thinks is facts, apparently, with little (if any) self-doubt. Unable to handle a second or diverse opinion. Simon Baron-Cohen's "Theory of Mind" becomes extremely relevant when talking to ShintaiDK.

 

[TeknoBug]

Good to know that i3's are capable of playing this game, it keeps up with a 5GHz OC'd AMD FX for god sake. lol

 

[Alatar]

I would rather see you say: I think it doesn't use it. Or you have a source from nvidia saying project cars has no gpu physx?

Does Nvidia's own drivers and PhysX indicator saying it doesn't have it count?

Here's as much smoke as I can give you with the PhysX indicator on. It's quite clearly CPU.

Or maybe these tests:

CPU PhysX is included with the game, GPU PhysX needs the driver.

Edit - I'm checking right now to see if PhysX is still on my computer or not. I did a clean install of the driver, but PhysX is still listed in my Programs and Features list. I can't uninstall from there, so I'm looking for another way...

Edit 2 - Ok, this time I am certain PhysX was uninstalled... It was gone from Programs and Features, and also gone from the System Information --> Components tab, under 3D Settings. Before, it showed the version number there, now it is gone.

Same outcome though...

Ahem, 32 cars, rain, max settings, CPU vs GPU physX:

3 runs for both CPU and GPU physX to minimize the variance between runs.

And here's the worst case scenario run that I ran 6 times in total:

https://www.youtube.com/watch?v=gTllE5LqgrM

Right at the start of a race with all the cars on the screen and lots of water effects everywhere because of it. Also the player car visible so you have to render the water effects for that as well including all the rain drops on the car.

(also excuse the keyboard driving)

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:

I did two quick bench runs with my main setup (5960X + original Titan). One with auto detect CPU/GPU physX and one with forced CPU physX from the Nvidia control panel:

The only test where someone has claimed that CPU vs GPU PhysX made a difference was this one:

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.

And no one has confirmed or retested it yet, unlike the other claim that it doesnt make a difference.

 

[Erenhardt]

What whould physx indicator say if some physx is done on CPU and some on GPU?

 

[mikk]

straight from horses mounth:
http://www.geforce.com/whats-new/articles/gamescom-2014-project-cars


So from now one I would love to see you put as much effort into spreading that information, as you did spreading the misinformation.

You have anything to support your claims?
Do you expect us to believe they backed off from GPU physx, that was showcased and is advertised on nv site?

At this time APEX Turbulence was planned.....it has been delayed for a post-release update later. There is no current info when it comes (or if). Once it comes you can expect decreased performance with Turbulence enabled since a GPU Physx eye candy won't be for free. Currently no GPU Physx is in use as said.

 

[Erenhardt]

At this time APEX Turbulence was planned.....it has been delayed for a post-release update later. There is no current info when it comes (or if). Once it comes you can expect decreased performance with Turbulence enabled since a GPU Physx eye candy won't be for free. Currently no GPU Physx is in use as said.

It says quite opposite. There is physx APEX Turbulence on the demo.

 

[mikk]

It says quite opposite. There is physx APEX Turbulence on the demo.

It doesn't matter whether or not APEX was enabled in some showcase demo for us. I just said how it is currently. As I said it was planned for release but they postponed it for a post-release patch/DLC. That was their statement some months ago. It may not acceptable to you as you prefer to blame others but not AMD. GPU Physx hinders performance in all games I'm aware of, I guess it would be the first game where a GPU Physx eye candy feature improves performance. Would be nice but this is nothing we can expect realistically. The whole Nvidia Physx thing has nothing to do with AMDs much worse performance.

 

[Erenhardt]

It doesn't matter whether or not APEX was enabled in some showcase demo for us. I just said how it is currently. As I said it was planned for release but they postponed it for a post-release patch/DLC. That was their statement some months ago. It may not acceptable to you as you prefer to blame others but not AMD.

You have anything to back this up. You are going against what nv said. Anything official...?

GPU Physx hinders performance in all games I'm aware of, I guess it would be the first game where a GPU Physx eye candy feature improves performance. Would be nice but this is nothing we can expect realistically. The whole Nvidia Physx thing has nothing to do with AMDs much worse performance.

This is another lie. I advise you too to post your opinions with a disclaimer, and not present such opinions as facts (spatially when those are lies)

Here is a comparison of GPU vs CPU physx

http://www.pcper.com/reviews/Graphi...and-PhysX-Comparison-GTX-680-and-HD-7970/GPU-

You need me to tell you exactly what those graphs show, or you can figure it out on your own?

PS. Can you quote me on saying amd drivers can't increase performance of project cars?

 

[Alatar]

What whould physx indicator say if some physx is done on CPU and some on GPU?

It would say GPU physX

You have anything to back this up. You are going against what nv said. Anything official...?

How hard is it to understand this:

?

I really don't understand what's the debate anymore. The huge majority of tests show absolutely no difference between CPU and GPU Physx, the only one that doesn't is a one off and NV's own tools say it's pure CPU physX.

 

[Erenhardt]

UPDATE: I did finally get an answer from Gearbox about the slow downs we were seeing on the AMD results. Apparently when larger collections of PhysX simulations are running on the CPU, those threads can take quite a bit longer than they would when running on the GPU. As a result, the CPU (and rest of the game engine code) becomes "blocked" waiting for a single thread to finish, which results in the lower CPU utilization we saw on the AMD results as well as the lower overall performance.

This is interesting. Anyone with amd and nv gpu can test it?

Alatar, it wouldn't be the first time when nv drivers do something wrong.

Nvidia says gpu physx,
dev says/hint gpu physx,
forum poster say cpu physx....

Hmm, who is right?

 

[mikk]

You have anything to back this up. You are going against what nv said. Anything official...?

Source is Slightly Mad Studios from the internal wmdportal forums. Because I have access since late 2011 and you don't it's no surprise my knowledge is much higher.

Nvidia did not say Retail version of pcars using GPU Physx. Once again, some showcase demo is irrelevant for our Retail version.

APEX Turbulence can't run on AMD by the way because it is a GPU only Physx feature.

 

[Erenhardt]

Source is Slightly Mad Studios from the internal wmdportal forums. Because I have access since late 2011 and you don't it's no surprise my knowledge is much higher.

Nvidia did not say Retail version of pcars using GPU Physx. Once again, some showcase demo is irrelevant for our Retail version.

APEX Turbulence can't run on AMD by the way because it is a GPU only Physx feature.

Then we have your word for it.. I see.

Also, how many shares you have? Made any profit yet?

For those not in the know - he is investor in project cars. Investors get their return and profit from market success of project cars. More info here:
https://www.youtube.com/watch?v=5eD7o_f7s0Y

 

[SlowSpyder]

This thread reminds me of the GTA V thread... "AMD gets hammered again" and then I get the game and it plays great. I'm not too interested in racing / car simulator games, but if a low of ~50FPS is all I have to worry about, I'm ok.

 

[monstercameron]

Then we have your word for it.. I see.

Also, how many shares you have? Made any profit yet?

For those not in the know - he is investor in project cars. Investors get their return and profit from market success of project cars. More info here:
https://www.youtube.com/watch?v=5eD7o_f7s0Y

I dont see how he would profit from the bad press the games has had...

 

[Reticula]

Such a controversy around Project Cars, like a good old Ageia/early NVIDIA times. Same "don't know what to blame - blame PhysX" stuff

Do you expect us to believe they backed off from GPU physx, that was showcased and is advertised on nv site?

Happens all the times. Some project live, some project die.
Deep Black, Shattered Horizon, PlanetSide 2, Warmachine Tactics, The Secret World are only few examples of games with promised, but not delivered hardware PhysX effects.

What whould physx indicator say if some physx is done on CPU and some on GPU?

PhysX indicator proves little. It is linked to basic CUDA context initialization or something like that. Even if it shows "GPU", it does not actually means GPU is actually crunching some physics calcs.

Too bad PhysX 3 does not support profiling tools like AgPerfMon anymore, so we can't do Mafia 2 style tricks
I think I can pull some old contacts and ask about the situation directly)
---------
Zogrim,
PhysXInfo.com

 

[el etro]

Crysis 3 and BF4 is much better well threaded than pCARs.

 

[Stuka87]

The title is surely a joke. This game looks very poorly threaded, especially compared to many other games.

The game looks pretty decent, but seems like it needs more optimizations considering how it runs.

 

[CubicZirconia]

This thread reminds me of the GTA V thread... "AMD gets hammered again" and then I get the game and it plays great. I'm not too interested in racing / car simulator games, but if a low of ~50FPS is all I have to worry about, I'm ok.

A lot of people are doing much worse than a low of 50 FPS, myself included.

 

[jhu]

Because ipc of fx is not overall lower than Phenoms...its slower only in some x86 code, but not in all variations. New games are ussually better at Fx cpus than at Phenoms 2.

This is really easy to figure out: somone check the game binary for AVX instructions.