[computerbase] Project CARS benchmarks

[Erenhardt]

FYI:



http://techreport.com/news/27910/nvidia-physx-joins-the-free-source-party

You have to make an account and accept an EULA but it's not like AMD can't see PhysX source if they really want to....

And? They can't change that code. It will run on the cpu anyway. What is the difference? it will eat your CPU regardless if you know every single line of code from your memory... it will still eat your cpu.

 

[cmdrdredd]

It is really awful if you are a baker and lateron the devs devide to incorporate this Gameworks things. Would it be possible to withdraw the money? Because the money was paid to develop a game and not to include this GW stuff that principally excludes all bakers on AMD cards from gaming. I hope the devs will get what they deserve.

Does the game run? Then you aren't excluded from gaming.

Way to exaggerate.

 

[Vesku]

The dev studio only ran a full profiling of the game on an AMD card after launch when people started to complain. Where they say it is completely CPU bound. Insist the issue is only 1% on them 99% on AMD, even though they don't say how much of that CPU use is PhysX calculations and how much AMD driver activity. They also initially misrepresented how much AMD tried to work with them.

Am I missing anything?

http://hardforum.com/showpost.php?p=1041593386&postcount=59

http://hardforum.com/showpost.php?p=1041593400&postcount=60

 

[Alatar]

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:

 

[Flapdrol1337]

Then agian that's a lot of cpu to burn.

 

[3DVagabond]

I think you passed way over in tinfoil country.

Whats your view on the large performance difference within AMD depending on driver version and power setting? Pretty much contradicts your entire post.

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

 

[Silverforce11]

http://steamcommunity.com/app/234630/discussions/0/613957600537550716/#p1

Interesting.

 

[SPBHM]

I remember Shift or Shift 2 also having some strange performance problems with Physx.

 

[lilltesaito]

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:

http://steamcommunity.com/app/234630...0537550716/#p1

Interesting.

Very interesting indeed.

 

[Silverforce11]

I remember Shift or Shift 2 also having some strange performance problems with Physx.

That's no coincidence. The lead dev was responsible for NFS Shift 1/2. He has a very anti-AMD history and his steam responses to gamers on AMD setups clearly shows his vitriolic side.

 

[Azix]

are there any of the usual physx effects in the game? Or is it just "physx"

 

[SPBHM]

are there any of the usual physx effects in the game? Or is it just "physx"

I think they have the same physx usage on the consoles as they do on the PC, so no extra stuff you are used to see on games like "Borderlands",

 

[railven]

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:

Quick question, since I didn't see if you posted it, but what CPU were you doing your Radeon testing on?

And did you notice any huge load spikes on the CPU?

Thanks for testing it. Wonder if anyone else cares to chime in with their findings. Some else here surely owns this game. I'm tempted to ask a Steam friend to let me access their account to test it.

 

[DarkKnightDude]

Thanks for testing it. Wonder if anyone else cares to chime in with their findings. Some else here surely owns this game. I'm tempted to ask a Steam friend to let me access their account to test it.

I'll give it a shot right now with my GTX 770 setup and see if physx affects it. I own the game, but so far been trying to set up my controls mostly.

EDIT: Basically did a quick lap for each test. 16 AI, Rain on, 3PM in Donington Park GP.

The game runs at 24-28 FPS with CPU Physx forced on through control panel.

With default GPU PhysX settings, it seems to be about 36ish-40 FPS. There's a noticeable difference there for me. Need to do a better graph test with MSI or something, but not sure how other posters do it. I was using DXTORY to gauge my framerate.

 

[RussianSensation]

From another thread:

"No, in this case there is an entire thread in the Project Cars graphics subforum where we discussed with the software engineers directly about the problems with the game and AMD video cards. SMS knew for the past 3 years that Nvidia based PhysX effects in their game caused the frame rate to tank into the sub 20 fps region for AMD users. It is not something that occurred overnight or the past few months. It didn't creep in suddenly. It was always there from day one.

Since the game uses GameWorks, then the ball is in Nvidia's court to optimize the code so that AMD cards can run it properly. Or wait for AMD to work around GameWorks within their drivers. Nvidia is banking on taking months to get right because of the code obfuscation in the GameWorks libraries as this is their new strategy to get more customers.

Break the game for the competition's hardware and hope they migrate to them. If they leave the PC Gaming culture then it's fine; they weren't our customers in the first place."
http://hardforum.com/showpost.php?p=1041593372&postcount=58

Does the game run? Then you aren't excluded from gaming.

Way to exaggerate.

Coding the entire physics of the game using PhysX, a proprietary NV physics engine that is offloaded to the GPU, while leaving Intel/AMD GPU users for dead? That's your idea of great PC game development? Are you kidding?! PC gaming was always about the developer making the game either accessible to as many gamers as possible (Blizzard - the gold standard), OR pushing the technical boundaries for years to come with no particular goal to favour anyone but PC gamers who would be excited for upgrading to "max out" a next gen game. (Crysis 1).

Who the hell codes a game from the ground-up on a proprietary source code that runs on a GPU for 1 vendor but can only be executed on the CPU for Intel/AMD users? Insane. Did they not have enough $ to create their own brand agnostic physics game engine? Pffft.

The reason it was never fixed even when their alpha backers told them is because it CAN'T be fixed. They built their entire game around PhysX.

That explains how AMD's R9 290X is never at 100% GPU usage as the game is so CPU starved with AMD cards. Performance goes from 24 fps to 23 to 37.4 fps by just swapping an FX8370 to an i7 4770K with the same 290X. 980 with the same slow FX8370 is hitting 54.5 fps.
http://www.computerbase.de/2015-05/project-cars-guide-grafikkarte-prozessor-vergleich/

It runs on CPU also for Nvidia users.

Umm...clearly most of the heavy stuff is being offloaded to the NV GPU. It's already been posted by the developer n the forums that PhysX is being offloaded to the GPU for NV.

FX8370 + 290X = 23 fps
FX8370 + 980 = 54.5 fps

Let's not kid ourselves here - under no circumstances can a 980 be 2.4X faster than a 290X unless something is artificially holding back the 290X on that 8370 rig.

Also, swapping FX8370 to an i7 4770K only increases performance 15% on the 980 but a whopping 63% for the 290X. If in NV's case, if the CPU was also responsible for all of the PhysX code, we would see a massive increase in performance too by just swapping out the 8370 with the i7 4770K with the 980 pairing.

 

[Udgnim]

I'll give it a shot right now with my GTX 770 setup and see if physx affects it. I own the game, but so far been trying to set up my controls mostly.

EDIT: Basically did a quick lap for each test. 16 AI, Rain on, 3PM in Donington Park GP.

The game runs at 24-28 FPS with CPU Physx forced on through control panel.

With default GPU PhysX settings, it seems to be about 36ish-40 FPS. There's a noticeable difference there for me. Need to do a better graph test with MSI or something, but not sure how other posters do it. I was using DXTORY to gauge my framerate.

assuming your #s are accurate between GPU & CPU-only PhysX, AMD isn't going to be able to fix their performance issues then

 

[DarkKnightDude]

assuming your #s are accurate between GPU & CPU-only PhysX, AMD isn't going to be able to fix their performance issues then

I'll do another run just to be sure with a different map with rain off or something. Need to quickly figure out how to set up a graph though for FPS numbers get through benchmarking with FRAPS or MSI.

There's definitely a notifiable difference though. I think its safe to say AMD is being crippled by the HUD and PhysX.

 

[RussianSensation]

There's definitely a notifiable difference though. I think its safe to say AMD is being crippled by the HUD and PhysX.

The HUD is nearly a 50% performance hit. How in the world did the developer not notice this but a random internet gamer recognized it just days after launch? Unbelievable.

http://forums.anandtech.com/showpost.php?p=37388928&postcount=256

AMDs last communication with them was in october. It seems AMD simply shifted whatever resources it had left back then to Omega and the like and considered Project Cars for done. The company simply sent AMD 20 keys in the hope something would happen since AMD didnt reply. And funny enough the first communication from AMD happens on twitter.

There is only one company to blame, and thats AMD.

Thats there is a recession in performance with newer drivers shows everything. So AMD had some optimziations for the beta that got lost along the way. And thats not a new thing either with AMD.

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:

 

[Silverforce11]

assuming your #s are accurate between GPU & CPU-only PhysX, AMD isn't going to be able to fix their performance issues then

This won't be something AMD or devs can fix. Its because of heavy usage of PhysX in their game engine from day 1, they knew performance would be horrible for AMD.

The console version is toned down to a point where console players complain the game feels like an arcade and not a racing sim. Its physics is already messed up. On the PC, PhysX is run to the max, their software engineer knew about it, he was the one who fessed up on their forums. His suggestion for AMD users is to tone down all the settings for now. No other way around it.

Under no normal situation would a gtx660 smack down on a R290X except for PhysX being GPU accelerated vs CPU only. This is the exact thing we've seen for years with running PhysX on AMD, and in Project Cars, the smoke is GPU PhysX accelerated. Do a burn out on an AMD rig, slideshow.

@DarkKnightDude Try it and see.

 

[RussianSensation]

The console version is toned down to a point where console players complain the game feels like an arcade and not a racing sim. Its physics is already messed up. On the PC, PhysX is run to the max, their software engineer knew about it, he was the one who fessed up on their forums. His suggestion for AMD users is to tone down all the settings for now. No other way around it.

Well it's going to be a long time before a CPU can run PhysX as fast as the GPU, which means "for now" could turn into 10+ years.

Will be interesting to see what HardOCP says on this one and if TechReport and PCPerspective will include this title into their GPU testing and keep it for years as a 'worthy' GPU demanding title under the veil that it is currently the editor's favourite game series in the genre....like Borderlands franchise, over and over. Let's not forget TechReport publicly withdrew Dirt Showdown testing because it heavily favoured AMD cards.

 

[casiofx]

This won't be something AMD or devs can fix. Its because of heavy usage of PhysX in their game engine from day 1, they knew performance would be horrible for AMD.

The console version is toned down to a point where console players complain the game feels like an arcade and not a racing sim. Its physics is already messed up. On the PC, PhysX is run to the max, their software engineer knew about it, he was the one who fessed up on their forums. His suggestion for AMD users is to tone down all the settings for now. No other way around it.

Under no normal situation would a gtx660 smack down on a R290X except for PhysX being GPU accelerated vs CPU only. This is the exact thing we've seen for years with running PhysX on AMD, and in Project Cars, the smoke is GPU PhysX accelerated. Do a burn out on an AMD rig, slideshow.

@DarkKnightDude Try it and see.

SMS devs dare to say that AMD didn't correspond to them about developing this game, when in reality there's not a single way AMD can even run Physx in Radeon GPU

 

[RussianSensation]

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:

Hold on a second...if the game's engine uses PhysX and automatically offloads the calculations to the GPU seamlessly, I do not think you can control whether or not your GPU or CPU is performing the calculations. This isn't like a dedicated PhysX ON/OFF function in other games like Batman series. It was already posted that the game doesn't have ON/OFF switch for PhysX which means you can't just dynamically isolate PhysX how you are trying to do it. The NV driver along with the game engine has full control of how the game functions and in this case you can't just toggle PhysX CPU vs. GPU. It's no wonder your results are 99.9% identical in both cases.

SMS devs dare to say that AMD didn't correspond to them about developing this game, when in reality there's not a single way AMD can even run Physx in Radeon GPU

But you already know how some users are going to spin it:

1. AMD has no money to optimize drivers (ignoring the fact that AMD cards run vendor agnostic titles flawlessly)
2. AMD has crappy drivers since forever (ignoring that CF often has superior scaling and lower frame times latency than even Maxwell and that R9 290X has tightened its performances standing with 780Ti/980 in almost all non-GW titles in the last 12 months)
3. AMD fell asleep at the wheel because they didn't optimizing the PhysX in the game. It's their fault! (ignoring that PhysX cannot be optimized on AMD GPUs)
4. AMD has horrendous DX11 CPU driver overhead (ignoring that physics calcs are being done on the CPU, thus completely bogging down the CPU when paired with the AMD GPU, no matter if you throw an i7 4770K @ 5Ghz in there)

 

[Silverforce11]

Hold on a second...if the game's engine uses PhysX and automatically offloads the calculations to the GPU seamlessly, I do not think you can control whether or not your GPU or CPU is performing the calculations. This isn't like a dedicated PhysX ON/OFF function in other games like Batman series. It was already posted that the game doesn't have ON/OFF switch for PhysX which means you can't just dynamically isolate PhysX how you are trying to do it. It's no wonder your results are 99.9% identical in both cases.

Some users on steam have said when they left PhysX as CPU only, they got terrible performance and poor GPU load % like on AMD.

Also, @DarkKnightDude above said his performance dropped when it force it to the CPU as well.

This combined with the HUD tanking performance on AMD = gtx660 destroying R290X.

 

[RussianSensation]

Some users on steam have said when they left PhysX as CPU only, they got terrible performance and poor GPU load % like on AMD.

I didn't know that. Was just going off based on what he posted. So we need a 5960X OC to handle PhysX in Project CARS then?

Also, @DarkKnightDude above said his performance dropped when it force it to the CPU as well.

His benchmarks hardly reflect that. No?

This combined with the HUD tanking performance on AMD = gtx660 destroying R290X.

They should just rename this game PhysXCars or ProjectNVIDIA. This will be a good gauge to see which review site includes this title into their long-term / regular review testing.

The crazy part is the average Joe (most PC gamers) will never read forums such as ours or reddit or the ProjectCARS community forums and automatically assume AMD cards and drivers are garbage. That's why I think GWs needs to be boycotted and banned outright. I just can't believe TW3 is GW. :$

 

[Silverforce11]

Read above at his post again,

"The game runs at 24-28 FPS with CPU Physx forced on through control panel.

With default GPU PhysX settings, it seems to be about 36ish-40 FPS. There's a noticeable difference there for me."

GPU PhysX: 36-40 fps
CPU PhysX: 24-28 fps

Quite a big difference.