[computerbase] Project CARS benchmarks

[Face2Face]

Looks like I'll stick with Assetto Corsa...

 

[sontin]

Project Cars runs better on Windows 10 with the latest driver - the performance is 20-25% better than on Windows 8.1: http://www.computerbase.de/2015-05/...ch/3/#abschnitt_amd_radeon__gibt_es_loesungen

 

[ShintaiDK]

Project Cars runs better on Windows 10 with the latest driver - the performance is 20-25% better than on Windows 8.1: http://www.computerbase.de/2015-05/...ch/3/#abschnitt_amd_radeon__gibt_es_loesungen

Yep, its quite clear its an AMD driver issue. Also that pre 15.4 drivers also perform better.

Its not exactly a new AMD thing either, that people have to sit and swap drivers depending on game due to lost optimizations

 

[stahlhart]

Thread cleaned and reopened.
-- stahlhart

 

[Alatar]

Now that this is open I'll just leave this here:

http://www.overclock.net/t/1554407/various-project-cars-pc-benchmarks/90#post_23886414

AMD perf jumps crazy amounts after you raise power limits from stock. With stock power limits in pCARS the cards hit the limit and downclock like crazy all the way down to 500mhz. (at least my 7950 did)

And that would be why most reviewers probably see terrible performance (because they run stock) and why actual users who overclock are seeing perfectly fine performance.

 

[RussianSensation]

Project Cars runs better on Windows 10 with the latest driver - the performance is 20-25% better than on Windows 8.1: http://www.computerbase.de/2015-05/...ch/3/#abschnitt_amd_radeon__gibt_es_loesungen

Update:
"H has now also carried out a benchmark run under Windows 10 to the latest Windows Update by partly filled with Catalyst drivers for the not yet released operating system. And lo and behold: Project Cars running faster. Even when compared with the same driver under Windows 8.1, the performance increases depending on the scenario by ten to 13 percent. compared to Windows 8.1 with the Catalyst 15.4 Windows 10 works by as much as 20 to 25 percent faster! Unclear, however, is still why the same driver works much faster than on Windows 8 on Windows 10."

Very strange game overall. Performance with rain esp. is all over the place and even Kepler performance makes no sense:

1. 770 has perfect scaling from 680 (+10%) but 780Ti is barely faster than a 780 (+7%). No way does that make sense.

2. 960 OC is within 6% of 780Ti. This is basically impossible in any PC game made until today. Even if we account that Maxwell has 35% increase in IPC, there is no way for a 1024 CUDA core Maxwell card with 2GB of VRAM to be just 6% within a 2880 CUDA core 3GB Kepler.

3. Titan X OC barely scales vs. 980 OC, just 2% faster.

http://www.computerbase.de/2015-05/project-cars-guide-grafikkarte-prozessor-vergleich/2/

AMD card performance is totally inexplicable too. There is only a 1% scaling between 290 and 290X and 290X OC is just 15% faster than a stock 280X. Not only that, but Fermi GTX580 beats 7970 and is 53% faster than the HD6970. Looks like there is something in this game that stops all AMD cards from working at 100%. GPU usage in videos shows AMD cards do not reach 100% GPU utilization at all.

 

[mikk]

GameWorks failed to recognize amd beta win10 drivers as amd driver and didn't execute amd performance shader paths.

This is purely nonsense. There is no Gameworks feature in this game. AMD massively failed, driver overhead too big. Suddenly they are starting to work on this game, this is so typically for AMD. Project Cars wasn't important enough for them apparently.

 

[3DVagabond]

This is purely nonsense. There is no Gameworks feature in this game. AMD massively failed, driver overhead too big. Suddenly they are starting to work on this game, this is so typically for AMD. Project Cars wasn't important enough for them apparently.

Well the dev said they just sent AMD 20 game keys for them to optimize their hardware. Such cooperation is astounding (not!). "Here have some game keys. See if you can figure out what we did that killed your performance since your last driver you sent us".

 

[cmdrdredd]

Well the dev said they just sent AMD 20 game keys for them to optimize their hardware. Such cooperation is astounding (not!). "Here have some game keys. See if you can figure out what we did that killed your performance since your last driver you sent us".

The older driver worked better. The game devs don't know what changed in the driver do they? I bet with an older driver the performance is better but who wants to have to test multiple drivers for each game they might play?

 

[ShintaiDK]

Well the dev said they just sent AMD 20 game keys for them to optimize their hardware. Such cooperation is astounding (not!). "Here have some game keys. See if you can figure out what we did that killed your performance since your last driver you sent us".

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.

 

[mikk]

Well the dev said they just sent AMD 20 game keys for them to optimize their hardware. Such cooperation is astounding (not!). "Here have some game keys. See if you can figure out what we did that killed your performance since your last driver you sent us".

AMD has access to the game since over 3 years. They also they were involved in the game development. Obviously their involvment wasn't enough from AMDs side and their Windows 7/8 drivers weren't optimized for this game. Unlike Nvidia who released some driver updates to minimize the driver overhead last year. In pcars it was a big help.

 

[Silverforce11]

Well, if you want to take things out of context to fit an argument be my guest. Just don't expect to be taken seriously.

AMD was involved in the development of this game which was mostly crowd funded to begin with. After a while AMD support/communication dropped off and a new driver started performing worse. Why? Who knows but it's not up to the developer to beg AMD to write a driver that works.

I wonder if backers who had access to alpha can shed light on why their communication with AMD dropped off.. I think it may explain the sudden appearance of NV logos everywhere in the game.

It's like the same thing in Batman & AC from 2008/9. Ubisoft at the time said "We didn't receive a dime from NV to make this game".. technically he wasn't lying, because they eventually found out from marketing, they got $2M from NV to help them market (not make) the game. The Batman was even more funny, with actual email exchanges on the public record as proof. AMD wanted to send in their own AA code, but the devs refused, saying legal department says no, they can't accept it, since they are already using NV's AA code.

As to Prj Cars, the dev is full of bile against AMD users in an old thread from early 2014 on Steam, when backers told them of poor performance on AMD, they refused to fix it but instead, blame AMD drivers.

All of this would fully explain the horrendous performance of Project Cars on AMD.

 

[railven]

Turns out all versions of the game run poorly on AMD hardware.

PS4 version:

Weird ghosting/blurring issue.

XBone version:
"Glitch on Xbox one self.pcars
Submitted 54 minutes ago * by jnrdingo
There is a glitch in my game where if I go off the track or brake, I lose all sound except backfire..
Edit: issue is with force feedback, turn it off and you are golden."

Bold on solution, which is just odd.

Both console versions suffer dips in performance and tearing out the ying-yang:
https://www.youtube.com/watch?v=tb4JIBaGBWU


Reading a little history on the dev side, seems some of these issues (PC specific since there were no early builds for consoles) were known and mentioned. But ignored. There is currently some kind of backlash at the main dev for the series (guess he has a sort of bad-track record, his team was behind Shift 1 and Shift 2, and you guys might remember that game's performance)

http://www.pcgameshardware.com/aid,...-2-accelerates-Radeon-graphics-cards/Reviews/

Need for Speed: Shift Patch 2 - Background
When Need for Speed: Shift was released, we criticized the surprisingly low performance of AMD's Radeon cards. Especially in scenes with many vehicles the framerate was bad - no matter if the resolution was set to 800 x 600, 1920 x 1200 or any other resolution. The first time the racing game was updated, the problem was not solved, but the second patch for Need for Speed: Shift delivers more frames per second - according to the readme because of "Improved ATI graphics card support”. The reason for the up to now poor performance of Radeon cards has not been unveiled although rumors say that certain Shader routines had not been optimized. Honi soit qui mal y pense - Shift is part of Nvidia's TWIMTBP program.

Shift 1 was also a Nvidia sponsored game and required a patch from the dev to fix Radeon performance, and believe it or not Shift 2 Unleashed had some similar issues that required a driver update and game patch to resolve.

Seems just bad code, from a team who's now only 3 games made to date heavily favored Nvidia (be it Nvidia involvement, dev bias, whatever).



On the Win10 discovery, I wonder if it has anything to do with Windows 10 changing how GPU's are addressed (if I remember right, aren't they doing some kind of homogeneous GPU ecosystem - how they can claim SLI/CFX regardless of GPU brand/maker?)

 

[cmdrdredd]

I'm not convinced the dev knows how to do a console game. They put settings in the game. They should have made sure the game runs the same for everyone on the console. If that means 1080p 60fps with some slightly reduced details then so be it. Adding optional settings means that when you are playing online someone can be running 30fps and someone 60fps. I'm not sure if that would affect the netcode on the consoles negatively. Maybe they transferred/ported over too much code from Windows.

As for controller issues (rumble feature turning off sound) there have been reports of the game not functioning correctly with a controller on PC. I think they still have some work to do to the overall game to begin with.

 

[mikk]

I wonder if backers who had access to alpha can shed light on why their communication with AMD dropped off.. I think it may explain the sudden appearance of NV logos everywhere in the game.

Contact didn't drop off. They have exchanged several hundred emails over the course of the project with lot's of communications around March and just before release. As Ian said they were invited to work with them for years.

 

[railven]

Contact didn't drop off. They have exchanged several hundred emails over the course of the project with lot's of communications around March and just before release. As Ian said they were invited to work with them for years.

At this point with some of the screen caps of the dev (Ian), what he says holds as much water to me as anything AMD could say. It's a bad case of "he said she said" but the fact that other issues (not even performance; such as handling, physics, etc) were all feedback to the devs as far back as 2012 only to still be persistent release date in 2015 tells me the devs were either A) over their heads or B) not listening.

EDIT: And worse, in light of the whole "hush-hush" Barbra Streisand bullblop the gaming communities have been experience, trying to silence negative criticism is not what you want to do right now:

Ooops forgot link of who they're trying to shut up:
http://pretendracecars.net/2015/05/06/the-community-assisted-review-of-project-cars/

 

[sontin]

Edit - More info below. Lot's of communications with them around March and more yesterday.

Gains that were made with an earlier driver they released were lost in a later release. Our internal analysis shows only very marginal gains available from our side (around 1%) with an excessive amount of work.

We've had emails back and forth with them yesterday also. I reiterate that this is mainly a driver issue but we'll obviously do anything we can from our side."

Some great gains we saw from an earlier driver they released have been lost in a later driver they released. So I'd say driver is where we start.

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

"And a quick update from me, since today I did some profiling with the various driver versions, all running on Windows 7SP1, i7-5960X@3.5ghz:

Brands, 9AI, 5PM, BMW GT4, graphics settings at High except reflections at Medium:

Driver v14.12 - 59FPS
Driver v15.4 - 67FPS
Driver v15.2 - 72FPS (engineering sample 15.200.1018.1 from Guru3D)

[...]

Note that this isn't conclusive yet - their engineering team are also doing some analysis across driver versions, and also another more general review (although it should be said that we've exchanged several hundred emails over the course of the project between the two teams, with many previous reviews and re-reviews as we've made progress during the games development)"

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

AMD is the only one to blame for the performance.

 

[railven]

http://hardforum.com/showpost.php?p=1041593386&postcount=59
[/I]
[/I]
http://hardforum.com/showpost.php?p=1041593400&postcount=60

AMD is the only one to blame for the performance.

Confused why his scores are almost double the score of OP? If those were the scores on the OP I don't think we'd even be having this conversation(s).

 

[railven]

http://hardforum.com/showpost.php?p=1041593386&postcount=59
[/I]
[/I]
http://hardforum.com/showpost.php?p=1041593400&postcount=60

AMD is the only one to blame for the performance.

Wow, why didn't you include this part?

Again, if there's anything we can do we will.

I've now conducted my mini investigation and have seen lots of correspondence between AMD and ourselves as late as March and again yesterday.

The software render person says that AMD drivers create too much of a load on the CPU. The PhysX runs on the CPU in this game for AMD users. The PhysX makes 600 calculations per second on the CPU. Basically the AMD drivers + PhysX running at 600 calculations per second is killing performance in the game. The person responsible for it is freaking awesome. So I'm not angry. But this is the current workaround without all the sensationalism.

That would definitely explain why Win10 with DX12 would see huge gains.

EDIT:

You also left this part out:

Finally this is what they are basing the assertion that it is an AMD driver issue on. Straight from the software engineer at SMS whom I think is a wonderful person and knows more about this than 99.9% of us. Note that they are NOT doing this test with the rain effects on, there are only 9 AI on the track, and the game is running at HIGH / Medium settings; not Ultra which Nvidia's slowest card (GTX 560Ti) from 4 generations ago will run "laps" around AMD's fastest card in this game currently. Also they are testing on a i7-5960X and it can't handle the load of the game and the AMD driver

Seems like they found the issue.

EDIT #2: IE this is a game issue, unless AMD can optimize PhysX code? I guess reducing CPU over head (DX12/Mantle) would also help. I wonder if GeForce cards get an even bigger boost in Win10 - anyone know if they tested that?

 

[cmdrdredd]

At this point with some of the screen caps of the dev (Ian), what he says holds as much water to me as anything AMD could say. It's a bad case of "he said she said" but the fact that other issues (not even performance; such as handling, physics, etc) were all feedback to the devs as far back as 2012 only to still be persistent release date in 2015 tells me the devs were either A) over their heads or B) not listening.

EDIT: And worse, in light of the whole "hush-hush" Barbra Streisand bullblop the gaming communities have been experience, trying to silence negative criticism is not what you want to do right now:

Ooops forgot link of who they're trying to shut up:
http://pretendracecars.net/2015/05/06/the-community-assisted-review-of-project-cars/

Read through that and yikes what a mess.

 

[railven]

Read through that and yikes what a mess.

Yeah, just seems the more I read about this game, it seems to be a mess (mind you not an absolute garbage game).

With so much hype following this game, many skeletons are going to be falling out of closest mostly because it was a kickstarter game. And peeps on kickstarters want answers! Haha.

Welps, I've done my morning readings off to do something constructive. Do hope they can fix what seems to be CPU issues on the AMD side. Turning off some of the PhysX options seems to alleviate a lot of the issues. Par for course, though.

 

[Paul98]

Looks like a lack of testing, and unwillingness to correct and recognize problems.

 

[mikk]

At this point with some of the screen caps of the dev (Ian)

This is not complete or I should say outdated.

Confused why his scores are almost double the score of OP? If those were the scores on the OP I don't think we'd even be having this conversation(s).

Not sure why are you confused here. These scores have nothing to do with each other. Different CPU/track/weather/car/visual options may result in different fps scores, especially in this case from different people.

 

[mikk]

The PhysX runs on the CPU in this game for AMD users.

It runs on CPU also for Nvidia users. It's not the first game out with CPU Physx, so what is the deal here? The 600 calculations per second thing is unrelated to Physx since it is used only for collisions. SMS is using their own physics engine for the tires and suspension physics on cars.

 

[Silverforce11]

It all makes so much sense now:

"The game runs PhysX version 3.2.4.1. It is a CPU based PhysX. Some features of it can be offloaded onto Nvidia GPUs. Naturally AMD can't do this.

In Project Cars, PhysX is the main component that the game engine is built around. There is no "On / Off" switch as it is integrated into every calculation that the game engine performs. It does 600 calculations per second to create the best feeling of control in the game. The grip of the tires is determined by the amount of tire patch on the road. So it matters if your car is leaning going into a curve as you will have less tire patch on the ground and subsequently spin out. Most of the other racers on the market have much less robust physics engines."

The devs built their game engine around PhysX, which on NV GPUs, can be offloaded from CPU -> GPU. But on AMD that cannot happen, and the CPU is hammered.

Way too sneaky, making people think AMD has bad drivers that can't handle "draw calls". DRAW CALLS. lol

Theres no goddamn way a game such as that gets 1M draw calls per second to flood AMD's driver. It's running PhysX simulation on CPU for AMD vs CPU + GPU on NV.

The devs fully know what the problem is, as I stated, they knew from alpha feedback that it ran poorly on AMD. They did not care, cannot be fixed because PhysX cannot be GPU accelerated on AMD. Instead, they blame AMD drivers. What a bunch of un-ethical pricks.