rFactor 2 vs iRacing SEBRING with MSI Afterburner Data

They are using extra cores now for graphic stuff (or was it only for shadows?), it was said some time ago. There was even a graphics bug on CPUs with 6 cores, they fixed it in last update

 

I simply testet the performance with 2 and 4 cores and noticed a clear performance increase.

 

Pick below is I7 8700k CPU usage in rFactor2. CPU1 - CPU6 are I7 8700k physical cores. Could you please explain to me why CPU1 usage is 72%, CPU2 usage is 0%, CPU3 usage is 100%, CPU4 usage is 0%, CPU5 usage is 30% and CPU6 usage is 0% if rFactor2 has true quad core support?

 

@Ari Antero Who said anything about 'true' quad core support? They haven't somehow split the physics or main graphics threads up, they only said Dx11 has allowed them to 'take advantage of multiple threads', and they mentioned shadows as a primary example of something that has benefited from using multiple threads. As mentioned above the recent bug with 6 core CPUs was also indicative of more than just 2 cores being used.

No, it's not being spread evenly across 3, 4, 5, ... cores, but it's no longer true to say that rF2 only ever uses two cores.

@Andregee 4 cores was always better than 2, but only because it meant background tasks weren't sharing the same cores as the game. The game itself was only using 2 cores, until the recent dx11 changes.

 

Technically, even until recent changes more than 1 thread was involved in rendering. At least with Nvidia, there are multiple driver threads running within rf2.exe process, and I believe that's even more true with DX11. Yes, that doesn't mean high load on all cores, but there's many more threads at work than just 2.

Now, how much FPS boost there is by going from processor that can execute 4 threads simultaneously to 6 vs how much impact single thread execution has, that's another question. But if we saw shadows parallelized, I won't be surprised if it is just the beginning. Further parallelization might be as well the only way to achieve good performance with high FOV headsets.

 

Exactly. Ever since Nvidia and AMD released multi-threaded drivers four or fives years ago, you would typically see the CPU load split between the cores when gaming. But this multi-threading is simply because the Nvidia driver works things out cleverly and divides the load, it doesn't mean that the game itself is multi-threaded. Though at this point as @Lazza mentioned, there was a specific mention in the build notes that the game now multi-threads things like shadow rendering. On the other hand, Ari's screenshot shows maxed out CPU usage for one core, so there is probably still some room for improvement.

But anyway, I think it would take some more expert person to comment on what the difference is between Nvidia driver dividing the load and native multi-threading support.

 

My graphics skills are forever stuck with DX8 days, but driver multithreading won't take care of some scene setup parts like managing vertex buffers etc. That is where the most parallelization potential lies (for graphical part), in scene setup. And for that, developers need to do things in a special way. It seems like rF2 started to move in that direction, and this is great news for all of us.

 

I guess you'd expect CPU load to be 100% on at least one core unless you limit FPS. Something has to be a bottleneck when running at max speed, so unless you manage to get equal load on multiple cores (so they're all on 100%) you'll end up with one higher than the others.

 

Limit FPS, or, have GPU that's not fast enough at given settings. As you say, something has to bottleneck.

 

When game have two core support it can not spread evenly across 4 or 6 cores.
i7s which have 4 physical cores with Hyper-threading turned on this results in 8 logical cores.
The reason why we call these logical or virtual cores is because they aren't full cores in the sense that they don't have their own dedicated execution units but they do have their own dedicated fetching and decoding pipelines.The idea is to keep the parts of the processor that does all maths/logic (execution units) occupied and not wasting time idling. With hyperthreading each physical core has 2 pipes attached to it where it can execute 2 threads.
If you look the pick in my post # 43 again you may understand why I claim that rF2 has two core support.

 

Update was 6 core CPU hyper threading graphics bug fix. I would love to get also 10 core hyper threading graphics bug fixed

 

Hmh... Ari you sure you don't have aggressive threading switched on or disabled multithreading from Nvidia CP? For me the CPU load on Sebring is almost equally divided, but I have only 4 cores to test with...

 

I7 8700k don`t seem to have equally divided CPU load. Usually CPU3 has 50-80% load.

 

Isn't weird to have it equaliy divided? I mean you'd expect the core running physics to be way higher than the others, that would be normal

 

im pretty sure when looking at MSI afterburner cores 1 , 3 ,5 ,7 are our physical cores and 2, 4 ,6 etc are your virtual cores. in this one could assume hyperthreading does do anything in RF2 according to the results?

I do believe RF2 is doing something with threads 5, 7, whether this is shadows or whatever. if it was only using 2 cores, wouldn't we see the behavior or cores 1, 3 maxed out? a nice test would be to disable cores in bios and see difference in performance.

in my screenshot attached core 1 is 38% core 3 is 59% core 5 is 19% core 7 is 20%.
so still not alot happening but better then DX9

 

Still playing around with MSI Afterburner OSD...thought I would load up some other Sims and see how they go...again just a noob and not very scientific....I can get very good playable results (I am happy with this) on most Sims with my GTX 690 / i7 3770k / 5760x1080 res...results below.
I did also try 'SimBins 'Race Injection' and Sector 3 Studios 'Raceroom but could not get a screenshot with the MSI AB data for some reason, even though it was displayed while playing them.
Some interesting numbers though...well, for my anyway...lol...PCARS 2 getting all the CPU Cores going and Assetto Corsa did as well even though core 7 was 0% it did fire up later.
Assetto Corsa

Automobilista

Dirt Rally

SimBin GTR2

Project Cars 2

Wreckfest

 

Does anyone know the latency threshold for the physics and graphics to run in unison,I know iracing is 16miliseconds.

Over on iracing they claim that they will never be able to split the graphics and physics to run on separate cores using parallelisation due to latency constraints of real-time sims. and it’s not upto the game devs to figure this out,it’s upto the main language creators.

Iracings used two cores when I was there,one for physics, graphic and sounds,the other core had windows and some other low level stuff afaik,they are looking into moving sounds to a separate corr in the future in the hopes of gaining some performance.

But David tucker is adamant that until parrallelism can work with such low latency requirements,16mls or below,then separating physics and graphics will only be a bad thing for low latency sims

 

The folks behind iRacing have tied physics to graphics since GPL. Back then getting 30 fps was crucial to achieving smooth gameplay.

 

Kunos from assetto Corsa made a video about this very subject and how assetto Corsa has achieved some results. As u can see there core utilisation is pretty decent. He talks in depth about parallisation etc. I had the video posted in another thread I'll see if I can find.

 

There were not any performance decreasing background tasks while I did my tests and even with 2 cores, there was CPU Headroom availeable. Using only 2 cores via Taskmanager, with the others in Idle to take over background tasks gave the same result I don't know why, but 4 cores were allways faster in Rfactor 2