VR performance with 1080ti

Authense

So fast I am slow
Hi, I have been working on this issue for quite a while now and I have decided to reach out for your help now as I am running out of options:

System: latest Core I7 I7700 (no k, standard, but constantly running at 4 ghz), MSI geforce 1080ti, 16GB, Windows 10 64bit, HTC Vive. Fanatec CSR wheel and pedals

Tweaks tried so far:
reduced in game graphic settings to a minimum. low details, low shadows, low smoke etc.
Supersampling on (1,5-2,0) / off
MSAA forced in NVIDIA settings
Anti-Aliasing on/off/ reduced
render target multiplier maxed out - nice for the VR steam menu, but no effect on in game performance

Overclocking the 1080ti fails with Assetto Corsa crashing when entering VR every time.

On top of that, distant rendering is quite blurry and I was astonished watching some Videos of other folks claiming that this would be there actual lense view quality. I do not get that at all.

Symptom:
I am getting nowhere near the VR performance that others pretend to have both in resolution quality and FPS. I cannot get 90 FPS together wit other cars on the track. At the start it is around 65 and lower and during race high 70ties and some low 80ties. On a long straight without other cars around I can hit 90 FPS depending the track, i.e. Monza, but not in Spa for instance.

Whatelse can I do or try?
Is there a reference setup with which you get to 90 FPS on a specific track with XY cars on the grid for testing? Reason is I do not know what I should expect as I cannot find references, only people on Youtube claiming with their settings they would constantly hit 90FPS, well, I don't.

Of course, if I only do training, I also hit 90 FPS constantly. But in a race during overtaking for instance it is in the low 70ties and really bad in corners with two cars side by side.

Many thanks in advance,

Chris
 
You might alienate potential helpers by saying "I am getting nowhere near the VR performance that others pretend to have both in resolution quality and FPS", as that seems to be insinuating that these people are lying.

Personally I'm very happy with everything maxed in daylight on my stock 6600k/980Ti combo with 45fps locked and ASW forced. 20 car grids look and feel smooth. Some people cannot get on with ASW though, and you might be one of them.
 
You might alienate potential helpers by saying "I am getting nowhere near the VR performance that others pretend to have both in resolution quality and FPS", as that seems to be insinuating that these people are lying.

Personally I'm very happy with everything maxed in daylight on my stock 6600k/980Ti combo with 45fps locked and ASW forced. 20 car grids look and feel smooth. Some people cannot get on with ASW though, and you might be one of them.

The Vive doesn't support Oculus ASW. I guess SteamVR reprojection is the way to go here, but have no experience with the performance.
 
Couple of findings today:

  • Switched off all Windows Services not required (Virus, Firewall etc.)
  • Set NVIDIA settings to default
  • Increased in game Super Sampling to 2.4 (max) which alters render target multiplier to the same figure
  • Kept Super Sampling in Steam VR Advanced Settings at 1.0
  • Switched off everything I could find that Steam wants to do in the background
  • Switched off replay and such in game
Launched the game from within Steam VR and put 25 cars (all the same) on the grid. Monitored render stats during race.

Settings in game:
1920X1080, Full screen rendering, Vsync off, AA 16/4, low details, low mirrors, smoke, basically all set to minimum. Use Post processing to launch Natural_mod, overall quality is normal, glare and others off. Blooming off. G-force 0,1/0,4/0,1.

With simple (i.e. Lotus) originally from within the game I get 90 FPS all the time. With more complex cars that are downloaded from this site I get 65 in the beginning and then around 80 during race. With more complex original cars I get to 86-90 FPS.

But with one of the downloaded Ferraris (F2002) I also get 90 FPS with 25 such cars on the grid. It doesn't look less complex to render than the McLaren MP4 which seems to be a lot more demanding on FPS.

FFB has impact. As soon as I have an accident and FFB starts to simulate a crappy car the frame rate drops to 70 FPS and does not recover.

The in game overlays have a huge impact on frame rate in and around 10 frames per sec.

The car model and some of the things the author does has impact on the frame rate.

In all cases: render stat shows low usage (3% to max 14%).

Chris
 
Hi,
to help you at all we need to find out if your CPU or GPU are the limiting factor. If we don't know that you can mess up the settings as much as you want without any real effect!

So please download "openhardwaremonitor" (just a little exe, reading out temperatures,loads etc.) and log your graphics card frequency and usage. If your 1080 ti isn't running in 3D mode (~1500-2100 MHz) you should set the power management mode in the nvidia settings to "prefer maximum performance".
If your card isn't above 90% load you will know that the CPU is the limiting factor. (Which I think)
Although your cores will only show 10-50% load, your "single thread limit" will be reached. RAM or CPU overclocking helps but regarding in-game-settings that reduce CPU load:
- Disabling shadows (patcher or content manager)
- reflection frequency
- reflection rendering distance (content manager)
- mirror rendering distance (content manager)

And that's about it. All other settings barely have any influence on the CPU. One big factor is the amount of overlay apps you're running. Try to deactivate them all and you should see quite an increase in fps!

But you really should download the hardware monitor and check it!
It should look like this: (Started Assetto Corsa and the graphs went up.)
[hovering over the X -or Y-Axis will make that axis "zoomable" with the mousewheel. If you hover over the graphs, you will zoom both! Grabbing one of the Axis will make it "scrollable"]
OpenHardwareMonitor_Settings.JPG

upload_2018-6-18_17-13-22.png
 
You might alienate potential helpers by saying "I am getting nowhere near the VR performance that others pretend to have both in resolution quality and FPS", as that seems to be insinuating that these people are lying.

Personally I'm very happy with everything maxed in daylight on my stock 6600k/980Ti combo with 45fps locked and ASW forced. 20 car grids look and feel smooth. Some people cannot get on with ASW though, and you might be one of them.

You are right, not intending to sound like that but could be misinterpreted.
 
Ok guys, problem solved. First of all, many thanks for your directions.

My findings will not be new for the experts but may help some newbies later.

Conflict:
First of all there is a conflict between my Fanatec CSR/Pedals and the HTC Vive controllers / Steam VR dashboard.
Solution:
I know there are direct solutions to this, I might try them later
Bypass:
Launch AC from within Content manager, disable VR dashboard, do not activate Controllers

From there I defined a reference race with 24 cars (downloaded McLaren, Ferrari, Williams) in Spa, only launched the renderstat in game overlay and monitored what happens.

BTW, I also ran Hardware Monitor and CPU goes up to 80% at the start of the race. Thanks for that. I am sure, this CPU causes some limitations.

Anyway, to achieve a huge increase in sharpness and immersion quality of the game I made the following changes to the steamvrsettings.ini:

"compositor" : {
"renderTargetMultiplier" : 1.2999999523162842

This settings has no effect on the game, I can also set it to 5, I believe, this has an effect on the dashboard only. Any thoughts?

"steamvr" : {
"allowInterleavedReprojection" : false,
"basestationPowerManagement" : true,
"directModeEdidPid" : 43521,
"directModeEdidVid" : 53794,
"enableHomeApp" : false,
"installID" : "...",
"ipdOffset" : 0.0060000047087669373,
"mirrorViewGeometry" : "0 0 1024 600",
"renderTargetMultiplier" : 3,
"supersampleManualOverride" : true,
"supersampleScale" : 3

I played with these three settings forward and backward for many hours to understand the logic of these (to me at least) confusion and redundant names. I restarted Content Manager and SteamVR every time I made a change to the ini file.

SupersampleScale = in game OpenVR overlay, but that scale ends at 2.4
renderTargetMultiplier = not referenced in game but if I reduce it 1.1 with 3 for supersampleScale I get less frame rate on the grid and throughout the first corner. Don't know why.

Setting 3/3:
In Spa/EauRouge I am in the 80ties FPS and then on the straight at 89.5 FPS for 99% of the remaining race distance. It drops to 86-88 sometimes in corners but that does not hurt.

Now, interestingly, I can set them both to even 4 or 5 and get almost the same frame rate but more clipping and little more FPS reduction but only small given the SS increase. That does not hurt either, but the image does not get better with these settings. 3 seems to be the sweet spot.

Effect:
All in all, the SupersampleScale / render targetmultiplier setting of 3 gives a fantastic image quality, no shimmering on the edges of buildings anymore, way better view on cars in front of me and much less blurriness in the distance.

FFB:
Also, whenever I bump into another car and FFB tries to simulate an issue with the body of the car or with the tires, frame rate also drops a few FPS. As per the FFB issues and wheel I tried all my USB ports and but doesn't make a difference. I have also put the HTC Vive into a 3.0 USB ports as there were reports on issues with the 3.1 port.

There were other reports that Bluetooth with the Base stations might cause trouble but in my case it makes no difference so I leave it on.

One more thing:
Regardless the supersample settings, even with setting = 1.1, I do not get more than 75 FPS at the start of the race. Of course, if I pick a course and cars with less details I can hit 89.5 at the start.

To summarize, renderstat FPS figures during race are not affected by Supersampling setting, only mildly when going beyond 3 to 4 or 5. The image is fantastic with setting 3/3

What do you think about these findings?

Best
Chris
 
Just found out that the other one is the setting in SteamVR in Manual override of Supersampling:

"steamvr" : {
"allowInterleavedReprojection" : false,
"basestationPowerManagement" : true,
"directModeEdidPid" : 43521,
"directModeEdidVid" : 53794,
"enableHomeApp" : false,
"installID" : "...",
"ipdOffset" : 0.0060000047087669373,
"mirrorViewGeometry" : "0 0 1024 600",
"renderTargetMultiplier" : 3, (This one is "Manual override" in SteamVR settings)
"supersampleManualOverride" : true,
"supersampleScale" : 3 (This is OpenVR supersampling in game)

I honestly doubt that is it 3x3 with stable 90 FPS but as I said, it is not as crips without settings renderTargetMultiplier to 3 in here.

Don't know what this one does as I cannot find it nowhere at this moment:
"compositor" : {
"renderTargetMultiplier" : 1.2999999523162842

Maybe it would become visible under Advanced settings on the dashboard as "Supersampling" there, because I know that rendertargetMultiplier in AdvancedSettings equals SupersampleScale on the ini file.

That is all quite confusing really.
 
Hi,

So please download "openhardwaremonitor" (just a little exe, reading out temperatures,loads etc.) and log your graphics card frequency and usage. If your 1080 ti isn't running in 3D mode (~1500-2100 MHz) you should set the power management mode in the nvidia settings to "prefer maximum performance".
If your card isn't above 90% load you will know that the CPU is the limiting factor. (Which I think)
Although your cores will only show 10-50% load, your "single thread limit" will be reached. RAM or CPU overclocking helps but regarding in-game-settings that reduce CPU load:
- Disabling shadows (patcher or content manager)
- reflection frequency
- reflection rendering distance (content manager)
- mirror rendering distance (content manager)

And that's about it. All other settings barely have any influence on the CPU. One big factor is the amount of overlay apps you're running. Try to deactivate them all and you should see quite an increase in fps!

Hi Rasmus,

back to the FPS limits on the grid. GPU and CPU show 50% as you are saying. Therefore I only get 60-70 FPS at the beginning no matter the settings in game or the level of supersampling.

The 1080ti is waiting for the CPU. I have the non-overclockable version of the 7700 running at 4Ghz constantly and that is not enough horsepower for AC in VR to run at 90 FPS all the way.

Strangely, the plot shows that the load is constantly between 40% and 50% throughout the race but the FPS recover quickly and I get 86 to 89,5 all the way. I guess the 86 FPS are the sparks when the CPU hits 50%, mostly in corners and with many cars in front of me.

Thanks for your advise,

Chris
 
Don't know what this one does as I cannot find it nowhere at this moment:
"compositor" : {
"renderTargetMultiplier" : 1.2999999523162842

This one is the renderTargetMultiplier in Advanced settings when you have it on autostart. SteamVR keeps it on the ini file even if the dashboard and Advanced settings aren't running at all.

I have set the manual override to 3 SS as this gives me the best image quality in AC.

I reduce this to whatever I need for other applications on the application tab in SteamVR Settings.

For those who do not know, the manual override figure (whatever it is) becomes 100% for the application tab slider per app. So i.e. 3 SS in manual override and 50% for any application will result in 1,5 SS for that particular application.

The warning message is a little misleading underneath manual override, the manual override should be used and maxed out for the application that allows the highest SS setting. AC then sits at 100% on the application tab.

Chris
 
Hi Rasmus,

back to the FPS limits on the grid. GPU and CPU show 50% as you are saying. Therefore I only get 60-70 FPS at the beginning no matter the settings in game or the level of supersampling.

The 1080ti is waiting for the CPU. I have the non-overclockable version of the 7700 running at 4Ghz constantly and that is not enough horsepower for AC in VR to run at 90 FPS all the way.

Strangely, the plot shows that the load is constantly between 40% and 50% throughout the race but the FPS recover quickly and I get 86 to 89,5 all the way. I guess the 86 FPS are the sparks when the CPU hits 50%, mostly in corners and with many cars in front of me.

Thanks for your advise,

Chris
Yep that's definitely CPU limit!

If you really want to know what's going on you can do it like I'll describe below but you have to use a second monitoring or run assetto corsa in window mode. Both difficult in VR...
Anyway, here's how you could read out the real CPU usage:

1. Download Process Explorer

2. Run it, right click on acs.exe -> properties

3. Navigate to "Threads" and sort it for "CPU"

upload_2018-6-21_17-16-26.png


4. Sadly only in real time, you see the cpu usage per thread!

5. Now calculate the single thread limit (one core limit) by dividing 100% with the amount of threads your CPU got. For me that's i7 2600k: 4 cores / 8 threads. 100%/8 = 12.5%

6. The moment one of the shown Threads in the process explorer window goes above 12.0% (there might be a spike higher than that but it won't show to you), your fps will go down as the CPU limit is reached!

7. Although taskmanager etc. show a nice and even CPU usage, it's not the truth! Windows will "shuffle the single thread around", which results in a better performance somehow. Imagine it like you have 1 nail that needs to go into a wall. That's your single thread. Now it might go in faster if you grab 7 other guys, line up and everyone will do one massive hit on this nail.
But as you can imagine, there's a limit to that.

Imagine you now have 8 nails and you are still 8 guys. It will be faster if everyone grabs one and starts hammering than it would be if everyone hits each nail once. Now the absolute best would be if the guys who are done or have more energy would help the others with some nice big hits in between.

Windows gets nails from the games and manages the guys quite well. You won't get 100% usage of all guys with one single nail though.

Now what I can see at my pc is that the one thread is more or less constantly at 12.4X so really close to my CPU limit. When the load changes the fps go up but the thread will stay at 12.4X.
That's why you see the same CPU load but the fps are fluctuating.

8. Now I'll show you how video rendering looks like in comparison and you'll see why it can reach 100% CPU load:
upload_2018-6-21_17-18-0.png


9. The problem with racing sims is that you can't really split it up into more threads. Project Cars uses 1 core per tyre for example but in AC it's really difficult. They were a small team back then and the engine has it's limitations. The overlay apps are in the same thread as the rendering and the physics is is quite "bad". But if you try to think about it and really try to thing about "multi-threading" your daily life you will see that it's not as easy as one might think.

10. Cooking is a good example. You can cut and wash all ingredients in parallel but if you cook a soup you can not cook all ingredients on its own and throw them together after it. You need to do it in one pot so they mix nicely.

11. I think and hope that unreal engine 4 offers better tools to split things up so we probably won't have to buy a new CPU for it :)

12. If you now think a current gen i3 at 5.5 GHz would give you more fps: sadly not.. I tried to disable hyperthreading and overclock my CPU higher and the results were the same. 8 threads at 4.4 GHz are equal to 4 threads at 4.7 GHz.
The "shuffling between the cores/threads although it's only 1 real thread" gives a higher boost than pure clock on the single cores.
 
Thanks. Well, some people hit a constant 90FPS with I7700K@4.9Ghz and 1080ti as they demonstrate on Youtube as well as here and there. So I am wondering, should I get a K version and overclock it?
 
Thanks. Well, some people hit a constant 90FPS with I7700K@4.9Ghz and 1080ti as they demonstrate on Youtube as well as here and there. So I am wondering, should I get a K version and overclock it?
Depends on whether you can get it cheap while selling your non-k.
Then it depends on your Mainboard chipset and quality if you would reach such clock speeds.
Then you need to read quite a lot to overclock it safely and get some knowledge about your cooling. Air cooling is totally fine but the case shouldn't be such an oven like my old sound dampened one. Maybe case fans and CPU cooler would need to be replaced..

In general: yep you should. It's absolutely easy to overclock and optimize the airflow but only if you are willing to get into this stuff. If you never touched a bios or the inside of a PC I would absolutely not recommend it.

If your BIOS supports the 8700k and you have DDR4 memory I'd say it's probably better to just slam an i7 8700k in it, OC it slightly about 200 MHz and be happy and future proof.
Here you can compare the cinebench single thread and multi thread performances:
TechreportLINK
 
Hi Rasmus,

thanks for your advise.

In the meantime I continued optimizing and the answer has always been right in front of me.

It came to me the moment I accepted - with your kind support - that the CPU is the limiting factors. In earlier times, when games were more CPU heavy the number of cars on the grid was causing the lagging. I knew it all the way, because earlier on this thread I reported that I get 90 FPS with one car easily. LOL.

So I went back to my PC, loaded my reference track and reduced the number of drivers from 25 to 5 and then increased it again:

10 opponents = 90 FPS all the way
15 opponents = 90 FPS all the way
20 opponents = 75 FPS on the grid to 88/90
25 opponents = 62-66 FPS on the grid to 88/90

There a simply too many cars on the grid for this CPU. That also explains why none of the other settings including Supersampling made a difference on the FPS. The 1080ti is simply not even close to its limits.

With 15 opponents being the sweet spot in my setup I was then able to raise world details to Ultra, use post processing, drive FFB to 100% gain and, regardless, the FPS was still 90 (89,5/89,6) all the way. Before I was under the impression, especially because of tons of stuff I had viewed and read, that the graphical settings determine the FPS. Yes, Shadows do, but for most of them, AC is a 2014 game I guess and cannot stress out a 1080ti even in VR.

A very good test for other people could be to see what FPS rate they get with only one car on the track, increase #cars, and then in case the number of cars is the only variable and the FPS comes down, then it is the CPU for sure. This way, everyone can easily find the sweet spot and then max out graphic settings.

Agreed?

Chris
 

What are you racing on?

  • Racing rig

    Votes: 528 35.2%
  • Motion rig

    Votes: 43 2.9%
  • Pull-out-rig

    Votes: 54 3.6%
  • Wheel stand

    Votes: 191 12.7%
  • My desktop

    Votes: 618 41.2%
  • Something else

    Votes: 66 4.4%
Back
Top