VR performance in Racesims (including "next" gen VR)

[HoiHman]

I have been a VR enthusiast from the beginning.

I bought a Oculus Rift CV1 at launch day and it has served me well over the last 3 years, but.... the image quality and performance compared to a monitor is terrible.

So in those 3 years i upgraded from an Intel I7-4770K to an Intel I7-7700K, sold the GTX1080 and bought a 1080Ti.

For upcoming new HMD's i did another couple of hardware upgrades:

-The 1080Ti has been replaces by a 2080Ti
-The I7-7700K is overclocked to 4,7Ghz
-DDR4 memory runs now on 3200Mhz


My system gets a 11.000 score in VR benchmarks




Still i'm struggling with every single race sim between having a decent performance and good image quality. Even though my PC is "better than 92% of all results"

Most race sims are running most of the time locked at 45PFS with ASW enabled. Mainly because i run them with higher multisampling settings to compensate for the low image quality in the rift.

Yesterday i tried ACC.

The 1.0 release of Assetto Corsa Competizione is the worst of them all. I did a fresh install yesterday and set the game at the lowest settings, Steam VR settings are at 100% and i'm still not getting 90fps at the start in Spa with 20AI.

At those low settings the game looks worse than Grandprix 4 on a 1024x768 monitor more than 20 years ago.

The strange thing is that even with ASW disabled, i'm still getting not more that 65-70% load on both my GPU and CPU.

On my 34" widescreen monitor ACC runs like a dream and looks drop dead gorgeous.

I'm all in on the whole VR and simracing combo, but having suffered the low image quality of VR for the last 3 years i'm kind of growing tired of it.

Because i love VR, i invested in both a HP Reverb and a Valve Index and will decide which one is best. However my latest experience in ACC lets me to believe that with 2080Ti, both headsets will struggle to maintain a steady fps.

The main selling point of the Reverb is the number of pixels, but i doubt that a 2080Ti will be fast enough to run any simrace game with 20 cars on track at 90fps.

The main selling point of the Index is it's refresh rate, but again i don't think that a single 2080Ti will be able to deliver a 120fps frame rate in most of the current race sims with other cars on track.

Now we are starting to see all these reviews of the new gen HMD, but most of them are for normal games, not the high-end demanding racesims we use.

With my current hardware spec, i don't see any useful upgrade that would really boost my performance even more.

So my conclusion is: VR performance in race sims sucks, even if you have a high-end PC.

How's your performance in VR and what's your VR benchmark score?

 

[HoiHman]

Yes but you are forgetting one very important aspect.

You can run Pcars 2.0 with high settings in practice mode during the day with only one car on track.

But try running it with the same settings at night in Long Beach with 32 car grid. Your frame rate will be cut in half or even more.

Since these testers are not sim racers, i highly suspect they did not do full races with an entire grid, but just ran some laps to see how the game looks with probably only one car on track.

That's the problem with VR performance in race games.

 

[RCHeliguy]

Yes but you are forgetting one very important aspect.

You can run Pcars 2.0 with high settings in practice mode during the day with only one car on track.

But try running it with the same settings at night in Long Beach with 32 car grid. Your frame rate will be cut in half or even more.

Since these testers are not sim racers, i highly suspect they did not do full races with an entire grid, but just ran some laps to see how the game looks with probably only one car on track.

That's the problem with VR performance in race games.

Given his status as someone who admittedly got nausea in VR racing sims, that is entirely possible. I guess we will have to find out first hand.

Is the limitation you saw CPU bound or GPU bound?

 

[RCHeliguy]

I find it very unfortunate that at least out of the gate a 2080 is beaten in VR by a 1080Ti.
This comparison was made last November. Hopefully NVidia has improved their drivers since then.
On paper the 2080 SHOULD be faster.

I'm very curious to see what NVidia does next. They will either double down on RTX, or maybe they could create something that is simply a scaled up 1080Ti with lots more cores, faster memory, and higher clock speeds.

https://babeltechreviews.com/the-rtx-2080-vs-the-gtx-1080-ti-in-vr/3/

 

[anton_Chez]

Worst GPU release in my memory of being on PC. Will gladly keep my 2080, oh wait sorry my 1080Ti from 3 years ago until Nvidia actually release something worthwhile for not a million dollars.

 

[RCHeliguy]

I basically agree with you and my 1080Ti is running very well. However if they started to actually utilize the additional pipelines in the RTX cards and saw meaningful improvements that I could actually take advantage of without being CPU bottlenecked, I'd break down and get one. Right now it makes little sense.

 

[Mascot]

The 'Ti' versions have always been much more powerful than 'non-Ti' cards though, haven't they, even from the previous gen?
Might be wrong, but that was my impression.

 

[RCHeliguy]

The issue with the RTX cards is that they have dramatically more power but a chunk of it is in unutilized pipelines. I suspect this will be unlocked eventually. The question is when?

Also on the synthetic tests they already show a large advantage, which means the drivers for them are not very well optimized yet.

There could also be other bottlenecks.

My Z390 motherboard has PCI 3 x 16, x8, x4 slots. My 1080Ti is currently in the x 8 slot which is fine for that card. The 2080Ti is the first GPU that just barely passes the bandwidth of an x8 slot by 3-4%.

So I don't think PCI bandwidth is an issue for the GPU. The only way PCI bandwidth can be pushed is by also using M.2 NVME cards which I haven't bothered with at this point. My MB has M.2 slots, one SATA and one for PCI. I'm not using either yet.

I'm currently running 4 x DDR4 8Gb sticks at 3200MHz. The motherboard is writing to the memory across pairs so data is simultaneously written across two pairs of memory sticks. I saw an increase in performance and drop in CPU load when I changed the memory clock from 2166 to 3200MHz. So the CPU was obviously busied up a bit waiting for memory. I was able to increase SS by about 0.2 in each game after bumping memory.

Currently I haven't OC'd my CPU. The i9-9600K will normally run two CPU cores at 5.0GHz and the rest at 4.86GHz, but I think that is only if there is not a load on the rest of the CPU.

Since I bumped my memory speed up, I've gotten very close to CPU/GPU parity in the sims I have. By that I mean that I can get to a point where I have CPU cores approaching 100% along with the GPU. Before I bumped the memory speed, I could max at least a single core of the CPU before maxing out the graphics card.

The main take away I have from this is that it currently takes a heck of a lot of CPU to feed a GPU all the way when you are using a sim in VR.

While I'm not saying everyone needs 16 cores for things to run well, having "extra" cores does allow load to be spread out such that the software driving transducers, motion systems, and reading from numerous USB devices can happen on other cores outside the cores used for the main game program and rendering. What I'm seeing are 8 cores that seem to be well utilized and another 8 that are seeing minor use. 2 cores are typically working harder than the rest. It does look like you could take the work being done by the 8 cores seeing minor use and fit that work into the less used 4 of the initial 8. Based on what I've seen in my load graphs, I think having at least 8 cores to run sims in is a good idea.

I do think an emphasis on multi-threading is becoming more important over time.

 

[Mr Latte]

While we wait for more on the performance abilities/benefits/demands of the upcoming units...

Just sharing a nice channel to keep an eye on below...

This vid will appeal to Rift owners

 

[sfdmalex]

Not sure why you are having such a hard time. My experience is with Assetto Corsa (vanilla).

I have a 2080Ti and an 8700@5.0 and I have plenty of performance headroom running 4xAA at 1.8SS at 90FPS solid. **4xAA came in after the shader patch optimizations have been all ticked.

I think 2.5 is overkill for SS as I can barely seen any difference after 1.8.......diminishing returns kick in way before 2.5

Other than static reflections I'm pretty sure the rest of the settings are all maxed out including post processing.

 

[HoiHman]

Obviously 2.5 is overkill but, since i can't maintain a solid 90fps with my 2080Ti, i might as well run constantly with ASW and turn up the SS as high as possible.

I'm still totally in the dark what to do with my I7 7700K @4700. Is it the cause of my low performance.

Does it need upgrading? How does the performance of my current CPU compare to a 9900K in VR race games only? Is the performance of the I7 7700 per clock tick so much lower?

What's the gain i can expect if i do decide to upgrade?

 

[sfdmalex]

Obviously 2.5 is overkill but, since i can't maintain a solid 90fps with my 2080Ti, i might as well run constantly with ASW and turn up the SS as high as possible.

I'm still totally in the dark what to do with my I7 7700K @4700. Is it the cause of my low performance.

Does it need upgrading? How does the performance of my current CPU compare to a 9900K in VR race games only? Is the performance of the I7 7700 per clock tick so much lower?

What's the gain i can expect if i do decide to upgrade?

Man even with my 1080ti I could do 90FPS in AC.

I'll double check my settings for you, but I'm pretty sure other than static shadows everything is maxed out.

As far as I recall my GPU is well in the ~90% and CPU not close to maxing out.

For your Ti what kind of settings are you running? One advice that I can give is to manually set your fans to 100% once temps reach 40C or something. I find that the automatic fan settings never went to 100% for me and the card would just throttle once it hit 63C~

Oh and kill hyper threading, it sure helps in benchmarks, but in Assetto Corsa at least game performance benefits from no HT.

 

[m_mirk]

Obviously 2.5 is overkill but, since i can't maintain a solid 90fps with my 2080Ti, i might as well run constantly with ASW and turn up the SS as high as possible.

I'm still totally in the dark what to do with my I7 7700K @4700. Is it the cause of my low performance.

Does it need upgrading? How does the performance of my current CPU compare to a 9900K in VR race games only? Is the performance of the I7 7700 per clock tick so much lower?

What's the gain i can expect if i do decide to upgrade?

You can check this page with cinebench-r20 single-thread performance (not sim racing but maybe related). With i7 7700K @ 4,8 Ghz at 100% performace, I get 102% for 9900 K @ Stock. So I think its not worth an upgrade. Maybe Ryzen 3000, but I dont think so.

https://www.computerbase.de/2019-03...agramm-cinebench-r20-single-thread-ergebnisse

 

[HoiHman]

Thanks @m_mirk

That was the info i was looking for 5% max performance is not enough to upgrade right now. Will certainly wait for the Rysen 3000.

@sfdmalex can you do 90 fps in AC with a full grid and at what SS are you running?

I'm running AC at 2.5SS and max AA to get a decent picture quality in VR. I have always found it dificult to get a steady 90 fps with a full grid. So that why i have turned the settings up and rely on ASW.

Are there any specific setting that really hurt performance without really adding to the picture quality in VR?

I will start from scratch hopefully by the end of this week if the HP Reverb arrives

 

[sfdmalex]

[QUOTE="HoiHman] can you do 90 fps in AC with a full grid and at what SS are you running?

I'm running AC at 2.5SS and max AA to get a decent picture quality in VR. I have always found it dificult to get a steady 90 fps with a full grid. So that why i have turned the settings up and rely on ASW.

Are there any specific setting that really hurt performance without really adding to the picture quality in VR?
[/QUOTE]

I'm running 1.8SS and to be honest I never run full grids as I don't play AI so the most I had was 15 or so in multiplayer, but I sure ran some benches and did some tests in the past but then again that would be with no more than 15 cars.

But seeing that my GPU is maxed and CPU at around 20-30% in solo mode, I guess big grids is all about CPU at that point.

 

[Deleted member 197115]

AI or MP should be the same, rendering thread spends time calculating draw calls for opponent cars.

 

[HoiHman]

So in general for all sims

Enable or Disable Hypertrading for VR performance?

 

[Insert Coin]

I have tested it in AC: performance is worse when HT is disabled (Core i7).

 

[HoiHman]

Thanks @Insert Coin

 

[Apelsin Elise]

So in general for all sims
Enable or Disable Hypertrading for VR performance?

Depends on a sim. In my experience ACC, PC2, R3E do not care one way or the other. On my system (8700k, 1080ti, Odyssey) AC definitely needs HT disabled in VR. I see big improvements in fps with HT disabled (maybe at the expense of grid size). Solid 90 fps (SS 2.0) with all settings on max except reflections.

Based on what others have said on this topic, it looks like HT on/off benefit is highly dependent on a particular PC build. Disabling HT brings frame rate improvements for some, with just an opposite effect for the others. Looks like you'll need to try it for yourself to see the effect on your particular setup.

IMHO, Custom Shaders Patch works like magic in improving VR frame rates in AC.

 

[RCHeliguy]

I'm sure there is some complexity as you mentioned.

For example, how fast is your memory and how many apps do you have running?

GPU and VR system
Transducer system
Motion system
How many separate USB controls do you have.

I found my control inputs and the reaction time of my NLRv3 improved dramatically with my i9-9900K. However when I look at the core utilization, it looks like 8 cores are being well utilized, but have more room and 8 cores are running pretty low single digits. Whether distributing all that out helps or not I couldn't tell you.