VR testing and findings
-
@Lippy said in VR testing and findings:
Finished VR setup
Asus ROG Strix G512LW Laptop
I7 10750H
RTX 2070
RAM 64G
Rift S 80Hz
The only thing I changed “set g_bVRNoPresent 1” turned off VR display
Set Antialiasing to 8 everything else default settings
The results 30 to 40fps on the ground 73to 80fps in the air single player
No jitter no black lines at the edges of the frames the only problem I have now
is temp runs at 83C ordered a new fanIf you are going to post FPS figures or do a comparison you really should be using the Benchmark TE, otherwise there is no way to correlate it to anything/anyone else… That TE is just sitting on the Runway looking around in the rain, record your FPS there.
-
@Seifer said in VR testing and findings:
Parallel Render
set g_bVRParallelRenderThread 1 // 0 disables, 1 enables parallel renderOne of the nastier bugs we had during VR development was understanding jittering/juddering (when you rotate your head and the movement is not smooth). This happens because in BMS we prepare the scene in one thread and render in another, like most modern games. The problem is that VR is a bit peculiar, so this creates a race between the thread that prepares the scene and the one that renders it.
To fix this issue, we serialized both threads for VR. The drawback is that we lost a lot of performance because of this. We were able to fix this issue, but for some headsets (or for virtual desktop) the fix doesn’t seem to work for reasons we still don’t understand.
So, if you have jitter/judder, try reverting to set g_bVRParallelRenderThread 0, which will make movement a lot smoother.I found this video that sums up the issue quite well in my opinion :
-
Tried to find a subject on this but couldn’t really… Has anyone been able to get OpenComposite (the SteamVR replacement) to work with BMS? I ask because trying to see if I can get OpenXR toolkit to run in BMS, which has lots of performance-enhancing settings.
-
@spotdott you won’t get opencomposite or openxr to work. You can get vrperfkit + FSR to work, however. I didn’t find much performance benefit with it.
-
@rubbra hello, where did you put the .dll and .yml files? I tried the bin folders for steamVR and BMS with no results.
-
@spotdott I had similar lack of success then used an automated installer/patcher ; annoyingly I’ve wiped and rebuilt my pc since then and I can’t remember what it’s called. Will do some googling and get back to you.
(Edit - I think it was this - https://github.com/tappi287/openvr_fsr_app )
-
@rubbra I don’t think FFR works with BMS but the upscaling might be working. I’m able to get about 25 ms timing overall in the benchmark TE (vs 35 ms) with 1.4x resolution , scaled by 83% with a sharpness of 89%. Picture quality wise it looks pretty decent (1.3x resolution sharpness for lower performance cost)
-
@spotdott no idea on ffr - my gpu doesn’t support it, but FSR does for certain.
-
@rubbra I have an RTX 3060 ti, using NIS mode able to achieve about 25ms frametime (40 fps) with shadows on in the benchmark TE. Set upscaling to 80 percent on VR resolution of 1.4, thread optimization enabled. very playable in campaign I’m very satisfied. Thank you very much
-
@spotdott no worries at all. I’m not using it at all, 1.8x supersampling on my rift s, shadows on, BMS res at 1.0, etc etc and rocking a solid 80fps nearly all of the time. It’s really only smoke that kills my performance now. (1080ti)
-
Hey guys, I’ve acquired an i5 12600K for testing and wanted to see how it would compare with the i3 12100.
GPU is an RTX 3060ti 8 GB, OC a little bit (+150 MHz core, +900 MHz memory)
Clock speeds are all stock (no Overclocking).
For the i3, the four cores will run at 4.1 GHz; for the i5, the six cores will run at 4.6 GHz
DDR 4 XMP profile 3200 MHz
on a B660 Motherboard
Windows 11 (which I’m not in love with)Quick comparison conditions:
Scenario: Benchmark TE
No shaders activated (except far rain)
half smoke densitySteamVR resolution set to 100% (3156x3092 - Reverb G2)
Shadows ON
Thread Optimization ON
VR Resolution set to 1.4
Object Detail set to 5
OpenVR FSR App set upscaling to 80%Quick comparison findings (all in frame timing (ms), not FPS)
i3: 12100: FPS 26ms, SIM 25ms , REN 13ms
i5 12600k: FPS 26ms, SIM 25ms, REN 11msenabling/disabling E-cores didn’t make a difference
enabling/disabling Hyperthreading made a bit of a difference for the i5 (25ms/24ms/10ms)Of course, I was also curious what if upscaling is turned off; I don’t want the game to be GPU-bound so reduced the resolution to 1.2x:
for the i5 (FPS/SIM/REN),
with shadow off and thread optimization’s on: 27ms/26ms/8.6ms
with shadow on and thread optimization’s on: 29ms/28ms/11ms
with shadows off and thread optimization’s off: 26ms/6.7ms/32ms
with shadows on and thread optimization’s off: 29ms/10ms/29msHave to note that with both CPUs, in the stock Korea campaign when in the air away from the air base, the frame rates are much better.
In conclusion, it would seem that BMS in VR is not taking complete advantage of the 12600K’s extra cores, its faster clock speed, nor larger cache.
-
@spotdott Great info.
Just a comment. I’m guessing you are using SteamVR to upscale to 3156x3092, (cannot for the life of me understand why SVR cannot recognise the reverb as square panels! ) and this represent nearly 1.5 upscaling. By setting g_fVRResolution in the game to 1.4 this is further upscales the render resolution to almost x2 the native panel resolution. You might want to do one or the other and compare this with both on as you might find the difference in quality does not justify the fps loss.
Just a thought
-
@Fish44 a few months ago I originally decided to set the steam VR resolution to close to the reverb g2’s native resolution but a member of this board told me that steam needed to increase super resolution in order to compensate for the loss in image quality due to lens distortion. There’s also a clear difference between setting vr resolution between 1 and 1.4. however after using a third party app to downsize the resolution and then upscale, I was able to achieve very good frame rates. The point I was making was, when it comes to 12 gen CPUs I was bottlenecked by my graphics card more than the amount of CPU cores and frequency. I really hope the developers fix thread optimization so that it doesn’t just flip fps reported from Ren to Sim. In the mean time, I can just stick with the i3 and put the money in getting a more powerful graphics card.
-
@spotdott said in VR testing and findings:
@Fish44 a few months ago I originally decided to set the steam VR resolution to close to the reverb g2’s native resolution but a member of this board told me that steam needed to increase super resolution in order to compensate for the loss in image quality due to lens distortion. There’s also a clear difference between setting vr resolution between 1 and 1.4. however after using a third party app to downsize the resolution and then upscale, I was able to achieve very good frame rates. The point I was making was, when it comes to 12 gen CPUs I was bottlenecked by my graphics card more than the amount of CPU cores and frequency. I really hope the developers fix thread optimization so that it doesn’t just flip fps reported from Ren to Sim. In the mean time, I can just stick with the i3 and put the money in getting a more powerful graphics card.
Me, i’m hoping for OpenXR compatibility…
-
Here’s another update to the quest for more FPS in the reverb G2: I upgraded to a RTX 3080 from the 3060 ti. Will give an update when I try the 3080 with both the i3 12100f and the i512600k.
-
@spotdott said in VR testing and findings:
Here’s another update to the quest for more FPS in the reverb G2: I upgraded to a RTX 3080 from the 3060 ti. Will give an update when I try the 3080 with both the i3 12100f and the i512600k.
Okay so after some testing here is my findings after upgrading to a MSI Ventus 3080 OC 10 GB:
i3 12100f + 3080: FPS 22ms, SIM 21ms, REN 16ms
i5 12600K + 3080: FPS 19ms, SIM 19ms, REN 14msFor comparison:
i3 12100 + 3060 ti: FPS 26ms, SIM 25ms, REN 13ms
i5 12600K + 3060 ti: FPS 26ms, SIM 25ms, REN 11msfor the i5, I had to disable the E-cores and multithreading to get the best results, in that case it’s just essentially an overpriced 12400.
In conclusion, super high resolution demands graphics cards that can do well in 4k or even 8k gaming, and less dependent on the core count of the CPU. I’m going to return the 12600K and continue to save up for a 3090, 4080 or 4090
-
@spotdott yes, I think this is overall a good plan. Part of the reason I went for “only” an i5-13600KF with all the way to 4090 for VR, which was the first time I have gone without an i7 in several upgrades. Also with VR resolutions, AA settings over even 2x really requires more than 10GB VRAM unfortunately, so that is a major limitation with Nvidia right now.
-
Just an F.Y.I. to some folks out there.
I recently obtained a G2 Reverb and was disappointed by the poor frame rates I was getting, (20-24 FPS in MP campaign, 12 FPS in benchmark TE). After trouble-shooting nearly everything, I finally settled on something wrong with the video card, (eVGA 3080 Ti).
Upon using Precision X1 to investigate its settings, it turns out it was set to throttle power at 20%. I know I never set it that way, but it turns out that others in the internets have had similar issues when updating Nvidia drivers with Precision X1 running in the background.
I’ve had the 3080Ti for over a year, and I think it’s been set this way since Day 1. All of my games are now running about 3x more FPS since I set the target to 100% power use so it doesn’t throttle the card. This is what the default target is supposed to be.
I’m now up to ~70 FPS in campaign and ~35 FPS in the benchmark TE with the Reverb G2, with nothing overclocked, (not my 8700k, nor memory, nor GPU) with shadows and environment mapping enabled. Perfectly playable.
Obviously not everyone has an eVGA and uses Precision X1, but it may be worth investigating whether your graphics card is being throttled due to some (not by your wishes) setting.
-
@Scoob_SBM said in VR testing and findings:
I’m now up to ~70 FPS in campaign and ~35 FPS in the benchmark TE with the Reverb G2, with nothing overclocked, (not my 8700k, nor memory, nor GPU) with shadows and environment mapping enabled. Perfectly playable.
Your 70 FPS in campaign is measured in the air?
-
@spotdott Yeah, campaign measured in the air. Benchmark TE measured on the ground.
-