P3D5 HF2 issues with VR

[truthknown]

First, let me say thank you for continuing to release hotfixes. Each one offers more improvement. Still a ways to go but getting better each time.

With VR I have a few issues I'm experiencing:

1. When transitioning from night to day, dynamic lighting goes off in general. However, in VR it is remaining on the right eye only. So the left eye has no dynamic lights on and the right eye has the dynamic lights on. It is causing a nauseating effect. Easily seen if showing double mode on the monitor while in VR. If I load the sim straight into daytime scenes, there is no issue.

2. Launching VR in Single Pass mode causes a large VRAM spike. It drives the VRAM up significantly for only a second, then falls back down to normal. Depending on your settings, this spike can cause the VRAM to hit the max budget and crash the sim. In Stereo mode, launching VR does not cause such a dramatic spike in VRAM, the VRAM curve is much more smooth when graphed out. Because of this issue, Single Pass is still unusable for me in VR without significantly reducing settings that actually cause meaningful VRAM decreases.

3. Enabling VR causes an increase in VRAM, but disabling it does not reclaim all of the VRAM it used. In my experience, about 0.4GB of VRAM is lost each time. As its not possible to do everything in VR, naturally one will enable and disable VR numerous times while in P3D. Losing 0.4GB of VRAM for each time you disable VR, you quickly run out of VRAM budget and cause a CTD.

4. When in VR, sometime my budget remains at 10.1GB (1080TI FTW3 11GB card). Sometimes it falls to between 8.7-9.3GB. Its random, I cannot seem to figure out what may be causing the reduction in budget. I know Windows Mixed Reality Portal and Steam VR in combination uses about 2GB of VRAM but since there are most cases where my P3D budget remains at 10.1 while in VR, that doesn't seem to be the issue as those things actually use the shared GPU memory as appropriate.

I would love some assistance with these issues and happy to provide more data and screenshots if necessary. I hope these can be tracked down and fixed, or if there are current solutions, please do let me know.

Thanks,
Chris

[whitav8]

@OP,
You need to tell us which versions of Windows 10 (hopefully 2004), which Nvidia driver (maybe 451.48), which SteamVR you are using (1.12.5 or so) since VRAM is such a hot topic with DX12.

[truthknown]

Using the latest versions of all.

Windows 10 (2004)
Nvidia 451.48
Steam VR 1.12.5
Odyssey+ HMD

i7 8700K delidded OC 5.0Ghz
32GB RAM
1080Ti FTW3 11GB

[MPO910]

I see all points the OP mentioned exactly on my rig too. I am using hp reverb with wmr steam vr. Especially the dyn light trasition rom night to day is awful. Butvthe op did describe the issue very well. I thought it was addon airport dyn light as with eddf or lxgb. I never tested it on stock airports ( have they dyn lights)?

But also rhe vram spike st dingle pass. Even on my titan rtx it is very heavy.

I9 10900k
Rtx titan 24gb
Latest Windows 2004 build, latest nvidia drivers. Latest beta and non beta steam vr are having the same issues with dynlight. So it is not related to them.

Marcus

[jellef]

I have been struggling with VR and P3D v5 ever since it came out. Lot's of investigations learned me a couple of things:

1. No changes in v5 have solved the problem so far, including HF2.

2. No external causes are the 'real' source of the crashes, including Nvidea drivers, hardware or software add-ons.

3. The conditions that lead to a crash are all due to system load, in particular the amount of VRAM in use and the percentage of the 'budget' used. Configuring more complex airplanes or scenery will bring this limit closer, even for the most powerful systems.

4. Most errors are the notorious DXGI_ERROR_DEVICE_HUNG. Mostly this leads to an unrecoverable crash, although HF2 sometimes let's P3D terminate on a more decent matter, but never returns to the situation before entering the VR mode.

5. The amount of VRAM nears the limit, but hardly ever seems to reach 100%. The driver never appears to solve the problems by reallocating some of the dedicated GPU memory to the Shared GPU memory pool, although the latter reserves 50% of the physical system memory.

My conclusion is that the GPU memory management is very poor. There are some indications that this is a de facto DX12 issue, but to me it seems the implementation of the driver requires a solid upgrade. The fact that dedicated VRAM does not return to the amount claimed before VR mode entry after this mode was shut down indicates a serious memory leak. The advise to restart P3D (as I've seen on this forum) after every change in scenery or airplane configuration is no the solution. This is going back to the Windows common practice of CTRL-ALT-DEL in recent days.

Also in my case I use Windows Mixed Reality and Steam VR (using the HP Reverb VR headset) and I know this also contributes to the use of VRAM, but this configuration is supported for V5. The advice to use the Singe Pass mode also for me leads to almost instantaneous crashes. The spike in VRAM usage this generates drives me mad. Whether I start WMR/Steam VR before entering the P3D VR mode or letting P3d to trigger this does not make any difference. Setting the HMDDisplayMode to stereo helps to prevent an imminent crash (mostly), but results in bad VR performance (much higher GPU and CPU frame times, much stutters).

The inability to reclaim VRAM after ending the VR mode usually results in HUNG errors the next time VR mode is enabled, because less VRAM is available. The 1st attempt success, the next one with unchanged settings fails! The loss of about 0.4 GB each time the mode is switched, also happens on my system.

My RTX 2080 Super has 8GB VRAM and should be able to handle this with ease. With a video card with 6GB or less the system becomes unusable in most cases, unless you are willing to run with only default scenery and planes.

The DXGI_ERROR_DEVICE_HUNG error in my case mainly occurs in running VR mode, but this is only because I have a pretty powerful system. If I screw up the graphics settings to the max in non-VR mode, I get the same crash. In the mean time I have seen multiple reports of a newer version of Windows 10: 2004 (I run on 1909), so I see no reason at this moment to update. I' am not convinced this results in a better behaviour. Windows in itself doesn't seem the main reason for these crashes, although the use of DX12 maybe.

I am counting on LM to develop a better graphics driver that improves VRAM management and has a better error recovery. This is urgently required!

Cheers,

Jelle

[Beau Hollis]

Thanks for the continued reports. We are able to reproduce the GPU memory leak with SteamVR and are looking into it. The VR emulator interface seems to be leak free so that should help narrow things down.

[Beau Hollis]

I wanted to reply to all 4 issues posted:
1) We'll look into this.
2) This is a known issue which we hope to resolve but it's tricky because it happens as a side-effect of a feature that's working as designed. It relates to how the engine caches resources internally. Render target caching and reuse for post processes saves VRAM and avoid stutters, but in this case the non-vr view resources remain in the cache for a few frames before being released.
3) As mentioned above. we've reproduced this one and will look into it.
4) We don't have any control over the budget provided by the OS, so there's not much we can do here. If you have any VR utility apps, perhaps they are detecting that VR is in use and allocating some resources.
Thanks

[jellef]

That sounds promising. If the single pass mode behaves the same as the stereo mode in terms of switching behaviour (not cause the spike) this will be a big step forward. And if the memory leak is solved the stability of the graphics system will improve a lot.

Thanks for the update, I look forward for an update.

Jelle

[vtracy]

When I read the reports and comments in this forum, I am somewhat astounded. LM, you must have tested the sim with HF2; ergo there must be suggestions - or min/max settings suggestions - that you can give us: OS version, RAM, and VRAM.
For me, it is not really helpful to read about crashes occurring with a 8 GB card as I have a 11GB card(GTX1080ti) and the same crashes happen when I fly.
BTW: In my case, I did a clean install of v4 HF2, EU Germany South and North, and EU England. I started flying with the default Mooney Bravo Retro, both from a North German airfield (Luneburg) and from Manston, Southern England. In both cases, everything went well. I then installed A2A Spitfire Mk IIb and tried to change to it after starting the sim with pre-loaded Mooney and Manston: CTD. I then started the sim; then, using the startup window, I changed from Mooney to Spit and started the sim. The Spit showed up, at the Ramp in Manston. I then made a mistake in starting the Spit, it nosedived. I hit „Resume“, the Mooney appeared on Manston Ramp. I changed to Spit: CTD.
My system: 7700K 4.3 GHz, 16GB RAM, nVidia GTX1080ti(11GB VRAM) all SSDs for OS and sim, Win10 PRO 2004, latest nVidia driver.
As you can see, it is a fairly powerful system that I put together in the expectation that later P3D version would require more powerful hardware. Looking at the latest sim behaviour, I wonder what more is expected in system resources.
Volker

[Britjet1]

Same issues here.
Oculus VR totally broken with hotfix 2. Stutters. GPU memory errors, screen blanking. CTD.
This is with 2080Ti.
Fed up.

[JorgenSA]

@vtracy,

I do not use VR myself, but reading your post it seems that your issues began when you installed the A2A Spitfire. Up until then, everything seemed to be fine.

Have you talked to A2A about complete, absolute v. 5 compatibility of the Spitfire?

Jorgen

[truthknown]

Thanks Beau,

I hope you guys are able to track down and solve these problems. Could we expect a quick patch if some of these are fixed? Particularly #1 and #3? I am happy to test out any potential solutions. I look forward to the upcoming fixes.

Thanks,
Chris

[nathantw]

What I don't like is that cars blink at night. It's just a big strobe effect of lights at night. FWIW, I found that version 4.5 with HF3 has the same problem.

[truthknown]

Hey Beau,

Any progress on these? Would love to know if you guys were able to find solutions!

Thanks,
Chris

[Beau Hollis]

I have an update on a couple of these:
1) We are fairly confident this will be fixed in the next release. I've made a ticket and it will be worked soon. It's a core engine issue so it's not something we could include an a VRHotFix add-on.
3) As of now it appears to be an issue with OpenVR. The vr compositor is holding a reference to the resources we submit. We're going to try a few things to mitigate.