Anyone have a stutter-free multi-channel setup?

[MistyBlue]

Please forgive the up front venting. I feel like I'm forever chasing the pot of gold at the end of the rainbow trying to get a completely stutter-free multi-PC setup in my home cockpit and I'm just tired of years of messing with it, upgrade after upgrade, hoping THIS time will be better only to be disappointed again and again.

I really don't understand why at this point in time it isn't a breeze to get today's high-end PC's with high-end graphics cards that can run at high frame rates across high speed local networks to run perfectly smooth and in sync. Maybe someone can explain because it eludes me. If it's still a limitation of the original code, I sure hope we can finally get over that by the next major version.

/over

I am perfectly able to get all my individual P3D v5.3 PC's running practically as smooth as glass between G-Sync and variable refresh rate in P3D (which interestingly for me has always been the smoothest at 59 FPS with VRR, VSync, and Triple Buffers on). But combine 2 or more PCs with Multichannel and no matter what it's a stutter fest. Host...perfectly smooth. Clients...not smooth.

For reference, I'm running the host on a 10700KF, and the clients on a 12700K and 11900K. Clients have RTX 3090's. M.2 980 Pro's for hard drives. G-Sync high refresh monitors. They're not slouch PCs.

My litmus test for smoothness is to turn down all P3D settings to minimum, then take an aircraft, put it on the apron at an airport, turn on slew, and set the aircraft spinning about the Z axis continuously. I then watch the airport and scenery closely as it moves by. If I see any ghosting, flickering, tearing, blurriness of scenery, inconsistent frames, etc, then I consider it not acceptable. Especially when using a collimator like I am where every inconsistency is magnified in your face and immediately noticeable.

I'm aware I can lock to 30 FPS via several methods but I have NEVER EVER seen that look any better than a cartoon flip book where the frame flicker is visibly perceptible. Sure, the flicker is consistent so some may say it's "smooth" but the strobe effect is not ideal.

So I'd like to know, has anyone gotten a perfectly smooth multichannel setup (per my litmus test above) without using frame-locked cards? If so, what was your configuration, both in P3D and NVidia CP, along with your display refresh settings?

[MistyBlue]

I guess no one has a stutter-free experience with multi-channel then? Wow.

Well, some things I've found in my setup so far that seem to have helped:

First, I've upgraded my host PC from a 9900K to a 11900K, and clients to 13900K's overclocked. I'm still running a single 3090 in each client PC. I'm sure if I upgraded to 4090's that would help but not even going to consider that route right now. I may further upgrade the host to a 13900KF.

Second, running P3D in the smallest possible window at the lowest possible settings on the host PC improves stuttering quite a bit. My assumption is that P3D is then able to run at the highest frame rate possible. Seems, though, that to work with ActiveSky and AI traffic, weather and traffic settings need to be turned up on the host so the clients inherit so that kills frame rates a bit. I may upgrade the host PC again to a 13900KF which I have lying around.

Third, running all PCs with VSync off, target frame rate unlocked with Variable Refresh Rate on seems to provide the smoothest frames. I'm still experimenting with that as I think I can still improve it somehow.

Still experimenting, but hope that helps someone.

[SorensenDK2]

Hi there.

I feel your pain here! I to feel like the option of having a stutterfree multi Pc setup for P3D is an impossible mission.

Does all your screens run with same refresh rate when using VSYNC?

Have you tried messing around with Wideview? I think this makes a stutterfree multi view possible, but then it seems to be difficult to have problemfree weather and traffic synchronization.

I have until now given up on achieving multiview with P3D, and hope for a solution with MSFS.

Best Regards
Bjarne Sorensen

[MistyBlue]

Hi Bjarne, glad to hear I'm not the only one.

All 6 displays are G-Sync high refresh monitors so I don't use VSYNC. I use the variable refresh setting in P3D instead, though I'm not sure it makes much of a difference when you have 6 displays all running at slightly different refresh rates. Anyway, I found VSYNC introduces more stutters, likely because the lower frame rates interferes with a consistent sync between host and clients.

I have used WideView for a long time and I have nothing really bad to say about it. It's a bit smoother, but as you said, there are some drawbacks that I think multichannel alleviates for me...like perfect weather sync, and AI traffic sync (except for missing AI lights on some displays, another topic I posted about), and just simpler to run than having to have several apps to do the same thing.

For now, with the host and 2 clients having been upgraded to 13900K processors I feel things are running a little better. Still not perfect, but more tolerable.

In the meantime I am still playing with MSFS with WideView. It's far from perfect right now. I just wish we could spend way more time flying and much less time, money and effort worrying about this stuff.

[SorensenDK2]

Hi Mistyblue..

I was wondering if you have had any luck experimenting with P3D multidisplay? I really want to change my setup to a multichannel, but are still struggling to get it right...

Best Regards
Bjarne

[MistyBlue]

Hi Bjarne,

I'm afraid I haven't gotten things to run any better since I last posted. I have been building the sim more than flying so I haven't done much more experimenting. But last time I fired up the sim to show to someone it was a stutter-fest again. Since I didn't have time to spend hours figuring it out I just walked away. When I have time and patience, I'll try again and let you know if I discover anything useful.

[SorensenDK2]

Hi Tony...

Thx for reply. It is weird that it is such a hazzle. I think I will also give it a try again, and of course also give a hint if I should find the wise answer ;-)

/Bjarne

[MistyBlue]

FWIW, here's an update. So far I still have not been able to get rid of the multi-channel stutters, even at very low visual settings across all PCs.

----
Here are my current specs,which are all top-end as of this writing:

Host PC:
13900K, 2x16GB DDR4 3800 RAM (if I recall), RTX 3070Ti, Samsung M.2 980 Pro
240Hz display dongle (no physical display)
P3D running over 600-700 FPS on mostly minimal settings and 1024x768 resolution.

Visual PC1
13900K, 2x16GB DDR5 6000 RAM, RTX 4090, Samsung M.2 980 Pro
3x 27" Dell G-Sync monitors running at 2560x1440 each
P3D running over 200 FPS on mostly minimal settings

Visual PC2
13900K, 2x16GB DDR5 5600 RAM, RTX 3090 Kingpin (soon to be a 4090), Samsung M.2 980 Pro
3x 27" Dell G-Sync monitors running at 2560x1440 each
P3D running over 150-160 FPS on mostly minimal settings

Network between host and visual PCs is 10 Gigabit fiber.

View group configured on the host that includes only the Visual PC 1 and 2 displays.

FlyElise warping software. Performance doesn't appear to change with it on or off.

----

I've reduced all visual settings to just about the lowest settings across the board to ensure highest frame rates and rule out any heavy graphics load causing stutters.

Adjusting frame sync in the host network.cfg doesn't seem to have any appreciable change unless you get below about 15 and over 60ish and the stutters get much worse.

Changing the P3D window size on the host PC still has the biggest effect on stutters. Larger window sizes tend to cause a lot more stutters on the clients. Smaller windows seem to improve smoothness some, though the overall frame rate doesn't seem to change much, if at all. I'm still experimenting with sizes that work best in my case.

I know Genlock is the supposed solution, but forgive me if I'm not able or willing to invest $6-8K/Quadro GPU plus another $500 per card for the Genlock add-in card when to date I haven't seen an appreciable ROI on the expensive hardware I've already thrown at the sim to date. Someone would have to show me that Genlock truly works to make it even worth considering.

That said, is a software Genlock even possible? Some sort of clock sync that gets passed over the network to keep the frames in sync? Does it really have to be on expensive hardware given the very powerful (and much cheaper by comparison) consumer CPUs and GPUs we now have available?

Sadly, there's no silver bullet as we have all be finding out over 20 years or more. It still seems like the best, and simplest wraparound visual solution is with one almighty visual PC and 2-3 display projectors. Anything more than 1 visual PC without Genlock (and I still haven't confirmed the claims it truly does work) seems like a waste of money and time.

Unfortunately, with 6 displays I'm stuck with a minimum of 2 PCs, so the experimentation continues.

[MistyBlue]

Following up some more...both visual clients are running RTX 4090's now. The stuttering is a little better at the lowest visual settings, but with both visual clients consistently running over 150 FPS and the host easily over 200 FPS it's not the PC's capability that's the problem.

Therefore, based on my experience and running out of ideas, I'm ending up at the same conclusion. Multi-channel in its current form is not viable for multiple PCs using consumer cards, not even with a 13900K and a 4090.

That leaves only Genlock as a possible answer, but I can't confirm. If I win the lotto, I'll let you know.

So, back to my original question. Has anyone with multiple PCs found a way to get a consistently smooth, stutter-free experience with Multi-channel?

[Chris Metel]

I believe Nvidia has documentation for hardware setup and calibration utilities for testing sync using moving lines patterns. There is no Prepar3D specific instructions provided because Prepar3D does not directly interact with sync hardware. The sync hardware allows the displays to present at the same time. With V-sync enabled, Prepar3D's rendering work will be delayed until a display buffer is available which allows Prepar3D's frame timing to be implicitly controlled by the sync hardware.

The host will require a GPU. It displays a view which represents the full view group. The host is not considered part of the view group however so it can be configured with v-sync disabled and should not require sync hardware.

In addition, the settings and content would need to be tuned so that all channels can hold the target framerate to keep the systems visually synchronized. Prepar3D's channel sync operates on the CPU workload. If one channel has a long CPU frame, the others can wait to present. If there is a long GPU frame on one channel, it may present later than the others.

We will investigate the dynamic lighting issue. Does this issue also show up when using a non multi-channel view group such as the " Auto-Fill Desktop" group that is included by default? Do you have some test steps we can utilize to allows us to replicate the issue?

Cheers,
Chris

[MistyBlue]

Thanks Chris. We don't have any other way to get help except these forums so it's nice to see a LM response. Much appreciated!

You answered some of my questions about GenLock, one of which is if there's anything specific P3D does with regard to frame sync. Sounds like the work is all done by the Sync card/NVidia driver. That's good as it seems it's not software dependent and therefore should work with any simulator.

Can the host GPU be just the onboard graphics or a low-end card? I'm running a 3070 Ti with a high refresh monitor dongle rather than a physical display, so I suspect the card is overkill. Seems like the GPU is only there to give Windows a desktop to run P3D. I'm not sure how much it influences frame rates on the host if you're using a virtual display.

So, uncap the frame limit on the host but set a frame limit on the client PCs? And V-Sync should be enabled for only the clients? I've tried that, but without GenLock cards, and I still get stuttering at the lowest graphics settings. So, my conclusion is the stutter must be due to the frames not being in sync across displays. I can't seem to find anything else that's causing it.

Regardless...I just want to make sure I have exhausted every option with my current configuration before I drop $10K for two GenLock graphics cards and the associated sync cards and find out it was a waste of money. So if you can think of anything outside of GenLock that will help I'm all ears.

Regarding the dynamic lighting issue...I have tried WideView instead of multichannel to sync the PCs and configured a local View Group on each client. Dynamic lights work correctly across all displays in that case. It seems only when I use Multichannel does it not work. I don't believe I tried auto-fill so I'll let you know what happens. I have a suspicion this issue may be due to zoom/LOD but can't confirm it.

A test case is pretty straight forward:

CONFIGURATION:

- Configure a host PC, with two visual client PCs with at least 2-3 displays on each client (I have 3 displays for each visual client PC with each display approximating fwd, 45 and 90 degree out-the-window views for left/right seats)
- Enable multichannel, and configure a view group on the host that spans the two client PC displays
- Ensure Dynamic lighting is enabled on the clients (seems it's not necessary on the host)
- Ensure you have AI traffic installed that has dynamic logo lights, landing lights and strobes. I use AIG aircraft. Probably set the traffic to 75-100% to ensure there's enough AI around to test with.
- Set your zoom to 1.0 which is what I'm using (although it doesn't seem to matter the zoom level)

TESTING:
- Switch to night time and move to a location at a busy airport where there are many AI traffic parked or moving with dynamic lighting on.
- In SLEW, start rotating your aircraft around the Z axis and keep an eye on the dynamic AI traffic lights as the object moves across the displays. If things are working correctly, the AI lighting should appear exactly the same from display to display.
- In my case, the dynamic lights on the AI disappear across displays. For me, dynamic lights appear in the fwd view, but then go dark as the AI traffic object passes to the 45 and 90 degree views.

Let me know if you need more clarification. Thanks again!

[Beau Hollis]

We typically advise our solution developer partners to get as much out of one box as they can before relying on multi-channel or multiplayer. You might want to try running all 6 of your views on one box with a couple monitors displaying two undocked views. This would give you a sense of how well your CPU can handle the 6 view case. If that test works better than your multichannel test, you would likely be better off investing in a dual GPU box and driving 3 displays off of each GPU. The other option for a one box solution would be using a hardware splitters to work around the 4 output limit.

While genlock hardware is needed to ensure that things are fully in sync, this is actually true even on a single box running multiple displays. The difference in presentation between genlock and not would be less than 16ms. The stutters you are describing sound like they are much bigger than that and are likely not going to be solved by genlock.

Thanks,

[MistyBlue]

Thanks Beau. I agree if I could somehow keep it to one visual PC that would be the best way to go.

I'll give your suggestion a try. I suspect I won't have any issues with a single box and undocked views with the RTX 4090 but it's worth a shot.

I'm not sure I can jam two RTX 4090's into one PC because they are so massive and power hungry, but I'll take a look. I could certainly do it with a couple of RTX A6000's since they are smaller form factor and much less power hungry, so I may give that a try. I can always return them if it doesn't help.

I thought about splitters, but something like a TripleHead2Go doesn't support a high enough resolution so that's out, unless you know of another solution.

I'm concerned that you're saying there may be bigger problems. What do you think those might be? I've tried to rule out everything I can. High-end CPU's/memory/M.2s/GPUs, fiber network, High refresh G-sync displays, different drivers, etc, etc. Not sure what else I can throw at it or optimize further but I wouldn't be surprised if there's something I'm missing. Sometimes it's just one obscure setting.

In any case, my gut tells me this is a refresh mismatch issue. The horsepower is there, and I notice smoothness varies wildly depending on the frame rate setting and monitor refresh. I just haven't found a display refresh to frame rate setting that works, so I'm grasping at straws now.

As an aside, somewhat related to figuring out the best performance...is there a way to develop a automated utility that goes through each P3D graphics setting to determine the optimal quality and performance settings? Maybe it runs a background test scenario for each setting, looks at the FPS hit (or frame consistency) then determines the best combination to maximize your hardware performance or at least provides a solid starting point. I wouldn't care if it took half a day to run if it ultimately helped me find the best settings for my particular hardware and saved me months of tinkering.

[SorensenDK2]

Hi Tony.

Did you ever find a solution for your multiview setup?

Best Regards
Bjarne