Multithreading Performance is Sensational in VR

[Flextremes]

1 hour ago, SneakyBastd said:

@Voyager many people are complaining that 40x0 series GPUs on Vive Pro 2 is stuttering and laggy, unplayable. Are you not getting any of those problems???

I have VP2 and reluctant to upgrade to 4090 if I'm only going to get problems!

For reference: https://forum.htc.com/topic/14223-nvidia-has-changed-the-render-pipeline-framedrops-on-vive-pro-2-rtx-4090/page/21/

Thanks.

G2/openXR on the 4090 runs extremely smooth for me... Is it possible to use openXR on the VP2? 

Also: check-out 

 

 

Edited March 11, 2023 by Flextremes

[PLUTON]

What should I do after updating the MT a setting or an operation to perform? Or does it work without doing anything?

[Supmua]

13 hours ago, Nahemoth said:

MT version seems to be fine with OpenXR support, but if I try with SteamVR I get a lot of spikes.

Someone else is getting this problem?

 

I only see this with Steam OpenXR + OXRTK (in my case, it was the Turbo mode on OXRTK that was causing this).  No frametime spikes with Steam OpenVR mode.  You can either disable OXRTK or do a setting reset to get rid of the spikes.

EDIT: having freezing issue with OpenVR mode with mt version.  OpenXR is the only API that is currently trouble free for me.

 

Edited March 12, 2023 by Supmua

[Supmua]

1 hour ago, SneakyBastd said:

@Voyager many people are complaining that 40x0 series GPUs on Vive Pro 2 is stuttering and laggy, unplayable. Are you not getting any of those problems???

I have VP2 and reluctant to upgrade to 4090 if I'm only going to get problems!

For reference: https://forum.htc.com/topic/14223-nvidia-has-changed-the-render-pipeline-framedrops-on-vive-pro-2-rtx-4090/page/21/

Thanks.

I no longer have the VP2 (sold mine), but if Varjo can fix the stuttering issue with the rtx 40series so can Vive which is a bigger company.  With the DCS MT version, being CPU bound becomes less of a problem and now it's mostly GPU that will dictate your final fps.

Now that DCS outputs OpenXR natively, it should work with VP2 (or any SteamVR HMD) via Steam.  I'm running OpenXR via Steam right now with the Aero.

Edited March 11, 2023 by Supmua

[Voyager]

3 hours ago, SneakyBastd said:

@Voyager many people are complaining that 40x0 series GPUs on Vive Pro 2 is stuttering and laggy, unplayable. Are you not getting any of those problems???

I have VP2 and reluctant to upgrade to 4090 if I'm only going to get problems!

For reference: https://forum.htc.com/topic/14223-nvidia-has-changed-the-render-pipeline-framedrops-on-vive-pro-2-rtx-4090/page/21/

Thanks.

I only got my 4090 and system set up only a few weeks ago and have not had time to do much in VR beyond configuring DCS. 

I haven't really seen anything like what they're describing though. I have had some issues with Steam setting the render resolution to ridiculous levels and flooding out the vram. 

I'll keep an eye on things and see if it shows up for me. 

[Flextremes]

Compare below the OpenXR CPU and GPU frame times in Single threaded and Multithreaded: F-18 on carrier ramp in Persian Gulf: 

1st: Single Threaded...

Secondly: Multi-Threaded

 

 

TOTAL CPU combined in ST: 41.1ms ...  in MT CPU combined is only 16.2 ms. Especially the 'app CPU' 22.5ms (ST) vs 4.0ms (MT) difference is stark.  

 

Edited March 11, 2023 by Flextremes

[BoneDust]

I'm running an Alienware New Aurora R9 | Windows® 10 Home Premium | 64bit, Intel(R) Core(TM) 9th Gen Intel Core i9-9900K 8C/16T R0| NVIDIA GeForce RTX 4070Ti 12GB | 2 X 1TB SSD | 64GB Dual Channel CORSAIR Vengence at 3200MHz | 850W EPA Gold Liquid Cooled power supply | G2 Rverb VR

I do not see any improvement in VR running MT.  Lost of folks are saying they're getting huge improvements, but for many of us, there's no improvement.  In fact, non-MT seems smoother.  The frame rates are comparable between both modes, drifting between 34-49fps depending on where I am, but I notice some texture tearing in the water from a distance in both MT and non-MT.  I can't list all of my DCS settings, but I have terrain set to low, water high, AS 16X, heat blur low, motion blur off, PD set to 1, clouds set to "standard", shadows low and flat.  I'm sure it could be a setting somewhere, but there's some many variables it is impossible to sort it all out.

Hopefully, the next iteration will see improvements for those of us seeing none.

[KCOOL]

5 minutes ago, BoneDust said:

I'm running an Alienware New Aurora R9 | Windows® 10 Home Premium | 64bit, Intel(R) Core(TM) 9th Gen Intel Core i9-9900K 8C/16T R0| NVIDIA GeForce RTX 4070Ti 12GB | 2 X 1TB SSD | 64GB Dual Channel CORSAIR Vengence at 3200MHz | 850W EPA Gold Liquid Cooled power supply | G2 Rverb VR

I do not see any improvement in VR running MT.  Lost of folks are saying they're getting huge improvements, but for many of us, there's no improvement.  In fact, non-MT seems smoother.  The frame rates are comparable between both modes, drifting between 34-49fps depending on where I am, but I notice some texture tearing in the water from a distance in both MT and non-MT.  I can't list all of my DCS settings, but I have terrain set to low, water high, AS 16X, heat blur low, motion blur off, PD set to 1, clouds set to "standard", shadows low and flat.  I'm sure it could be a setting somewhere, but there's some many variables it is impossible to sort it all out.

Hopefully, the next iteration will see improvements for those of us seeing none.

Many of us are GPU limited, so would prefer DLSS to MT. Hopefully in the next patch next month.

[Flextremes]

46 minutes ago, BoneDust said:

I'm running an Alienware New Aurora R9 | Windows® 10 Home Premium | 64bit, Intel(R) Core(TM) 9th Gen Intel Core i9-9900K 8C/16T R0| NVIDIA GeForce RTX 4070Ti 12GB | 2 X 1TB SSD | 64GB Dual Channel CORSAIR Vengence at 3200MHz | 850W EPA Gold Liquid Cooled power supply | G2 Rverb VR

I do not see any improvement in VR running MT.  Lost of folks are saying they're getting huge improvements, but for many of us, there's no improvement.  In fact, non-MT seems smoother.  The frame rates are comparable between both modes, drifting between 34-49fps depending on where I am, but I notice some texture tearing in the water from a distance in both MT and non-MT.  I can't list all of my DCS settings, but I have terrain set to low, water high, AS 16X, heat blur low, motion blur off, PD set to 1, clouds set to "standard", shadows low and flat.  I'm sure it could be a setting somewhere, but there's some many variables it is impossible to sort it all out.

Hopefully, the next iteration will see improvements for those of us seeing none.

Honestly, that sucks! It annoys me because I do not understand. Did you compare your CPU and GPU frame times instead of general FPS? I just cannot understand why I am experiencing such a dramatic improvement across the board and you do not. Look at my CPU frame times improvements: those are not dependent on GPU.

On a CPU level, you should at least also experience a difference. Is your 4070ti the main bottleneck? I do not believe so. 

With that said.. I think we are all missing something here: why are the results so mixed. I just cannot fathom what we are overlooking. Grrr...

You did start DCS.exe from the Bin-MT folder right? (apologies for such a stupid question, better safe than sorry)

Edited March 11, 2023 by Flextremes

[Hippo]

48 minutes ago, BoneDust said:

I'm running an Alienware New Aurora R9 | Windows® 10 Home Premium | 64bit, Intel(R) Core(TM) 9th Gen Intel Core i9-9900K 8C/16T R0| NVIDIA GeForce RTX 4070Ti 12GB | 2 X 1TB SSD | 64GB Dual Channel CORSAIR Vengence at 3200MHz | 850W EPA Gold Liquid Cooled power supply | G2 Rverb VR

I do not see any improvement in VR running MT.  Lost of folks are saying they're getting huge improvements, but for many of us, there's no improvement.  In fact, non-MT seems smoother.  The frame rates are comparable between both modes, drifting between 34-49fps depending on where I am, but I notice some texture tearing in the water from a distance in both MT and non-MT.  I can't list all of my DCS settings, but I have terrain set to low, water high, AS 16X, heat blur low, motion blur off, PD set to 1, clouds set to "standard", shadows low and flat.  I'm sure it could be a setting somewhere, but there's some many variables it is impossible to sort it all out.

Hopefully, the next iteration will see improvements for those of us seeing none.

I think that there is an improvement, as I have documented/ droned on about here.  Unfortunately, it seems pretty minor, especially when compared with numerous YT videos out there for pancake with fps improvements ranging from 50%-100%.  Although my initial excitement has been somewhat tempered, hopefully we will see further improvements in the future.

I don't mean to offend anyone, but I've learnt to take comments such as "Multithreading Performance is Sensational in VR" with a generous pinch of salt unless they're backed up by screenshots / videos with numbers, charts, etc.

Edited March 11, 2023 by Hippo

[Flextremes]

17 minutes ago, Hippo said:

I think that there is an improvement, as I have documented/ droned on about here.  Unfortunately, it seems pretty minor, especially after seeing multiple YT videos for pancake with fps improvements ranging from 50%-100%.  Although my initial excitement has been somewhat tempered, hopefully we will see further improvements in the future.

I don't mean to offend anyone, but I've learnt to take comments such as "Multithreading Performance is Sensational in VR" with a generous pinch of salt unless they're backed up by screenshots / videos with numbers, charts, etc.

I understand your perspective. No offense taken  But just check the improvements in frame times with the carrier on-ramp screenshots here as an example. These are not just some numbers: actually flying in VR with all eye candy enabled is butter smooth, a big difference with Single Threaded performance. I can only double down on that 'sensational' statement.

If you focus on fps only, you probably will not fully understand the impact of MT. For instance: the Marianas map, Syriah and the SuperCarrier maps were clearly CPU bottlenecked. Now with MT it makes a big difference that is immediately noticed while flying.

Edited March 11, 2023 by Flextremes

[Hippo]

54 minutes ago, Flextremes said:

I understand your perspective. No offense taken  But just check the improvements in frame times with the carrier on-ramp screenshots here as an example. These are not just some numbers: actually flying in VR with all eye candy enabled is butter smooth, a big difference with Single Threaded performance. I can only double down on that 'sensational' statement.

If you focus on fps only, you probably will not fully understand the impact of MT. For instance: the Marianas map, Syriah and the SuperCarrier maps were clearly CPU bottlenecked. Now with MT it makes a big difference that is immediately noticed while flying.

 

Sorry if you've already mentioned it, I couldn't see it mentioned anywhere, but what VR hardware are you using?

If you have some time to kill, I posted a track in the thread I linked to, and I'd be very interested to hear about your experience replaying that track.  In particular towards the end where I'm flying low over the terrain.  Look out to your side at the ground when I'm flying / banking low over it.  Is it smooth without stutter?  What sort of fps do you get?  Is the frame rate stable? Is there any reprojection going on?

Edit:  the maps you mention I haven't tried.  I don't own Syria, and gave up on Marianas the first time I tried it (although I had a 1080ti at the time - unusable slide show).  I've only tested on Caucasus, Persian Gulf, and a bare Supercarrier on missions with very little happening.  Perhaps I might see greater improvements with the maps you mention, and with missions with more going on.  I did notice that the CPU was doing a lot less on the SC missions, perhaps because there was no terrain nearby.

Edited March 12, 2023 by Hippo

[ISantus]

I'm in love with the new MT and couldn't agree more that VR is sensational

• Windows 11 (all updates)
• 5900x (no OC)
• 4090 (8% OC) - latest drivers
• 32gb (4x8gb@3600)
• Reverb G2
• OpenXR via composite

null

[dburne]

1 hour ago, BoneDust said:

I'm running an Alienware New Aurora R9 | Windows® 10 Home Premium | 64bit, Intel(R) Core(TM) 9th Gen Intel Core i9-9900K 8C/16T R0| NVIDIA GeForce RTX 4070Ti 12GB | 2 X 1TB SSD | 64GB Dual Channel CORSAIR Vengence at 3200MHz | 850W EPA Gold Liquid Cooled power supply | G2 Rverb VR

I do not see any improvement in VR running MT.  Lost of folks are saying they're getting huge improvements, but for many of us, there's no improvement.  In fact, non-MT seems smoother.  The frame rates are comparable between both modes, drifting between 34-49fps depending on where I am, but I notice some texture tearing in the water from a distance in both MT and non-MT.  I can't list all of my DCS settings, but I have terrain set to low, water high, AS 16X, heat blur low, motion blur off, PD set to 1, clouds set to "standard", shadows low and flat.  I'm sure it could be a setting somewhere, but there's some many variables it is impossible to sort it all out.

Hopefully, the next iteration will see improvements for those of us seeing none.

 

Are you overclocking the cores on that 9900k cpu? I run mine  at 5.1 Ghz and combined with my 4090 GPU saw a nice improvement running the MT version.

[Flextremes]

58 minutes ago, Hippo said:

Sorry if you've already mentioned it, I couldn't see it mentioned anywhere, but what VR hardware are you using?

If you have some time to kill, I posted a track in the thread I linked to, and I'd be very interested to hear about your experience replaying that track.  In particular towards the end where I'm flying low over the terrain.  Look out to your side at the ground when I'm flying / banking low over it.  Is it smooth without stutter?  What sort of fps do you get?  Is the frame rate stable? Is there any reprojection going on?

 

My system: Highly optimized Win 10, AMD 5800x3D, 4090, Reverb G2, Native OpenXR, almost maxed out DCS settings with 80% FSR upscaled resolution to 3700x3600 with 100% sharpening.

With your track I got 90 fps, until we got close to the island. Then 'app GPU' frametimes go above 9ms: reprojection kicks in and fps is 45. It is all perfectly smooth, reprojection is fine and acceptable for me.  

I created screenshots at full resolution for you during the track: note the frame times and fps reported by openXR in the right corner below. 

Click on the image to get the high-res picture, and download (right-click, download save as) it to appreciate the insane fidelity DCS-MT is rendering on my Reverb. So freaking beautiful.

Look at both "rdr CPU" and "app CPU" frametimes: they remain ultra low because of MT. This would be way higher in ST.

 

For context: These are my DCS settings...

I

Edited March 12, 2023 by Flextremes

[EightyDuce]

Looking at majority of the threads where folks aren't seeing significant improvements are those with current gen mid-tier or high-tier last gen GPUs and were already largely GPU bound. Being completely GPU bound now that CPU bottleneck is gone isn't going to change much in raw FPS.

Those folks will benefit from DLSS far more than MT, however, MT still will have a significant impact in MP.

[KCOOL]

26 minutes ago, EightyDuce said:

Looking at majority of the threads where folks aren't seeing significant improvements are those with current gen mid-tier or high-tier last gen GPUs and were already largely GPU bound. Being completely GPU bound now that CPU bottleneck is gone isn't going to change much in raw FPS.

Those folks will benefit from DLSS far more than MT, however, MT still will have a significant impact in MP.

MT is nice but I also feel DLSS will provide with much more happiness to most users (most of us can't / won't have 4090s...).

Hopefully we can try some early DLSS experience in the next patches.

Edited March 12, 2023 by KCOOL

[Khangirey]

22 hours ago, nikoel said:

Not bugged, you're getting exactly the performance that you're supposed to. Your CPU is giving you a theoretical 250ish FPS (4ms), whilst your graphics card is giving you a practical 58FPS (17ms)

Therefore you're limited by your graphics card and you get your 58FPS

It's sometimes just easiest to use this calculator to give the FPS figures -https://fpstoms.com/

 

Thanks. As I understand, I'm actually GPU bound. I'm still not quite understanding why the in-game readout tells me I'm CPU bound.

What are some options that I can use to increase my GPU frame times? I tried the OXRTK FSR and NIS options, going all the way down to 80%, and I'm not getting increases in FPS.

I don't know if this has something to do with it, but my GPU utilization hovers around 30-40 percent. 

[TED]

Some mixed reviews and results here but for me this has been the single best improvement ED has done for years. I was sceptical and holding back from further purchases until vr performance increased. 

Running-

Win11

AMD 6900XT

AMD 5800X3D

REVERB G2.

Yesterday I did a test mission I have in the caucuses with f18 strike mission with accompanying a10c's and ah64's and quite a few ground assets and targets. I ran the mission first in traditional ST, then repeated in MT. The difference for me was huge. Low level in the f18 with most settings maxed or high except msaa (off) I would get around 40 fps. Now getting 55-60. 

I subsequently bumped up openxr to 120% and fix my fps to 45 via amd adrenaline chill mode and it's is completely stable and smooth. Probably I can bump up a few more settings but it's safe to say for me this is a massive improvement and way exceeded any expectations I had for performance. 

[MarkyMarkUK]

I'm on i7-10K 32gb ram 3090 and reverb g2.

On medium or high preset I seem to mostly be GPU bound. I can get around 60fps but it's jerky. I have to switch on OpenXR toolkit reprojection which helps but not amazing.

[Hippo]

9 hours ago, Flextremes said:

My system: Highly optimized Win 10, AMD 5800x3D, 4090, Reverb G2, Native OpenXR, almost maxed out DCS settings with 80% FSR upscaled resolution to 3700x3600 with 100% sharpening.

With your track I got 90 fps, until we got close to the island. Then 'app GPU' frametimes go above 9ms: reprojection kicks in and fps is 45. It is all perfectly smooth, reprojection is fine and acceptable for me.  

I created screenshots at full resolution for you during the track: note the frame times and fps reported by openXR in the right corner below. 

Click on the image to get the high-res picture, and download (right-click, download save as) it to appreciate the insane fidelity DCS-MT is rendering on my Reverb. So freaking beautiful.

Look at both "rdr CPU" and "app CPU" frametimes: they remain ultra low because of MT. This would be way higher in ST.

Firstly:  thank you so much for taking the time to do that.

I am unfamiliar with the G2 so I assume that it runs "native" at 90 Hz, and drops to 45 Hz with reproj when the system can't keep up.  Reprojection is acceptable for me, up to a point.  I can live with it in helos, but the artifacting it introduces in fast jets as maneuvers can be much more violent is a problem.  Ideally I would want not reproj, or for it only to kick under very heavy loads.  I prefer to drop the eye candy to maintain native if possible.  I understand this point is very subjective.  Maybe the G2 does reproj better?  And throwing FSR into the mix?

For me, I was able to maintain 72fps (native) throughtout (i.e. I could not cause a single drop into reproj) with MT.  With ST I could induce ocassional drops to reproj if I looked around while flying close to the ground.  Unfortunately in MT, I could see visible terrain stuttering when looking out to my left/ right while banking over the island, even though I was running at a constant 72 Hz.  I did not see this stuttering with ST.  Sorry to be pedantic, but could you unequivocally confirm that you did not experience this?

Forgive me, but I am unsure as to what exactly your frame times are telling you.  Unless I am misunderstanding things frame time is just the inverse of fps, and has the advantage that comparisons between numbers are in direct proportion.  I am old school (or just old) and have been using fps for decades, so have a good "feel" for what they represent in reality, and continue to use fps in comparisons, though I probably should switch.  As another poster very reasonably suggests, if the cpu frame time to which you refer is a theoretical max whilst in reality what you are actually seeing is limited by your gpu frametime then I would suggest that it is not all that relevant.

I hate to impose on your good nature, but what would be really useful is if you could also run the track ST and MT and see if you notice any differences in fps, especially when looking directly at the ground beneath you when the plane is in a bank.  For extra credit it you could fire up msi afterburner and screenshot a graph of what all your cores and gpu %util are doing throughout, this would be very useful information to diagnose why you are seeing fps improvements that others are not.  It is only where the aircraft is over the island that the system is under stress, so a pretty poor choice of track from me... sorry.

This chap demonstrates a performance increase between an older ST version and the current MT version, but what is interesting is that he states in the comments that he's noticed no difference between the current ST and MT versions in VR.  The concerning implication is that he would have seen exactly the same increase if he had compared the new ST version with the old ST version, and one has to be wonder whether any improvements people are noticing in VR are down to other changes and not MT itself.  I unfortunately did not run tests with the version before the update, but from memory I'm almost certain that performance was considerably worse.  Most (all?) of the other YT videos that I've seen demonstrating large performance increases are in pancake.

I don't doubt that you are seeing what you say, and again I'm sorry if any of my remarks came across as snarky - it really isn't my intention, but from all the other evidence I've seen, VR performance improvements appear to be marginal at best.

Edit and note:  I notice that many posters say things like I got 35 fps before and now I get 50 fps.  I can only assume that they are running w/o reprojection, i.e. with unlocked constantly varying fps.  I gave up on doing this years ago as I found it too uncomfortable, and always run with reproj, so I only ever see 36/72 fps, which is why I compare cpu / gpu loads at the same fps.

Edited March 12, 2023 by Hippo

[Flextremes]

2 hours ago, Hippo said:

Firstly:  thank you so much for taking the time to do that.

I am unfamiliar with the G2 so I assume that it runs "native" at 90 Hz, and drops to 45 Hz with reproj when the system can't keep up.  Reprojection is acceptable for me, up to a point.  I can live with it in helos, but the artifacting it introduces in fast jets as maneuvers can be much more violent is a problem.  Ideally I would want not reproj, or for it only to kick under very heavy loads.  I prefer to drop the eye candy to maintain native if possible.  I understand this point is very subjective.  Maybe the G2 does reproj better?  And throwing FSR into the mix?

For me, I was able to maintain 72fps (native) throughtout (i.e. I could not cause a single drop into reproj) with MT.  With ST I could induce ocassional drops to reproj if I looked around while flying close to the ground.  Unfortunately in MT, I could see visible terrain stuttering when looking out to my left/ right while banking over the island, even though I was running at a constant 72 Hz.  I did not see this stuttering with ST.  Sorry to be pedantic, but could you unequivocally confirm that you did not experience this?

Forgive me, but I am unsure as to what exactly your frame times are telling you.  Unless I am misunderstanding things frame time is just the inverse of fps, and has the advantage that comparisons between numbers are in direct proportion.  I am old school (or just old) and have been using fps for decades, so have a good "feel" for what they represent in reality, and continue to use fps in comparisons, though I probably should switch.  As another poster very reasonably suggests, if the cpu frame time to which you refer is a theoretical max whilst in reality what you are actually seeing is limited by your gpu frametime then I would suggest that it is not all that relevant.

I hate to impose on your good nature, but what would be really useful is if you could also run the track ST and MT and see if you notice any differences in fps, especially when looking directly at the ground beneath you when the plane is in a bank.  For extra credit it you could fire up msi afterburner and screenshot a graph of what all your cores and gpu %util are doing throughout, this would be very useful information to diagnose why you are seeing fps improvements that others are not.  It is only where the aircraft is over the island that the system is under stress, so a pretty poor choice of track from me... sorry.

This chap demonstrates a performance increase between an older ST version and the current MT version, but what is interesting is that he states in the comments that he's noticed no difference between the current ST and MT versions in VR.  The concerning implication is that he would have seen exactly the same increase if he had compared the new ST version with the old ST version, and one has to be wonder whether any improvements people are noticing in VR are down to other changes and not MT itself.  I unfortunately did not run tests with the version before the update, but from memory I'm almost certain that performance was considerably worse.  Most (all?) of the other YT videos that I've seen demonstrating large performance increases are in pancake.

I don't doubt that you are seeing what you say, and again I'm sorry if any of my remarks came across as snarky - it really isn't my intention, but from all the other evidence I've seen, VR performance improvements appear to be marginal at best.

Edit and note:  I notice that many posters say things like I got 35 fps before and now I get 50 fps.  I can only assume that they are running w/o reprojection, i.e. with unlocked constantly varying fps.  I gave up on doing this years ago as I found it too uncomfortable, and always run with reproj, so I only ever see 36/72 fps, which is why I compare cpu / gpu loads at the same fps.

 

Let me start off by saying that this is a great conversation. It is very clear that many, like you, do not experience the significant improvement in fluidity so many are reporting. We need to figure out what is going on. 

I tested your track again until the end: I experienced zero stutters, perfectly smooth. The only reason I went 46 fps is that I have enabled all eye candy (that would destroy performance in ST, btw) and I am running at a very high resolution.

1. Seems like there is a trend: folks on streamVR are generally not happy. It seems like this first version of MT DCS is optimized for native openXR. ED is aware of these stutter problems: they mention it specifically at release. I am hopeful they can fix this: because it seems for open XR native this does not seem a problem. I also suspect that SteamVR and Oculus have a considerably higher CPU overhead than native Open XR. (conversely: reprojection is no way near as good in openXR) 
2. Also, it is clear that DCS MT seems to prefer newer CPU's with higher core counts. MT can even render worse results on old CPUs with only a few cores. This makes sense because managing MT processes instead of ST does incur a sizable overhead for tracking and syncing all those threads. If you do not have enough cores, ST will the better option.
3. Lastly, it is also clear that when somebody is GPU limited you should not expect higher FPS. But this is only half of the story: DCS always has been severely CPU bottlenecked: just check multiplayer, Supercarrier, high complexity missions, and even maps like the Marianas: CPU latency is all over the place. With MT these are fundamentally more fluid now. That is why I put so much emphasis on the Carrier example: Total CPU latency combined (app + rdr) in ST: 41.1ms, and in MT CPU combined is only 16.2 ms: this is x2.5 faster. Especially the 'app CPU' 22.5ms (ST) vs 4.0ms (MT) difference is stark: this is more than 5x faster. In VR this means that the chances of the CPU bottlenecking your VR experience is so much lower. 

I am a casual flyer since Flanker 2.0. (Yep I am that old). Many moons ago: only the DX11 update improved performance in DCS. But never in the history of DCS updates have we got this kind of performance improvement. 

But I do think we need to manage expectations: Always 90fps in VR without any drops in a simulation like DCS is close to impossible.

DCS is such a complex program, simulating the constraints of reality like AI, war logic, physics, technology, weather, multiplayer, and graphics. DCS IMHO dwarfs MSFS in complexity: There is absolutely nothing (for consumers) that comes close. That is why I am so appreciative of MT in DCS. 

 

 

Edited March 12, 2023 by Flextremes

[Hippo]

1 hour ago, Flextremes said:

Let me start off by saying that this is a great conversation. It is very clear that many, like you, do not experience the significant improvement in fluidity so many are reporting. We need to figure out what is going on. 

I tested your track again until the end: I experienced zero stutters, perfectly smooth. The only reason I went 46 fps is that I have enabled all eye candy (that would destroy performance in ST, btw) and I am running at a very high resolution.

1. Seems like there is a trend: folks on streamVR are generally not happy. It seems like this first version of MT DCS is optimized for native openXR. ED is aware of these stutter problems: they mention it specifically at release. I am hopeful they can fix this: because it seems for open XR native this does not seem a problem. I also suspect that SteamVR and Oculus have a considerably higher CPU overhead than native Open XR. (conversely: reprojection is no way near as good in openXR) 
2. Also, it is clear that DCS MT seems to prefer newer CPU's with higher core counts. MT can even render worse results on old CPUs with only a few cores. This makes sense because managing MT processes instead of ST does incur a sizable overhead for tracking and syncing all those threads. If you do not have enough cores, ST will the better option.
3. Lastly, it is also clear that when somebody is GPU limited you should not expect higher FPS. But this is only half of the story: DCS always has been severely CPU bottlenecked: just check multiplayer, Supercarrier, high complexity missions, and even maps like the Marianas: CPU latency is all over the place. With MT these are fundamentally more fluid now. That is why I put so much emphasis on the Carrier example: Total CPU latency combined (app + rdr) in ST: 41.1ms, and in MT CPU combined is only 16.2 ms: this is x2.5 faster. Especially the 'app CPU' 22.5ms (ST) vs 4.0ms (MT) difference is stark: this is more than 5x faster. In VR this means that the chances of the CPU bottlenecking your VR experience is so much lower. 

I am a casual flyer since Flanker 2.0. (Yep I am that old). Many moons ago: only the DX11 update improved performance in DCS. But never in the history of DCS updates have we got this kind of performance improvement. 

But I do think we need to manage expectations: Always 90fps in VR without any drops in a simulation like DCS is close to impossible.

DCS is such a complex program, simulating the constraints of reality like AI, war logic, physics, technology, weather, multiplayer, and graphics. DCS IMHO dwarfs MSFS in complexity: There is absolutely nothing (for consumers) that comes close. That is why I am so appreciative of MT in DCS. 

 

Thanks again, many great points there.

At the risk of being tedious,

Quote

I tested your track again until the end: I experienced zero stutters, perfectly smooth. The only reason I went 46 fps is that I have enabled all eye candy (that would destroy performance in ST, btw) and I am running at a very high resolution.

did you try my track with the ST exe, and was performance destroyed or noticeably worse?

Quote

Lastly, it is also clear that when somebody is GPU limited you should not expect higher FPS. But this is only half of the story: DCS always has been severely CPU bottlenecked: just check multiplayer, Supercarrier, high complexity missions, and even maps like the Marianas: CPU latency is all over the place. With MT these are fundamentally more fluid now. That is why I put so much emphasis on the Carrier example: Total CPU latency combined (app + rdr) in ST: 41.1ms, and in MT CPU combined is only 16.2 ms: this is x2.5 faster. Especially the 'app CPU' 22.5ms (ST) vs 4.0ms (MT) difference is stark: this is more than 5x faster. In VR this means that the chances of the CPU bottlenecking your VR experience is so much lower. 

I think this is going over my head.  What is meant by CPU latency?  Am I understanding correctly that the likelyhood of CPU bottlenecks tanking frame rates in specific situations with very high settings is greatly reduced in MT, but that in situations like my track with lowish settings like mine it will make no noticeable difference to frame rates?

I refer you again to the YT video I linked to.  That guy has a top end system meeting all of your requirements above and yet sees no difference between the ST and MT versions of the exe.  But I take your point on focusing exclusively on the fps number.

The ED FAQ states:

Quote

Will VR and flatscreen have an equal performance improvement?
MT is agnostic between a 2D monitor and VR. Both will see the same proportion of improvement. Note that VR is in general more demanding so you will see lower FPS than flatscreen performance given the higher rendering overhead.

I can find numerous pancake YT videos demonstrating fps increases varying from 50%-100%, yet am unable to find VR videos showing similar gains.  My own view is that the above statement is at worst incorrect, and at best needs to be qualified (as to what exactly they mean by "performance").

Edited March 12, 2023 by Hippo

[Zoijar]

Hmm I'm not sure yet. Data point:

• Default overclocked RTX 3090, Ryzen9 5900X 12-core, 64GB ram
• SteamVR / Valve Index
• 90Hz, 120% steam upscaling, motion reprojection turned off
• DCS pretty maxed out, almost all high/ultra, medium shadows, 4x anisotropic filtering, 2x MSAA, none of SS(AO etc), 1.7 pixel density, all sliders to the right, default shadows + global illumination
• Mostly 45 FPS in flight (90hz / 2). Looks good, feels smooth. Except when looking 90 degrees to the ground, there's this weird stutter or warp every second or two, mostly when stuff moves parallel from you (motion reprojection is off). That's pretty annoying.
• F16 training startup mission with some extra traffic and clouds it alternates between 30 and 45 FPS (90 / 2, 90 / 3) on the ground. 45 FPS in the air.
• DCS itself says I'm either CPU rendering thread or main thread bound. It says of the 22ms frame time (45 FPS) 15ms is spent on CPU. So with 15ms CPU I'll never get 90 FPS even if my GPU would be infinitely powerful...

I'll have to do some more in-depth timing to see where my cores and GPU are. I'm OK with 45 FPS, but I feel I could even increase the fidelity more. Even pixel density 2.0 doesn't seem  to change my frame rate. So the challenge seems to be either how far I can push my GPU as I'm 12-core CPU bound anyway, or what to reduce in settings so I might hit 90 FPS -- not the GPU settings it seems.

Edited March 12, 2023 by Zoijar

[Flextremes]

3 hours ago, Hippo said:

What is meant by CPU latency?  

I am lazy, so I asked Chat GPT explain:

Can you explain how both CPU frame time latency and GPU frame time latency affect fps

Certainly! Both CPU and GPU frame time latency can affect frame rate in Virtual Reality (VR) applications.

The CPU is responsible for managing the overall system resources and directing the flow of data between the VR application and the GPU. If the CPU is unable to keep up with the demands of the VR application, it may take longer for the data to be sent to the GPU for processing, resulting in higher CPU frame time latency.

When CPU frame time latency is high, the GPU may have to wait for the CPU to finish processing before it can start rendering the next frame, causing the frame rate to drop. In other words, if the CPU is not able to feed enough data to the GPU in time, the GPU will have to wait, and the frame rate will suffer.

Similarly, GPU frame time latency measures the time it takes for the GPU to render and display a frame. If the GPU is unable to keep up with the demands of the VR application, it may take longer to render each frame, resulting in higher GPU frame time latency.

When GPU frame time latency is high, the CPU may have to wait for the GPU to finish rendering before it can send more data to be processed, again leading to a drop in frame rate.

Overall, CPU and GPU frame time latency are closely linked to frame rate in VR applications. If either the CPU or GPU frame time latency is high, it can cause delays in the frame rendering process, leading to a lower frame rate and potentially causing motion sickness or breaking immersion. Therefore, it's important to ensure that both the CPU and GPU are capable of handling the demands of the VR application to achieve a smooth and responsive VR experience.

What CPU frame time latency and GPU frame time latency do you need to render at 90fps?

To render at a consistent 90fps frame rate in Virtual Reality (VR), the CPU frame time latency and GPU frame time latency should ideally be under 11 milliseconds (ms) each.

This is because 90fps means each frame needs to be rendered within 11.11ms (1 second divided by 90 frames), and any delay beyond this threshold can cause lag, motion sickness, or break immersion.

Therefore, both the CPU and GPU need to be capable of processing and rendering the VR application's data within the required time frame to maintain a smooth and responsive experience.

It's worth noting that achieving this level of performance can depend on various factors such as the complexity of the VR application, the graphics settings used, the hardware specifications of the CPU and GPU, and other factors. However, as a general rule, maintaining low frame time latency for both the CPU and GPU is critical for achieving a consistent and high frame rate in VR.

Edited March 12, 2023 by Flextremes