investigating Large performance drop in VR after updating to 2.8

[speed-of-heat]

7 hours ago, pastranario3 said:

Well I've got the fps back by lowering pixel density to 0.5 and increasing the multiplier since steam vr. For hp reverb 1: pixel density 1.0+steamvr 100% = pixel density 0.5+steamvr 400%. In the forum there is a calculator that helps to find the equivalences, search as PDSSCalc

I wrote pdsscalc, there should be no difference in performance or clarity between if the total number of pixels are the same (which with 0.5 and 400, they would be) The processing is all done on the total pixels.  I will give it a go later.

Edited November 11, 2022 by speed-of-heat

[slughead]

8 hours ago, RuskyV said:

Yes, I have cockpit reflections disabled as a personal preference. Although I have noticed that the option seems to have no impact on the cockpit instruments, especially the warbirds with the traditional dials.

 

You don't say if the performance inside the cockpit is better without reflections on. I'd expect it would be worse but that would be the same for 2.7. You should compare the difference in performance in 2.7 with and without cockpit reflections and the difference in 2.8 with and without reflections. Then we/you would know if there is any new impact in 2.8.

[pastranario3]

hace 3 horas, speed-of-heat dijo:

I wrote pdsscalc, there should be no difference in performance or clarity between if the total number of pixels are the same (which with 0.5 and 400, they would be) The processing is all done on the total pixels.  I will give it a go later.

 

I have already talked about this several times, there are even tests that show it just by searching reddit, YouTube, even in the dcs forum. I don't have time to make a video but the test is very simple. Logic says that it would be the same to insert the values but practice shows otherwise, in vr there is no logic.

[WipeUout]

38 minutes ago, pastranario3 said:

Logic says that it would be the same to insert the values but practice shows otherwise, in vr there is no logic.

There is logic in it.  Depending which software layer you use for supersampling, there are slight differences.  I found that SteamVR is better for upscaling than using DCS pixel density or even ''render quality'' in PiTool.  There are many ways to upscale an image and all those software do not use a universal upscaler. 

[RuskyV]

2 hours ago, slughead said:

You don't say if the performance inside the cockpit is better without reflections on. I'd expect it would be worse but that would be the same for 2.7. You should compare the difference in performance in 2.7 with and without cockpit reflections and the difference in 2.8 with and without reflections. Then we/you would know if there is any new impact in 2.8.

Good point, I'll check it later 

[Baltic Pirate]

On 10/29/2022 at 1:56 AM, Sr. said:

Show of hands... how many here are running a 4090? 

Not yet. Waiting for mine. But really what I cannot wait is that DCS introduces multicore. These are nice, but not. That is a huge performance drop in VR. I use it exclusively. Not monitor at all. Alas I will not be updating until issues are solved.  One would think that multicore and VR support would be a priority. VR and AR are the present and the future with dropping prices on hardware and more available options.  I like the platform a lot, but if we take a step back with new updates in performance, well not much to look forward to.

[5ephir0th]

51 minutes ago, pastranario3 said:

I have already talked about this several times, there are even tests that show it just by searching reddit, YouTube, even in the dcs forum. I don't have time to make a video but the test is very simple. Logic says that it would be the same to insert the values but practice shows otherwise, in vr there is no logic.

Some of those "Youtube tests" was wrong as they are not comparing same pixel count, i did my test over Oculus software and, as @speed-of-heatsaid, comparing similar end pixel counts gets similar performance and visual quality, theres no magic

[speed-of-heat]

1 hour ago, 5ephir0th said:

Some of those "Youtube tests" was wrong as they are not comparing same pixel count, i did my test over Oculus software and, as @speed-of-heatsaid, comparing similar end pixel counts gets similar performance and visual quality, theres no magic

This pretty much, but, it won’t take much effort to run around the buoy again… 

[Jarmak]

On 11/5/2022 at 4:50 AM, BIGNEWY said:

People appreciate you are passionate, but keep it friendly ( rules are at the top of the forum ).

As mentioned in the newsletter we are looking into the VR performance dip some have seen, it maybe related to the new features and clouds moving, it maybe something else. It is not 100% clear yet. 

My performance on my rig has not changed, I am seeing no performance issues which makes it harder for me to investigate it. 

I would suggest if you want help to post as much info as possible
DXDIAG
DCS.log
Screenshots of GPU control panel settings
VR headset settings and res. 

we need data to check to try to figure it out, people who are posting that their performance has dropped but giving no detail isnt very helpful. 

Also to add. 

Like it or not open beta is and always will be a public test version. 

It is where we find and try to resolve the crashes and issues we can not find in closed testing before we push a build to stable. 

DCS will always work this way, we are constantly evolving and updating, for more than a decade now and we will continue to update and bring new features to you all. 

thanks

As someone who actually does project management as a job IRL you don't have a public test build, you have a primary public release that you are releasing as WIP, and a backup stable build that your community barely touches.

There are a wide variety of methods used by software firms to keep their open betas and test builds from becoming the primary version their customer's are using, in order to avoid this very issue. ED does pretty much exactly the opposite of that and spends all of it's marketing promoting the open beta releases while barely mentioning the stable version. As an extremely predictable result the primary product your customer base is consuming is the open beta, and that customer base is going to have some expectations of a base level of functionality.

And honestly, it really doesn't matter what you think about it, what you label things, or how much you argue about it. Your customers have developed expectations as the very predictable result of decisions you've made in how you position the product, and they're going to be upset if you don't meet them. Telling them they're not allowed to be upset is just going to make them more upset; this is marketing 101 stuff. 

Positioning it as a WIP/Open Beta buys you a lot of leeway to have glitches or temporary downtime in functionality, but tapping the open beta sign at your power users when they're upset it's just straight not working for them anymore is not going to get you anywhere productive. 

[Antonisrho]

Hey folks, same here with huge FPS drop (I have an old I7 4770K processor topping at 4.2 and a 3080, oculus quest 2 and 32 GB of RAM....I know time for an upgrade!!!). 

I was having the same issues but having such an old processor my performance drop made the game unplayable. So I reverted back to the last 2.7 version.  Then after reading about people trying this and the other I though to try 2.8 once more. Since I updated to the last version (which included the hotfix) I didn't have such a bad performance anymore! There was some loss but not as bad as before (having gone from latest 2.7 to 2.8 and then to the hotfix).

Having a Quest 2 I also followed this tutorial below and my performance has jumped to 2.7 levels....or even better  

 

 

I know we shouldn't be going through all this tweaking....but what can you do so maybe try to revert back to latest 2.7 and then try to update again to 2.8 (which now is cleaner as it includes the hotfix).

 

Obviously do the necessary deletion of FXO metashaders2 etc....

 

 

Edited November 11, 2022 by Antonisrho

[YoYo]

I dont see any advantage of this for WMR and HP Reverb. Now, only terrain shadows OFF gives quite ok result. Im waiting for the next beta patch with optimalization . Personally I dont want to use any performance mods too, like VR shaders etc. Its a case of next issues and problems mostly. Here I saw infos about mouse issues.

Edited November 12, 2022 by YoYo

[speed-of-heat]

I saw an issue, that was perf related on some games on some driver subsets, allegedly nvidia have fixed it in their recent driver, so that will be my testing for today

[speed-of-heat]

On 11/11/2022 at 10:09 AM, pastranario3 said:

I have already talked about this several times, there are even tests that show it just by searching reddit, YouTube, even in the dcs forum. I don't have time to make a video but the test is very simple. Logic says that it would be the same to insert the values but practice shows otherwise, in vr there is no logic.

Ok so if it works for you I am really happy for you, but I can say it really doesn't work for me. In a normal test, with fixed head position, (its mounted so it doesn't move to minimise any impact on viewed objects etc...) ; these are my results, not I am not using steam but OpenXR so that might be a factor, but it never was before I moved from steam so I don't believe it will be now:

FPS	CPU	GPU	Notes
58.1	7.68	15.95	PD 1.0 Render Scale 100%
57.6	7.65	16.15	PD 0.5 Render Scale 400%

frame times are in miliseconds.

The test was conducted in a F/A-18, over the Caucuses, using a replay track (attached) to conduct each test after allowing the machine to cool and reloaded after each adjustment.

Whilst there is a nominal loss of performance at PD0.5 and 400%; this is inside the margin of error for these kinds of tests, and I would have to run multiple tests against the same set to get a real sense of if its real or not, and honestly, I can't be bothered 

Conclusion

This is still performance driven on the total number of pixels, and how you derive them PD or Render scale is totally irrelevant, it is the total number that is the determining factor. e.g. no magic.

performance track 2.8.0.trk

Edited November 12, 2022 by speed-of-heat

[speed-of-heat]

2 hours ago, speed-of-heat said:

I saw an issue, that was perf related on some games on some driver subsets, allegedly nvidia have fixed it in their recent driver, so that will be my testing for today

so I also tried the latest nVidia driver, no real substantive difference, but the numbers are here:

FPS	CPU(ms)	GPU (ms)	Notes
58.10	7.68	15.95	PD 1.0 Render Scale 100%
58.10	7.53	15.64	526.86 Driver

there is a very slight improvement, but this is margin of error and it has no practical impact in anycase.

[pastranario3]

hace 6 horas, speed-of-heat dijo:

Ok so if it works for you I am really happy for you, but I can say it really doesn't work for me. In a normal test, with fixed head position, (its mounted so it doesn't move to minimise any impact on viewed objects etc...) ; these are my results, not I am not using steam but OpenXR so that might be a factor, but it never was before I moved from steam so I don't believe it will be now:

FPS	CPU	GPU	Notes
58.1	7.68	15.95	PD 1.0 Render Scale 100%
57.6	7.65	16.15	PD 0.5 Render Scale 400%

frame times are in miliseconds.

The test was conducted in a F/A-18, over the Caucuses, using a replay track (attached) to conduct each test after allowing the machine to cool and reloaded after each adjustment.

Whilst there is a nominal loss of performance at PD0.5 and 400%; this is inside the margin of error for these kinds of tests, and I would have to run multiple tests against the same set to get a real sense of if its real or not, and honestly, I can't be bothered 

Conclusion

This is still performance driven on the total number of pixels, and how you derive them PD or Render scale is totally irrelevant, it is the total number that is the determining factor. e.g. no magic.

performance track 2.8.0.trk 1.78 MB · 3 descargas

 

As has become usual in this forum try to demonstrate each thing with which you try to help. Here I upload two videos where the first is PD 0.5 - SteamVR 400% and the second PD 1 - STEAMVR 100%. I understand that each rig is different, one shares trying to help, the one who is served welcome to the one who doesn't and well...

 

PD 0.5 - STEAMVR 400% -45FPS

 

 

PD 1.0 - STEAMVR 100% - 35FPS

 

 

 

Edited November 12, 2022 by pastranario3

[speed-of-heat]

as i said i cant replicate it, if you can good for you 

[WipeUout]

1 hour ago, pastranario3 said:

As has become usual in this forum try to demonstrate each thing with which you try to help. Here I upload two videos where the first is PD 0.5 - SteamVR 400% and the second PD 1 - STEAMVR 100%. I understand that each rig is different, one shares trying to help, the one who is served welcome to the one who doesn't and well...

 

PD 0.5 - STEAMVR 400% -45FPS

 

 

PD 1.0 - STEAMVR 100% - 35FPS

 

Is this some kind of a joke?  It not even the same video and you compare FPS?  At least go through the effort of making a track and test with a minimum of rigor!!

With the same settings, if pixel count is the same, performance will be the same and image quality will be the same.  As it was said and demonstrated, there is no magic.   

Edited November 12, 2022 by WipeUout
Typo

[pastranario3]

hace 20 minutos, WipeUout dijo:

Is this some kind of a joke?  It not even the same video and you compare FPS?  At least go through the effort of making a track and test with a minimum of rigor!!

With the same settings, if pixel count is the same, performance will be the same and image quality will be the same.  As it was said and demonstrated, there is no magic.   

 

haha too much, until here comes the one who wants to try it the one who doesn't. Make your own videos. Greetings.

[exil]

3 hours ago, pastranario3 said:

As has become usual in this forum try to demonstrate each thing with which you try to help. Here I upload two videos where the first is PD 0.5 - SteamVR 400% and the second PD 1 - STEAMVR 100%. I understand that each rig is different, one shares trying to help, the one who is served welcome to the one who doesn't and well...

 

PD 0.5 - STEAMVR 400% -45FPS

 

 

PD 1.0 - STEAMVR 100% - 35FPS

 

 

 

 

Sorry, but these are two different setups. 

I would measure GPU frametimes instead of FPS (to cleaning out any CPU factor) and do a track file and replay that with the headset lying on a table or a mount. 

No rant here, but this way it's not really comparable. 

[WipeUout]

12 hours ago, pastranario3 said:

haha too much, until here comes the one who wants to try it the one who doesn't. Make your own videos. Greetings.

Oh but I tried, and results were just the same with 0.3 FPS apart.  Use a track, it is more rigorous and credible.

[speed-of-heat]

Well I don’t “know” I have my suspicions, but, they are unfounded as I have no evidence, I can say that such and such is different (such as the shadows in the cockpit canopy) but without the ability to turn that off  and measure the delta, it’s pretty hard (for us) to say *this is the cause*, what makes it hard for ED is it’s not consistent, some people got a perf bump, some not, some a hit.  The code paths are quite complex when you take in all the options that are possible, from different versionsnof hardware, to windows, different types of vr,  to variations in settings and mods and on and on…

[fab.13]

20 minutes ago, AirMeister said:

Does anyone know the actual cause ?

And i don't mean - turn of these options for higher fps -. 
Is it here to stay because they did something with the clouds ?

Or are the shadows all calculated by the 1 core in the CPU instead of the graphics card ?

What is the actual cause and is it actually a bug or just not solvable with the current code or engine or whatever ?

Hello

if New code has been added to dcs ( added and not replaced ), then frametime must be affected ( more computing )

And new code comes with bugs sometime…( beta   )

Same about gpu: more Light source and so on

we already met DCS open beta fps dropdown when the previous  Lightning systèm was introduced

 

and we have been drived to redo settings from scratch while ed has improved performances in the same Time ( it taked time )

 

 

 

 

 

 

 

 

 

 

 

 

[Burt]

The only way is to go back to 2D. I cant tweak this perf issue out. 

[fab.13]

The last hotfix about shadows is an example, let’s see if another patch will improve this point or not

loading Time in some past version was very bad and have been improved..,

So improvment or bug fix about performances exist

 

[C3PO]

On 11/11/2022 at 4:49 PM, Jarmak said:

As someone who actually does project management as a job IRL you don't have a public test build, you have a primary public release that you are releasing as WIP, and a backup stable build that your community barely touches.

There are a wide variety of methods used by software firms to keep their open betas and test builds from becoming the primary version their customer's are using, in order to avoid this very issue. ED does pretty much exactly the opposite of that and spends all of it's marketing promoting the open beta releases while barely mentioning the stable version. As an extremely predictable result the primary product your customer base is consuming is the open beta, and that customer base is going to have some expectations of a base level of functionality.

And honestly, it really doesn't matter what you think about it, what you label things, or how much you argue about it. Your customers have developed expectations as the very predictable result of decisions you've made in how you position the product, and they're going to be upset if you don't meet them. Telling them they're not allowed to be upset is just going to make them more upset; this is marketing 101 stuff. 

Positioning it as a WIP/Open Beta buys you a lot of leeway to have glitches or temporary downtime in functionality, but tapping the open beta sign at your power users when they're upset it's just straight not working for them anymore is not going to get you anywhere productive. 

Couldn't put it better myself. Excellent post.