2.8 update has 20 % fps loss on pimax8kx and 15 fps loss on Varjo Aero (RTX 4090)

[slughead]

I've reverted to 2.7 and intend to spend my time flying than second guessing what has gone wrong. ED will be back on Monday and will no doubt look into this and make an announcement in the week or release a fast patch.

[Moxica]

20 hours ago, Burt said:

Things have to bake ! Give it time.

Actually.. Ooor as a matter of fact.. I'll be damned..
(I had to do some manual baking in Oculus debug tool, though. Some changes had happened there without my knowlegde; "Link sharpening" was reset to auto, "Asynchronous spacewarp" was locked to 45 frames. Those were not my settings))

Edited October 30, 2022 by Moxica

[exil]

I've just benched the difference between 2.7 and 2.8. My system specs: Ryzen 5800X3D, RTX 4090, 64GB RAM @ 3400Mhz, HP Reverb G1 with 100% Resolution, no overclocking

I used three different scenarios for the tests with the results in regards to in- or decrease in frametimes. All results are taken with the 'high' preset and with secondary shadows to off. Everything else was left as within the preset.

1. F18 Ready on the Ramp @ Caucasus

CPU Avg in/decrease: +1.1%

GPU Avg in/decrease: -7.9%

2. F16 Free Flight @ Syria

CPU in/decrease: minFT +7.0%, maxFT 6.2%, avgFT +6.6%

GPU in/decrease: minFT -9.8%, maxFT -8.7%, avgFT -9.3%

3. F18 on the deck @ Caucasus

CPU in/decrease: minFT +9.6%, maxFT +11.8%, avgFT +10.7%

GPU in/decrease: minFT -3.0%, maxFT -4.0%, avgFT -3.5%

Worth to note, that this was no sophisticated test because OXR doesn't have a good measuring tool like steamVRs FPSVR. So all i could do is write down min and max frametime values and compare them. But it at least gives an impression what is currently happening here.

To me, the uplift in CPU frametime was somehow unexspected but comes with a grain of salt because of the decreased GPU performance. Given that the game was more GPU than CPU limited for me (at least with the 5800X3D), this drives me faster into a 45 FPS reprojectoion (or even lower) because the GPU frametimes have always been a bit higher than the CPU ones.

The question now is, why are there such tremendous differences between various rigs. I'm on the happy side, only loosing about 10% performance, but other people are actually bleeding with loosing about 10 - 15 FPS total (not %).

 

[slughead]

3 minutes ago, exil said:

I've just benched the difference between 2.7 and 2.8. My system specs: Ryzen 5800X3D, RTX 4090, 64GB RAM @ 3400Mhz, HP Reverb G1 with 100% Resolution, no overclocking

I used three different scenarios for the tests with the results in regards to in- or decrease in frametimes. All results are taken with the 'high' preset and with secondary shadows to off. Everything else was left as within the preset.

1. F18 Ready on the Ramp @ Caucasus

CPU Avg in/decrease: +1.1%

GPU Avg in/decrease: -7.9%

2. F16 Free Flight @ Syria

CPU in/decrease: minFT +7.0%, maxFT 6.2%, avgFT +6.6%

GPU in/decrease: minFT -9.8%, maxFT -8.7%, avgFT -9.3%

3. F18 on the deck @ Caucasus

CPU in/decrease: minFT +9.6%, maxFT +11.8%, avgFT +10.7%

GPU in/decrease: minFT -3.0%, maxFT -4.0%, avgFT -3.5%

Worth to note, that this was no sophisticated test because OXR doesn't have a good measuring tool like steamVRs FPSVR. So all i could do is write down min and max frametime values and compare them. But it at least gives an impression what is currently happening here.

To me, the uplift in CPU frametime was somehow unexspected but comes with a grain of salt because of the decreased GPU performance. Given that the game was more GPU than CPU limited for me (at least with the 5800X3D), this drives me faster into a 45 FPS reprojectoion (or even lower) because the GPU frametimes have always been a bit higher than the CPU ones.

The question now is, why are there such tremendous differences between various rigs. I'm on the happy side, only loosing about 10% performance, but other people are actually bleeding with loosing about 10 - 15 FPS total (not %).

 

You can capture a trace with OpenXR Toolkit....

[exil]

12 minutes ago, slughead said:

You can capture a trace with OpenXR Toolkit....

I know, but I am obviously too stupid to read that file. Tried it with excel... 

[mbucchia]

2 hours ago, exil said:

I know, but I am obviously too stupid to read that file. Tried it with excel... 

Capture trace is not at all the option being mentioned in the release notes.

Capture trace is for advanced debugging. The performance recording is accessible from the OpenXR Toolkit in-game menu in the Performance tab, called Record statistics On/Off. That one will create a .csv file that you will have no problem to open.

Edited October 30, 2022 by mbucchia

[exil]

28 minutes ago, mbucchia said:

Capture trace is not at all the option being mentioned in the release notes.

Capture trace is for advanced debugging. The performance recording is accessible from the OpenXR Toolkit in-game menu in the Performance tab, called Record statistics On/Off. That one will create a .csv file that you will have no problem to open.

 

Thank you. I am already that far, but I don't know how to read out that csv file. I am able to open it in excel, but there ain't the values I'm looking for. 

With fpsVR I was able to record let's say a 60 seconds track which would output avg cpu and avg gpu frametimes recorded over the 60 seconds track. 

[MoleUK]

I can at least point to one indicator of the new low FPS. The update ms time in the DCS FPS breakdown indicator has spiked.

Included two screenshots. Same module, same server, same settings, minutes apart. Once that update time drops my FPS come back.

Apologies if it's hard to make out, hard to screenshot VR in my current setup.

If you can't quite make it out, in the first SS my simulation time is at 4.4ms, update at 5.2. Hitting around 60 FPS.

2nd SS, simulation time 6.2, update either 13.4 or 15.4. Gets down to 30 FPS. (edited)

For reference, my FPS in 2.7 was mostly dictated by my simulation ms. Once it got up to around 10, I'd drop from 60 to 40 FPS. When it hovered around 5-7 I'd maintain 60. The update ms was always consistently around 6-7, so this is a big change in behavior. (edited)

 

Edited October 30, 2022 by MoleUK

[markturner1960]

Interesting! How do you get those stats showing? 

[crazyave]

I have a 3090, i7@5Ghz, 32MB RAM, HP G2.  I saw a ~2ms GPU frame time increase with my existing settings/missions.  I fly with a friend that has a 3080ti and we both saw an increase in VRAM usage of about 1-2GB.  While I have a lot of headroom with VRAM, he does not and so he saw some huge impacts to performance and some missions became unplayable, >50% loss in FPS.  Everything is still playable for me though.  Certain missions and conditions do crash to desktop now though, where they didn't before.... weird stuff like if I destroy a certain train in one my of missions I get a CTD, but doesn't happen with any of the other trains?!?!? 

It seems to me the variance in impact to performance depends on if your settings are close to a threshold like VRAM usage.  If you step over that line, you get a gigantic impact.  If you happen to have some headroom you might not see as much, like in my case.  I maximized my settings for high FPS, whereas my buddy maximized for quality at a certain minimum acceptable FPS.  I only saw a slight drop in FPS, but he became unplayable because he exceeded his VRAM requirement for his settings.

All this heat wasted whining, what a waste.  A wise company would find a way to harness this energy and put it to good use.  

[MoleUK]

4 minutes ago, markturner1960 said:

Interesting! How do you get those stats showing? 

Hit ctrl + pause a couple times. On some keyboards it will be scroll lock instead of pause.

2 minutes ago, crazyave said:

I have a 3090, i7@5Ghz, 32MB RAM, HP G2.  I saw a ~2ms GPU frame time increase with my existing settings/missions.  I fly with a friend that has a 3080ti and we both saw an increase in VRAM usage of about 1-2GB.  While I have a lot of headroom with VRAM, he does not and so he saw some huge impacts to performance and some missions became unplayable, >50% loss in FPS.  Everything is still playable for me though.  Certain missions and conditions do crash to desktop now though, where they didn't before.... weird stuff like if I destroy a certain train in one my of missions I get a CTD, but doesn't happen with any of the other trains?!?!? 

It seems to me the variance in impact to performance depends on if your settings are close to a threshold like VRAM usage.  If you step over that line, you get a gigantic impact.  If you happen to have some headroom you might not see as much, like in my case.  I maximized my settings for high FPS, whereas my buddy maximized for quality at a certain minimum acceptable FPS.  I only saw a slight drop in FPS, but he became unplayable because he exceeded his VRAM requirement for his settings.

All this heat wasted whining, what a waste.  A wise company would find a way to harness this energy and put it to good use.  

Oof, hope that's not the case. I'm stuck with 8 gigs of vram on my 3070 for the forseeable future.

[WipeUout]

20 minutes ago, crazyave said:

It seems to me the variance in impact to performance depends on if your settings are close to a threshold like VRAM usage.

Seems consistent with my 7.5% FPS drop.  My HMD is at 72hz and my CPU frametime was around 11ms with a bit of headroom prior to 2.8.  Now I am exceeding the limit of 13.7ms quite often hence the drop.  CPU not a problem, still bellow 10ms in my test track but higher than before a bit. 

[imacken]

I know I'm the odd one out here, but on my Vive Pro 2 with SteamVR, I am seeing no change in frame times at all with 2.8.  At times, I see improvements.  Obviously, no Open Composite here.

[BMGZ06]

I lost more than 10 fps in the MP server I play a lot compared to 2.7 and even lowering shadows to off in cockpit and flat terrain then lowering clouds from standard to low and view distance from ultra to high does nothing to improve it. I then created a new savedgames folder hoping it would clean things up but no joy. I also was running 2x MSAA before and in the MP server got a really constant 40-45with few dips to low 30's. Even turning MSAA off has not helped at all. 

[exil]

51 minutes ago, crazyave said:

I have a 3090, i7@5Ghz, 32MB RAM, HP G2.  I saw a ~2ms GPU frame time increase with my existing settings/missions.  I fly with a friend that has a 3080ti and we both saw an increase in VRAM usage of about 1-2GB.  While I have a lot of headroom with VRAM, he does not and so he saw some huge impacts to performance and some missions became unplayable, >50% loss in FPS.  Everything is still playable for me though.  Certain missions and conditions do crash to desktop now though, where they didn't before.... weird stuff like if I destroy a certain train in one my of missions I get a CTD, but doesn't happen with any of the other trains?!?!? 

It seems to me the variance in impact to performance depends on if your settings are close to a threshold like VRAM usage.  If you step over that line, you get a gigantic impact.  If you happen to have some headroom you might not see as much, like in my case.  I maximized my settings for high FPS, whereas my buddy maximized for quality at a certain minimum acceptable FPS.  I only saw a slight drop in FPS, but he became unplayable because he exceeded his VRAM requirement for his settings.

All this heat wasted whining, what a waste.  A wise company would find a way to harness this energy and put it to good use.  

Yeah, there must be something wrong which affects individual systems. 

I was on the lucky side 'only' losing about 10% of Gpu performance while CPU has increased about the same amount. 

Is it just placebo on my side or does the cockpit sharpness increase? At least that's what I am seeing actually. Besides the shimmering clouds, visually it's a great patch if the performance thing wouldn't be there. 

[markturner1960]

2 hours ago, MoleUK said:

Hit ctrl + pause a couple times. On some keyboards it will be scroll lock instead of pause.

Oof, hope that's not the case. I'm stuck with 8 gigs of vram on my 3070 for the forseeable future.

Thanks, can someone explain what the text signifies that appears when you hover the mouse over parts of the graph? I get "Name, frame , simulation and present" In the upper graph, I see FPS & Then "T" - what is T? 

[Blinde]

On 10/29/2022 at 12:06 AM, dutchili said:

Performance improved with the Oculus Quest 2. Higher settings and stable FPS

I lost 15fps with Quest 2 compared to 2.7

[zildac]

I lost 15fps with Quest 2 compared to 2.7

You're lucky I lost 25 with a Varjo Aero! lol

[Slammin]

I guess I'm the lucky one.  As soon as Apache was available on the non-Beta, I switched and never looked back.

I am a little dismayed by some folks screaming bloody murder at the devs over this though.  If you are running the beta so that you can provide useful feedback to the devs, hats off to ya.  But some of the responses here sound like people are consulting their attorney or something!

[imacken]

I have both stable and OB installed at all times, which makes life a lot easier at times!

[Slammin]

Me too.  I try to keep my controller configs in sync.  I did just try the beta and am seeing the same as most are reporting here as well.

[mbucchia]

11 hours ago, exil said:

Thank you. I am already that far, but I don't know how to read out that csv file. I am able to open it in excel, but there ain't the values I'm looking for. 

With fpsVR I was able to record let's say a 60 seconds track which would output avg cpu and avg gpu frametimes recorded over the 60 seconds track. 

Can you send me one of those csv files?

Are you leaving recording on for the whole test?

 

Edit: It looks like I misunderstood. You are seeing values in csv file? You can do the average yourself, or even plot them. That's the whole point of having all the data. You can do what you want with it. It takes 10s to do an average in Excel.

Edit 2: but really tho, average is a terrible metric to look at. You can have really good looking average values but horrible lows or spikes. Average literally looses information. Don't look at average. Use Excel to plot the values.

Edited October 31, 2022 by mbucchia

[lj0076]

Hi, I am confirming the issues like others...
AMD 5800x, 64MB RAM, RTX 3080, HP G2... Open XR set, Open composite
With the update to 2.8 performance loss, on average 25-30%, the same test mission, in some from 90 fps to 60, on a map like Syria the drop is much more noticeable.
Even on Friday before the update, I was thinking how everything runs like butter, high performance, beautifully sharp, MSAA 2x, shadows and visibility medium... and it's gone.
I'm getting very erratic performance, I tried switching to Steam VR and it just resulted in one crash after another and a blue screen in VR.
Unfortunately, even when I went back to stable 2.7, the problem remained and I can't get the original performance with the same settings.
Huge disappointment...

[nachomaga]

3 hours ago, lj0076 said:

Unfortunately, even when I went back to stable 2.7, the problem remained and I can't get the original performance with the same settings.
Huge disappointment...

Really? I'm thinking to going back to 2.7, now I have doubts

By the way, I remember years ago that was a terrible VR update and Nevada maps loss 50% of performance (again, in VR). I think ED fix was more or less quick, if I remember correclty.

The problem this time is the perfomance loss is not only in VR. Hope ED will release a fix soon.

[exil]

7 hours ago, mbucchia said:

Can you send me one of those csv files?

Are you leaving recording on for the whole test?

 

Edit: It looks like I misunderstood. You are seeing values in csv file? You can do the average yourself, or even plot them. That's the whole point of having all the data. You can do what you want with it. It takes 10s to do an average in Excel.

Edit 2: but really tho, average is a terrible metric to look at. You can have really good looking average values but horrible lows or spikes. Average literally looses information. Don't look at average. Use Excel to plot the values.

 

Thanks again! 

Figured it out now. I was simply to stupid to import the csv into excel in a correct manner. 

I've got everything I need now and will roll back to 2.7 for some benchmarks.