AMD 6900xt tuning and settings for VR in dcs. My optimal recipe.

[Topo]

On 8/1/2022 at 11:02 AM, VirusAM said:

Did you try to use older drivers?

As I wrote above, using 21.5.2 everything is smooth.

I tried the latest drivers (22.7.something) and 22.5.1 and it was nightmare (especially the newest one). Maybe other drivers will be good as well, somewhere they introduced something that broke VR.

Anyway I am going to file a proper bug report to AMD tonight, I don't know if other people did the same.  

I'm away from DCS now; from what I remember, I use an old "VR friendly" driver, probably 21.5.2 or earlier.

In fact, in my experience motion reprojection is worse in my 6900XT than in my old Nvidia 2070 Super; for this reason I preferred to disable motion reprojection altogether, setting the G2 reverb to 60 hz

 

EDIT:

I run a test with 21.5.2 driver; forced 30 hz in Opencomposite and on SteamVR, and the image looked like jelly. 

Edited August 2, 2022 by Topo

[Svsmokey]

8 hours ago, Topo said:

I'm away from DCS now; from what I remember, I use an old "VR friendly" driver, probably 21.5.2 or earlier.

In fact, in my experience motion reprojection is worse in my 6900XT than in my old Nvidia 2070 Super; for this reason I preferred to disable motion reprojection altogether, setting the G2 reverb to 60 hz

 

EDIT:

I run a test with 21.5.2 driver; forced 30 hz in Opencomposite and on SteamVR, and the image looked like jelly. 

 

I run 60 hz native on an AMD card also . Motion smoothing works seamlessly and very well when on airfields or Supercarrier .

[VirusAM]

I run 60 hz native on an AMD card also . Motion smoothing works seamlessly and very well when on airfields or Supercarrier .

Which drivers do you use?

[Svsmokey]

20 minutes ago, VirusAM said:

3 hours ago, Svsmokey said:

I run 60 hz native on an AMD card also . Motion smoothing works seamlessly and very well when on airfields or Supercarrier .

Which drivers do you use? 

 

22.3.1 . I tried 22.5.1 , but it introduced a problem , so i reverted . I do not use motion reprojection , but motion smoothing works perfectly . I am always at 60 frames in the air , 30 on the ground , and the transition is seamless . I run in Steam VR .

Edited August 2, 2022 by Svsmokey

[VirusAM]

12 hours ago, Svsmokey said:

22.3.1 . I tried 22.5.1 , but it introduced a problem , so i reverted . I do not use motion reprojection , but motion smoothing works perfectly . I am always at 60 frames in the air , 30 on the ground , and the transition is seamless . I run in Steam VR .

 

OK, thanks....but motion smoothing is just a motion reprojection technology (steamvr name for it) 

[Svsmokey]

What i'm saying is that motion smoothing works with AMD gpu's , while motion reprojection apparently does not .

6 hours ago, VirusAM said:

OK, thanks....but motion smoothing is just a motion reprojection technology (steamvr name for it) 

 

[Topo]

Just now, Svsmokey said:

What i'm saying is that motion smoothing works with AMD gpu's , while motion reprojection apparently does not .

So, AMD cards are ok for motion smoothing (SteamVR) but aren't suitable for motion reprojection with OpenXR or Oculus ASW?

In my test with an RX6900XT MR forced always on (1/2, 30 fps) in OpenXR is a jelly mess, while in SteamVR is a bit better but still unusable.

[Svsmokey]

I don't know anything about OpenXR or ASW with AMD . but yes , motion smoothing works with AMD , SteamVR and the G2 .

[VirusAM]

2 hours ago, Topo said:

So, AMD cards are ok for motion smoothing (SteamVR) but aren't suitable for motion reprojection with OpenXR or Oculus ASW?

In my test with an RX6900XT MR forced always on (1/2, 30 fps) in OpenXR is a jelly mess, while in SteamVR is a bit better but still unusable.

I use the valve index and for me it does not work (motion smoothing) with drivers newer then 21.5.2 (may 2021 drivers)

[Hartsblade]

1 hour ago, VirusAM said:

I use the valve index and for me it does not work (motion smoothing) with drivers newer then 21.5.2 (may 2021 drivers)

Have the Index as well, and motion Smoothing doesn't work fir me either, never has.

[VirusAM]

Have the Index as well, and motion Smoothing doesn't work fir me either, never has.

Thanks for sharing your experience.
I am a new AMD Radeon user, I did file a bug report, in which i described that only with 21.5.2 drivers the motion smoothing is working.
I hope every AMD and VR user is doing the same.
I also hope I won't be forced to spend more and switch to Nvidia again

[TZeer]

[Sr.]

On 8/4/2022 at 2:40 AM, TZeer said:

 22.8.1 fixes this

[alessander]

9 часов назад, Sr. сказал:

22.8.1 fixes this

I confirm. Now there are no problems with flashing G2 displays and there is a significant increase in FPS, which began with the release of 22.6.1

[Sr.]

11 hours ago, alessander said:

I confirm. Now there are no problems with flashing G2 displays and there is a significant increase in FPS, which began with the release of 22.6.1

Roger that!

Looks more incredible now.

[dooom]

hey guys - im on 22.5.1 with my 6900XT and G2.  What exactly are you refering to for the 22.8.1 drivers??  Are you refering to the 60Hz flicker in the headset or are folks reporting a general FPS increase with 22.8.1?

[Sr.]

58 minutes ago, dooom said:

hey guys - im on 22.5.1 with my 6900XT and G2.  What exactly are you refering to for the 22.8.1 drivers??  Are you refering to the 60Hz flicker in the headset or are folks reporting a general FPS increase with 22.8.1?

For me, with 22.7.1 I was getting flicker at any rate, fixed with 22.8.1. And based on what I saw last night using the OXR toolkit counter I've gained a minimum of 4-6 fps GPU side.

[Picchio]

I confirm that 22.8.1 fixed the flickering issues brought by previous versions.

On a side-note, has anyone ever measured the performance difference (and TGP values) between different levels of Range Visibility, when facing a huge city (Vegas)?

Edited August 16, 2022 by Picchio

[WipeUout]

On 8/16/2022 at 10:19 AM, Picchio said:

On a side-note, has anyone ever measured the performance difference (and TGP values) between different levels of Range Visibility, when facing a huge city (Vegas)?

Never did before but thought this could be interesting.  So I did a benchmark track file and tested at three visibility range values: Extreme, High and Low.  Here are the results:

Extreme : 47.6 FPS

High:        50.6 FPS

Low:         57.3 FPS

I was using the settings (bellow) that I use only for becnhmarking in order to push the GPU to max througout the benchmark.

And here is the track file: Benchmark City Scape - F-18 over Dubai.trk

I use a bit of supersampling (126%) with Steam VR.  I also use VR Shader mod 2.7.12.23362 to gain a few FPS and Reshade_VR_Sharpen_color to improve sharpness.  My AMD driver is the latest (22.8.1).  And finally, my 6900XT is OC'ed at max 2825 Mhz, min 2725 Mhz.

 

In order to have a common reference point to compare results, here is my test procedure:

1. Load track file up to point you are in the cockpit but do not press ready yet.

2. Take off HMD and point mouse to load fpsVR.  Be ready to put back HMD quickly in order to start track as soon as fpsVR is started.

3. Once fpvVR is started, the screen brigntness will change and this is your cue to start the replay.

4. Keep your head straight, do not move.  The track file is short, the F-18 will do one 360 degree turn only.

5. Once the turn nears the end, the plane will start leveling, at that point check fpsVR average and take not of last value before end of track.

6. Try new setting and restart test procedure.

 

As you can see with my results, there is a significant difference with visibility range values when flying over dense city.

[Picchio]

31 minutes ago, WipeUout said:

Never did before but thought this could be interesting.  So I did a benchmark track file and tested at three visibility range values: Extreme, High and Low.  Here are the results:

Extreme : 47.6 FPS

High:        50.6 FPS

Low:         57.3 FPS

I was using the settings (bellow) that I use only for becnhmarking in order to push the GPU to max througout the benchmark.

And here is the track file: Benchmark City Scape - F-18 over Dubai.trk

I use a bit of supersampling (126%) with Steam VR.  I also use VR Shader mod 2.7.12.23362 to gain a few FPS and Reshade_VR_Sharpen_color to improve sharpness.  My AMD driver is the latest (22.8.1).  And finally, my 6900XT is OC'ed at max 2825 Mhz, min 2725 Mhz.

 

In order to have a common reference point to compare results, here is my test procedure:

1. Load track file up to point you are in the cockpit but do not press ready yet.

2. Take off HMD and point mouse to load fpsVR.  Be ready to put back HMD quickly in order to start track as soon as fpsVR is started.

3. Once fpvVR is started, the screen brigntness will change and this is your cue to start the replay.

4. Keep your head straight, do not move.  The track file is short, the F-18 will do one 360 degree turn only.

5. Once the turn nears the end, the plane will start leveling, at that point check fpsVR average and take not of last value before end of track.

6. Try new setting and restart test procedure.

 

As you can see with my results, there is a significant difference with visibility range values when flying over dense city.

 

Thanks for contributing!

I'll post my own results as soon as I can - meanwhile, I would propose to try and use different settings for a comparison, so that I can verifiy my thesis a bit better (more on that later).

Could you please try these settings as well?

Shadows - High (or Ultra, if you've enabled it)

Terrain Objects Shadows - Default

MSAA - 2x

Clouds - Ultra

Then, try setting up your benchmark mission with the Scattered 6 preset (the cloudier, the better, let's see what you get with this one) and setting mission time at 16:30.

On the GPU side, more critical information that is required for now:

6900XT model and settings - MPT, Wattman, type of cooling - if not on air then our results can't be compared!

In Adrenalin, edit your DCS profile with Chill activated and set at 45/45 - essential to compare GPU readings!

HWiNFO GPU related values taken during the test: GPU Temperature, GPU Hotspot Temperature, TGP Power - critical, don't forget to note this one!

 

I'm also using a G2, mine is set at 100% SS via OpenXR, Motion Smoothing disabled.

 

In my own test will I set up my F-14 at Nellis, bearing 210 (so that Vegas is in front of me in the distance), and my measurements will be from the cockpit, looking forward.

For the moment, you may try a simple test by adjusting only the weather in your own mission (with the graphics settings above), perhaps we might find interesting elements to compare. Hornet and Tomcat might measure differently but what I'm particularly interested in are the values read from HWiNFO.

Edited August 17, 2022 by Picchio

[WipeUout]

Ok, with clouds and shadows at max, MSAA at 2x, time 16:30 and cloud set at scattered 6: here are my results, but with a new track file since clouds template can't be changed once track is saved.  I did not use chill since this is useless in a bechmark.  If you limit your FPS then how can you get an accurate FPs reading?  Also understand that the limit refresh rate of your HMD will also corrupt your FPS result if your FPS exceed the refresh rate during the bechmark. I think you might use chill because you use reprojection?  That's the case for most G2 user as you can't achieve smooth video at native refresh rate.

Very similar results but you can't directly compare with the previous results since it is a new track file: Benchmark City Scape - F-18 over Dubai V2.trk

Vis Range Extreme :  46.4

Vis Range at High :    49.4

Vis Range at Low:      54.3

 

My card is an ASUS RX 6900XT TUF Gaming OC Edition air cooled.  As for other values from HWinfo:

GPU Temp (max) 63 deg C

GPU HS Temp (max) 83 Deg C

TDP (max) 318 Watts

Reliable and repeatable testing with the F-14 is not possible as the track recording is still broken.  A very important aspect of testing is repeatability, that's why you have to use a track file and be very rigorous with your testing procedure.  If you repeat the same test twice, you should get consistently the same result +/- 0.1 FPS. 

Edited August 17, 2022 by WipeUout
New info and findings

[Picchio]

1 hour ago, WipeUout said:

Ok, with clouds and shadows at max, MSAA at 2x, time 16:30 and cloud set at scattered 6: here are my results, but with a new track file since clouds template can't be changed once track is saved.  I did not use chill since this is useless in a bechmark.  If you limit your FPS then how can you get an accurate FPs reading?  Also understand that the limit refresh rate of your HMD will also corrupt your FPS result if your FPS exceed the refresh rate during the bechmark. I think you might use chill because you use reprojection?  That's the case for most G2 user as you can't achieve smooth video at native refresh rate.

Very similar results but you can't directly compare with the previous results since it is a new track file: Benchmark City Scape - F-18 over Dubai V2.trk

Vis Range Extreme :  46.4

Vis Range at High :    49.4

Vis Range at Low:      54.3

 

My card is an ASUS RX 6900XT TUF Gaming OC Edition air cooled.  As for other values from HWinfo:

GPU Temp (max) 63 deg C

GPU HS Temp (max) 83 Deg C

TDP (max) 318 Watts

Reliable and repeatable testing with the F-14 is not possible as the track recording is still broken.  A very important aspect of testing is repeatability, that's why you have to use a track file and be very rigorous with your testing procedure.  If you repeat the same test twice, you should get consistently the same result +/- 0.1 FPS. 

 

Thanks!

I purposely asked to enable Chill so that the GPU is put under a constant load - Chill works in that direction by limiting your maximum framerate - which is not exactly the focus of the tests I'm running, I'm mainly interested in power consumption at certain settings.

Would you run the same test you did before, but with a clearer weather setting like High Scattered 1? I would appreciate it if you attached a screenshot from HWiNFO with your complete GPU readings as well. Thank you in advance!

[WipeUout]

1 hour ago, Picchio said:

I purposely asked to enable Chill so that the GPU is put under a constant load

Constant load?  I don't understand what point you are trying to make.  Constant load will simply result in lower temps and power consumption.  With the test I performed, my GPU was at 98-99% load most of the time and this is consistent with the temps/power observed.  Scene with lots of details will be more demanding to your GPU, as long as your CPU can feed the data to the GPU which is another important factor here.  Complexity also increases your CPU load which can have its frametime value exceed the GPU frametime value and result in decrease load on the GPU. 

Can you expand more about what you are trying to achieve with those tests?  It might help me to understand...  ...and motivate me to invest more time into this!

 

[Picchio]

I'll try to expand a bit more once I collect data from my own tests - I am trying to pinpoint the cause of the unusually high TGP values I've been getting during my last sessions, after I updated to 22.8.1.

My card (Reference) has an increased power limit via MPT at 325W, a voltage cap and offset and GPU clock at 2400-2500.

During the same NTTR mission (except for different weather settings and drivers), I've been reading up to 322W TGP while before it didn't reach 290W. That, and Hotspot temperatures of nearly 110°C - again, I've never seen these values in DCS before (only when benchmarking in Timespy).

UPDATE: I found the cause of the high power readings, when moving very close to the clouds and through them, that's when they happen.

I'm not quite sure that's good of the graphics engine to just squeeze the hardware that way...

Edited August 17, 2022 by Picchio

[WipeUout]

2 hours ago, Picchio said:

UPDATE: I found the cause of the high power readings, when moving very close to the clouds and through them, that's when they happen

Did you try with different cloud quality setting?  I find that clouds in VR are awful not matter what the setting, so I leave it to low and gain some FPS. 

BTW I did some testing with a very cloudy mission (Overcast 4) and had a difference of 11 FPS between Low and Ultra settings!  This setting is very demanding and should be kept at low.  At least until we get the next gen GPUs...  In November hopefully! 

Edited August 17, 2022 by WipeUout
New info and findings