SYN_Vander BENCHMARK v6 to measure IL-2 performance in monitor & VR

[chiliwili69]

It seems there is a parameter of SteamVR called "maxRecommendedResolution" which is fixed for every VR headset.

 

The Pimax user with large resolution had to bypass that:

 

https://community.openmr.com/t/default-vrsettings-maxrecommendedresolution-steamvr/21330?page=2

 

In my steamvr.vrsettings file located at c:\program files(x86)\Steam\config\steamvr.vrsettings  my parameter is 3240. So I believe this is the cap.

The point is that I think SteamVR overwrite this value everytime it is started.

So that limit is 2921x3246x2=19 million pixels. So it means the VRtest3 would not be possible without changing the file settings.

Edited October 22, 2022 by chiliwili69

[MilitantPotato]

18 minutes ago, chiliwili69 said:

It seems there is a parameter of SteamVR called "maxRecommendedResolution" which is fixed for every VR headset.

 

That's super disappointing. Does using OpenXR not bypass this? I was getting similar results on both.

 

I tried finding a setting in the game that would increase framerate, but the only time i get high GPU usage and framerates in the benchmark is when there's almost no smoke on screen (as it pivots around the rail yard, or when it drops to a lower level of detail in the distance.) 

I was able to turn most settings to their max values and get the exact same score, but at higher GPU usage and power draw, which means something in-game seems to be causing issues with the render pipeline, i think.  I'm unsure how hard it would be, but a benchmark run with no smoke would be an interesting thing to try, since i seem to get significantly higher GPU usage and framerates when there's no smoke on screen (though it could be some other asset.)


Here's the same run as before, but with a few settings turned up:
Frames: 3393 - Time: 60000ms - Avg: 56.550 - Min: 48 - Max: 77

 

 

[MilitantPotato]

Hey, I get 55fps and very low GPU usage when looking specifically at this spot.  If I look anywhere else I get 85-90FPS and much higher GPU usage in this portion of the benchmark.  
There's definitely *something* here causing issues.

@chiliwili69 Is there any way to get free cam here so I can try and narrow down what in this SS is causing issues with FPS and low GPU usage?

Edited October 22, 2022 by MilitantPotato

[MilitantPotato]

I flew there on a quick mission, with the same settings and similar weather, got 85-90fps instead of 50-55. It's either the smoke, or maybe the train causing issues, but I believe the smoke chosen for the benchmark is more likely. 

EDIT:

After LOTS of testing and struggling with the mission editor, it seems to be the AI causing a CPU bind. After I deleted every plane and vehicle I could without breaking the replay, i almost completely stopped having CPU bind issues and was maxing out my GPU. Maybe a separate VR benchmark with less (no) AI would be more useful, as VR1 and 2 already show how well a CPU handles AI for single player VR.

Edited October 22, 2022 by MilitantPotato
Updating post.

[DBCOOPER011]

Below is a quick VR2 test in openxr w/o NIS/FSR, with a stock 12900K, stock gigabyte 4090 OC and DDR5 at XMP (6400Mhz). I fought with this thing for a day since I was getting severe stutters and the computer was shutting off sporadically. Turned out it was a holographic shell conflict in win11. shut that off and this thing runs sweet! I'm able to add much more MSAA and other eye candy and maintain 90fps even in multiplayer. Kind of strange that it maintains higher frames in multiplayer with a stock 12900k, versus when I had it overclocked with my 3090. This is strictly openxr as I havent tried it out yet in steam yet...

 

2022-10-22 15:13:26 - Il-2
Frames: 4045 - Time: 60000ms - Avg: 67.417 - Min: 52 - Max: 91

 

Edit- reran the test after ensuring the holographic shell program was off which is below. Apparently, once it's shut off in computer management it will reenable once the computer is rebooted...

 

2022-10-22 15:52:01 - Il-2
Frames: 4367 - Time: 60000ms - Avg: 72.783 - Min: 59 - Max: 91

 

Edited October 22, 2022 by DBCOOPER011
Incorrect info

[RAAF492SQNOz_Steve]

23 minutes ago, DBCOOPER011 said:

Below is a quick VR2 test in openxr w/o NIS/FSR, with a stock 12900K, stock gigabyte 4090 OC and DDR5 at XMP (6400Mhz). I fought with this thing for a day since I was getting severe stutters and the computer was shutting off sporadically. Turned out it was a holographic shell conflict in win11. shut that off and this thing runs sweet! I'm able to add much more MSAA and other eye candy and maintain 90fps even in multiplayer. Kind of strange that it maintains higher frames in multiplayer with a stock 12900k, versus when I had it overclocked with my 3090. This is strictly openxr as I havent tried it out yet in steam yet...

 

2022-10-22 15:13:26 - Il-2
Frames: 4045 - Time: 60000ms - Avg: 67.417 - Min: 52 - Max: 91

I look at your Avg fps result of 67.417 and start to wonder if changes to the game have had a massive negative impact on the benchmark we are using.

From memory, you were getting much higher average fps in test results posted some time ago.

 

Looks like I will have to re-do the benchmark and see if I get the same issue. 

[MilitantPotato]

Something drastic has changed to make everyone CPU bound with local AI present, I think.

I'm not skilled enough to remove all but the plane being chased from the benchmark file, but if someone is willing, it'd be a nice test.

[DBCOOPER011]

5 hours ago, RAAF492SQNOz_Steve said:

I look at your Avg fps result of 67.417 and start to wonder if changes to the game have had a massive negative impact on the benchmark we are using.

From memory, you were getting much higher average fps in test results posted some time ago.

 

Looks like I will have to re-do the benchmark and see if I get the same issue. 

 

Yea, I dont know. Was the benchmark changed? I just followed the instructions..

 

All I know is that this GPU is a beast. I'm getting a lot more steady fps then what I was getting with my 3090. I took a video of one of the missions that I believe was limited by the CPU, and ran the mission at 90fps. It dipped down to 89 at times but was pretty consistent. I was using 2x MSAA and could use 4X but it would drop into the mid 80's for fps. I could never use any MSAA with the 3090 and it would dip down at times. This is with stock 4090 and stock 12900k. The video is heavily compressed BTW since thats what took to get it on a free server....

 

https://dai.ly/k4hxkHZjNty97oyq83a

https://imgur.com/02cVWK3

Edited October 23, 2022 by DBCOOPER011
https://dai.ly/k4hxkHZjNty97oyq83a

[chiliwili69]

14 hours ago, MilitantPotato said:

I tried finding a setting in the game that would increase framerate, but the only time i get high GPU usage and framerates in the benchmark is when there's almost no smoke on screen (as it pivots around the rail yard, or when it drops to a lower level of detail in the distance.) 

I was able to turn most settings to their max values and get the exact same score, but at higher GPU usage and power draw, which means something in-game seems to be causing issues with the render pipeline, i think.  I'm unsure how hard it would be, but a benchmark run with no smoke would be an interesting thing to try, since i seem to get significantly higher GPU usage and framerates when there's no smoke on screen (though it could be some other asset.)

 

When runing in VR  and not achieving the 90fps, means that either GPU or CPU or both are giving frametimes above the threshold of 11.1 ms (1000ms/90Hz).

If the CPU or the GPU use more than 11.1 ms to calculate the frame then it means that for that specific second you will be below 90fps.

 

If you (with a CPU:5800X and GPU:4090) want to maximize your framrate and also improve your visual quality you will need to adjust the settings of IL-2 by:

1.- Lowering the settings which load the CPU

2.- Maxing out the settings which load the GPU.

 

The items which load the CPU were analyzed here:

 

The items which load the GPU are Clouds and MSAA. Maybe we could maxout this two items for the VRtest3 in order to stress to the max the GPU.

 

BTW, what is the "maxRecommendedResolution" parameter in your steamvr.settings file? (for your G2)

12 hours ago, MilitantPotato said:

After LOTS of testing and struggling with the mission editor, it seems to be the AI causing a CPU bind. After I deleted every plane and vehicle I could without breaking the replay, i almost completely stopped having CPU bind issues and was maxing out my GPU. Maybe a separate VR benchmark with less (no) AI would be more useful, as VR1 and 2 already show how well a CPU handles AI for single player VR.

 

Yes, the main bottleneck in the becnhmark was AI objects. And that was made on purpose.

We agreed with SYN_VANDER to put a good amount of AI objects (planes and ground) to simulate a moderate dense scenario, so can analyze which are the best CPUs to tackle that problem, in both monitor and VR. 

The CPU test in monitor (BTW, you can run it just to measure the CPU only) has not a cap, so it can go as high as we have better CPUs.

The GPU test is done in 4K, with settings which only load the GPU but not the CPU. So we can measure how good is a GPU in monitor.

So, if one combine the best CPU in IL-2 and the best GPU in IL-2  then it could achieve the best VR performance.

The VRtest1 was designed to load only CPU

The VRtest2 is the same but with 19.5Mpixels, so it is a mix of both

In this new VRtest3, (since there is a cap in maxSS%) we could do as VRtest2 but adding Extreme clouds and MSAAx4).

10 hours ago, DBCOOPER011 said:

Frames: 4367 - Time: 60000ms - Avg: 72.783 - Min: 59 - Max: 91

 

OK, thanks for this.

 

Could you also run the CPUtest and VRtest1 benchs, so we can use this info for the table?

 

Also, could you also tell us how we could design a VRtest3?  same than VRtest2 but with Clouds: Extreme and MSAAx4? or MSAAx8? will the 4090 be 100%loaded then?

10 hours ago, RAAF492SQNOz_Steve said:

Looks like I will have to re-do the benchmark and see if I get the same issue

 

The SYNVANDER mission is the same, but I changed the instructions to increase the settings of teh VR tests. (since we reached 90).

It is explained in this post from yesterday.

Edited October 23, 2022 by chiliwili69

[MilitantPotato]

It may be a day or two but I'll do a few runs with different msaa settings and extreme clouds. I'll also try a run or two at very high SS levels with steamvr and those settings to see if the SS cap applies to games too.  Hopefully it pushes the 4090 under 50fps to avoid the cpu bind.

[MilitantPotato]

@chiliwili69This will be a two part post. The first, confirming the SteamVR 100% super sample cap that you discovered, the second showing that the only setting that can make the benchmark use the 4090 to 100% is 8x MSAA with mostly maxed settings.  I can do more fine tuning of settings and try to reduce CPU overhead if you feel it's needed (presets, shadows, etc.) 

My take away is if the VR3 is meant to test the differences in video cards and not be another CPU test, a different benchmark or the same with no AI except the followed aircraft is needed. 

Here's the settings I used for every test, the only change between each run is the MSAA level: 

 

 

Here is 100% Super Sampling vs 150% super sampling for the G2, I'll also include the GPU and frametime graph from each run. These are run at 8XMSAA to ensure only the GPU is being loaded.

 

This is the key used in each graph, to save screen space:

 

 

100% (3176x3096)

Frames: 2666 - Time: 60000ms - Avg: 44.433 - Min: 39 - Max: 46

 

150% SS that SteamVR claims is 3888x3792
Frames: 2640 - Time: 60000ms - Avg: 44.000 - Min: 39 - Max: 46

 

There is zero difference between the runs, showing a hard limit of ~3200 pixels per eye.

 

 

-------------------------------------------------------------------

 

Finally, here are the runs testing 2, 4, and 8x MSAA.  The only thing of note here is that until 8xMSAA I'm almost fully CPU bound for the tests. For the last two runs my GPU usage is never 100%, a clear indication of CPU bottlenecking.

 

8X MSAA Using the settings in the first picture of the post.

 

100% 3176x3096 Super sampling for all the following tests:

Frames: 2666 - Time: 60000ms - Avg: 44.433 - Min: 39 - Max: 46

 

4X MSAA, 100% super sampling:

Frames: 3075 - Time: 60000ms - Avg: 51.250 - Min: 44 - Max: 64

Please note the GPU usage is much lower showing a clear CPU bottleneck.

 

 

MSAAx2 100% super sampling:

Frames: 3206 - Time: 60000ms - Avg: 53.433 - Min: 46 - Max: 70

Finally, I'd just like to restate my concern that the current benchmark will continue being a mostly CPU based test for many more generations of CPUs, as a near 100% increase over a 5800x would be needed to keep the CPU bottleneck of 50-52FPS from affecting tests meant for GPUs. If VR3 isn't meant to be a GPU test, then that's fine, but for folks wanting to see if $2000 for a video card is worth while over their current card, they'll just need to find data elsewhere and hope it carries to IL-2 I guess.

 

13 hours ago, chiliwili69 said:

The items which load the CPU were analyzed here:

To try and be as thorough as possible, i also lowered the settings using the super helpful guide you posted.

Here's the framerates and graphs with the settings that affect the CPU lowered to their minimum.  While my results are *much* higher, my GPU usage is still very low, showing a clear CPU bottle neck still.

Here's the settings I used for this final run:

 

Here are the results:

Frames: 4520 - Time: 60000ms - Avg: 75.333 - Min: 58 - Max: 91

 

Edited October 23, 2022 by MilitantPotato

[MilitantPotato]

One point of note in all these graphs, if under a CPU bottleneck FPSVR reports a higher GPU frametime, since it has no real way of knowing the theoretical output if the card wasn't being held back.  This can make some runs seem like the GPU frametimes are unusually high, which they are, but not from GPU limitations, and can be contrasted with GPU usage, which if low is a problem.

[MilitantPotato]

If anyone is using SteamVR and fpsVR, you can make FPS display your actual resolution and not just what steamvr claims it's set to.

 

Edited October 24, 2022 by MilitantPotato

[chiliwili69]

21 hours ago, MilitantPotato said:

There is zero difference between the runs, showing a hard limit of ~3200 pixels per eye.

 

Very good test, this proof the theory of the cap. Thanks!

21 hours ago, MilitantPotato said:

Finally, I'd just like to restate my concern that the current benchmark will continue being a mostly CPU based test for many more generations of CPUs, as a near 100% increase over a 5800x would be needed to keep the CPU bottleneck of 50-52FPS from affecting tests meant for GPUs

 

For the VRtest1, using the settings specified in the instructions, I obtain around 62fps with a 5600X. I believe you with your 5800X should achive at least that.

You can run the CPU test, the 4K test (if you have a 4K monitor o 4K TV), then VRtest1 and Vrtest2. So we can compare.

21 hours ago, MilitantPotato said:

If VR3 isn't meant to be a GPU test, then that's fine, but for folks wanting to see if $2000 for a video card is worth while over their current card, they'll just need to find data elsewhere and hope it carries to IL-2 I guess.

 

VRTest3 is not meant to be a GPU benchmark (There are already many GPU benchs out there), but just how VR would run with very high resolution devices (Pimax8KX, Aero, etc) or with the maxout graphics (clouds extreme, MSAAx8)

21 hours ago, MilitantPotato said:

the settings that affect the CPU lowered to their minimum.  While my results are *much* higher, my GPU usage is still very low, showing a clear CPU bottle neck still.

Yes, you are right here. Perhaps reducing the horizon draw distant to the lowest and putting off the HDR, SSAO, sharpen you could be closer to 90fps.

 

19 hours ago, MilitantPotato said:

If anyone is using SteamVR and fpsVR, you can make FPS display your actual resolution

Hey!! this is a good discovery and confirmation of the cap!!

 

Since you discovered this, could you please repost this in the VR section to people with different headsets can report what are their own limits (just to know if it is 3240 for all headsets).

In principle the Index and the Reverb both has the maximun at 3240.

[MilitantPotato]

I'll run your suggested settings and benchmarks as soon as i have time, thanks for responding.

 

As a side note i can confirm openxr has no resolution limits, which lets you run at very high super sampling levels. Unfortunately the motion smoothing is very poor and causes far more artifacts compared to steamvrs implementation. Ignoring the hassle of getting it working, I'm unimpressed with it's usability and performance.

Edited October 25, 2022 by MilitantPotato

[=SFG=BoostedStig]

UPDATE: Changed from 3080Ti to 4090

 

System Specs

Motherboard:  ASUS TUF Gaming X570 Plus (Wi-Fi)
CPU:                AMD Ryzen 7 5800X3D
CPU Freq:        4500mhz
Cores:              8
Threads:          16
RAM size:        32Gb (4x8GB)
RAM Freq:       3800 MHz
RAM timings:  16-16-16-35 1T
GPU:                Gigabyte 4090 OC

 

Windows 10

IL-2 v5.002

 

GPU Test:

Frames: 10709 - Time: 60000ms - Avg: 178.483 - Min: 156 - Max: 217

Frames: 10706 - Time: 60000ms - Avg: 178.433 - Min: 154 - Max: 214

 

[chiliwili69]

9 hours ago, =SFG=BoostedStig said:

Avg: 178.483 - Min: 156 - Max: 217

 

WOW! this 4090 is really a monster card.  You increased +34fps in avg and +50fps in the min!  I don´t know if you touched the CPU limit during this 4K test.

 

Would you be able to run the other tests (CPU and VR with new settings).

Thanks

[=SFG=BoostedStig]

17 hours ago, chiliwili69 said:

 

WOW! this 4090 is really a monster card.  You increased +34fps in avg and +50fps in the min!  I don´t know if you touched the CPU limit during this 4K test.

 

Would you be able to run the other tests (CPU and VR with new settings).

Thanks

 

yeah give me a few I'll give it a crack and report back

[=SFG=BoostedStig]

Alright so I re-ran all the tests just so they're all done on the same day back to back to back.

Results were consistent between passes, so I'm only posting one result per test.

Something definitely changed in this most recent build and/or test settings as my VR results are somehow worse on a 4090 than they were on the 3090. Have we figured out an exact reason for that yet?

 

Please Note: My result from yesterday was added to the spreadsheet, but it incorrectly lists my GPU at a 3090. This is a 4090. That result has been beaten by this one anyway.

 

System Specs

Motherboard:  ASUS TUF Gaming X570 Plus (Wi-Fi)
CPU:                AMD Ryzen 7 5800X3D
CPU Freq:        4500mhz
Cores:              8
Threads:          16
RAM size:        32Gb (4x8GB)
RAM Freq:       3800 MHz
RAM timings:  16-16-16-35 1T
GPU:                Gigabyte 4090 OC

 

Windows 10

IL-2 v5.002

 

VR Tests done with an HP Reverb G2 using OpenComposite

OpenXR Runtime Version: 112.2209.3002

 

CPU Test:

Frames: 7840 - Time: 60000ms - Avg: 130.667 - Min: 113 - Max: 180

 

GPU Test 4K:

Frames: 11142 - Time: 60000ms - Avg: 185.700 - Min: 157 - Max: 231

 

VR Test1:

Frames: 4279 - Time: 60000ms - Avg: 71.317 - Min: 58 - Max: 91

 

VR Test2:

Frames: 4131 - Time: 60000ms - Avg: 68.850 - Min: 55 - Max: 91

 

New VR Test3:

Frames: 2904 - Time: 60000ms - Avg: 48.400 - Min: 43 - Max: 58

Edited October 27, 2022 by =SFG=BoostedStig
Edited for incorrect GPU 4K test result shared

[1Sascha]

On 10/23/2022 at 1:52 AM, MilitantPotato said:

Something drastic has changed to make everyone CPU bound with local AI present, I think.

As mentioned in another thread: After going non-VR for a change in offline career mode, this seems to be an issue for me there as well. Especially on two of the newer maps (Kuban and especially Normandy), but I *think* the main factor is AI and mostly bomber AI. Flying multiple career-mode missions, I was getting considerable slowdowns/FPS-drops especially in missions with large flights of bombers present. OTOH: "Free hunt" type of career-missions on Rheinland with only ~16 fighters in the air still ran very smoothly.

 

In non-VR this isn't really noticeable unless you're monitoring your FPS - the game will still run smoothly on my system, but the drop is quite pronounced and I assume that this is what makes these missions a pain in VR. In non-VR I'm getting 140 to 185 FPS with very high settings and in 1440p (Ultra-preset, MSAA 4x, high clouds, etc), but if there's a flight of 111s/B-25s/B-26s on the map I'll get drops into the high 80s to low 90s. Which is still fine gameplay-wise but a pretty significant drop in performance.

 

I'm no programmer, but it seems to me that bombers will probably require more "AI-brainpower" (multiple engines, gunners) so I suspect the AI might at least be part of the problem here. I did monitor my GPU's performance and while it was maxed out nicely, it didn't run into its power-limit or any other limits. Strangely though, I also didn't observe any obviously alarming results monitoring my CPU's performance - it didn't look overly taxed usage-/power-draw-wise and temps were surprisingly low.

 

S.

[chiliwili69]

6 hours ago, =SFG=BoostedStig said:

Something definitely changed in this most recent build and/or test settings as my VR results are somehow worse on a 4090 than they were on the 3090. Have we figured out an exact reason for that yet?

 

Thank you very much for running this full set of test. They bring a lot of information.

 

First at all, the lower values you are having now in VR is because the settings of the test has been changed (to load more CPUs and have values below 90, so we can measure differences with others CPUs).

Because of that your 5800X3D is the only bottleneck in the VRtest1.

In the VRTest2, you fps decrease a very little bit, so your GPU perfectly handle the 19.5 Mpixels.

In the VRtest3, we load the GPU to the max and then it became also a bottleneck, so fps are reduced by -20fps.

 

I need to run my VRtest3, so you will see how low a 3080 goes.

Edited October 27, 2022 by chiliwili69

[chiliwili69]

I could also run my VRTest3, using the same settings than Vrtest2 but with clouds extreme and MSAAx8:

Frames: 2281 - Time: 60000ms - Avg: 38.017 - Min: 29 - Max: 46
Frames: 2274 - Time: 60000ms - Avg: 37.900 - Min: 29 - Max: 46

 

So, here my humble 3080 is having a hard time , I didn´t get out the 45fps zone!

[DBCOOPER011]

Greetings,

 

Below is what I get with my daily driver setup. Getting a 13700K next week, will see how that performs..

 

Motherboard: EVGA Z690 Dark
CPU:         12900K
CPU Freq:    55x2,54x3,53x4,52x5,51x6,50x8,Ecores-40
Cores:       8P+8E
Threads:     24
RAM size:    32Gb (2x16GB)
RAM Freq:    6930 MHz
NB Freq:     Auto 800-4700Mhz (ring auto)
RAM timings: 32-39-39-80 CR1
GPU:         Gigabyte 4090 OC (stock)
HMD: Reverb G2

Windows 11
IL-2 v5.002
Opencomposite-OpenXR Runtime Version: 112.2209.3002
SteamVR Beta 1.24.6

 

CPU Test:
Frames: 8301 - Time: 60000ms - Avg: 138.350 - Min: 121 - Max: 185

VR Test 1:
(Opencomposite) Frames: 4683 - Time: 60000ms - Avg: 78.050 - Min: 62 - Max: 91
(SteamVR) Frames: 4456 - Time: 60000ms - Avg: 74.267 - Min: 61 - Max: 91

 

VR Test 2:
(Opencomposite) Frames: 4447 - Time: 60000ms - Avg: 74.117 - Min: 60 - Max: 91
(SteamVR) Frames: 4179 - Time: 60000ms - Avg: 69.650 - Min: 58 - Max: 91

 

VR Test 3:
(Opencomposite) Frames: 2876 - Time: 60000ms - Avg: 47.933 - Min: 44 - Max: 55
(SteamVR) Frames: 2745 - Time: 60000ms - Avg: 45.750 - Min: 43 - Max: 52

 

[MilitantPotato]

Motherboard: X570 AORUS MASTER
 CPU:                 Ryzen 5 5800X
 CPU Freq:        4.9 Ghz
 Cores:               8 
 Threads:           16 
 RAM size:        32Gb (2x16GB)
 RAM Freq:       3600 MHz 
 NB Freq:          1800 MHz 
 RAM timings:  14-15-15-35
 GPU:                 Gigabye Gaming OC 4090
 HMD: Reverb G2

CPU Test
Frames: 7223 - Time: 60000ms - Avg: 120.383 - Min: 101 - Max: 166

4K Test
Frames: 9857 - Time: 60000ms - Avg: 164.283 - Min: 137 - Max: 205

VR1
Frames: 3390 - Time: 60000ms - Avg: 56.500 - Min: 49 - Max: 79

VR2
Frames: 3285 - Time: 60000ms - Avg: 54.750 - Min: 47 - Max: 75

VR3
Frames: 2691 - Time: 60000ms - Avg: 44.850 - Min: 42 - Max: 46

Looks like the new VR3 settings really push modern GPUs, cool to see the average framerates between myself and cooper's newer system so close.

Edited November 16, 2022 by MilitantPotato
4K results added

[DBCOOPER011]

Greetings,

 

Took out my 12900K and put it in my backup 3090 system. The only difference was the motherboard and DDR5 frequency, since the MSI board is 4 dimm and cant do the same frequency. The 3090 is on a waterblock, everything else is pretty much the same settings . Pretty much just a few fps gain going from a 3090 to 4090 in the IL2 test with this cpu..

 

Motherboard: MSI Z690-A
CPU:               12900K
CPU Freq:       55x2,54x3,53x4,52x5,51x6,50x8,Ecores-40
Cores:             8P+8E
Threads:         24
RAM size:       32Gb (2x16GB)
RAM Freq:      6000 MHz
NB Freq:        Auto 800-4700Mhz (ring auto)
RAM timings: 32-38-38-80 CR2
GPU:              EVGA 3090 FTW (Stock settings)
HMD:             Reverb G2

Windows 11
IL-2 v5.002
Opencomposite-OpenXR Runtime Version: 112.2209.3002

 

CPU Test:
Frames: 8397 - Time: 60000ms - Avg: 139.950 - Min: 124 - Max: 192

 

VR Test 1:
Frames: 4571 - Time: 60000ms - Avg: 76.183 - Min: 60 - Max: 91

 

VR Test 2:
Frames: 4318 - Time: 60000ms - Avg: 71.967 - Min: 56 - Max: 91

 

VR Test 3:
Frames: 2686 - Time: 60000ms - Avg: 44.767 - Min: 36 - Max: 51

Edited October 29, 2022 by DBCOOPER011

[MilitantPotato]

Hmm, I wonder if 8x msaa is just saturating the memory bus.

[Voyager]

I suppose the question ends up being, is the 4090 strong enough to lock 120fps at 3k*3k*2 VR at ultra with extreme clouds, any mirrors and a sufficiently fast CPU? 

[DBCOOPER011]

As an update, I put this 13700KF in the system and ran a few benchmarks. Everything is stock except I have two cores at 5800Mhz and allcore at 5500Mhz.

 

CPU Test:
Frames: 8794 - Time: 60000ms - Avg: 146.567 - Min: 129 - Max: 199

 

VR Test 2:
Frames: 4761 - Time: 60000ms - Avg: 79.350 - Min: 60 - Max: 91

 

VR Test 3
Frames: 3141 - Time: 60000ms - Avg: 52.350 - Min: 44 - Max: 58

 

[MilitantPotato]

21 hours ago, Voyager said:

I suppose the question ends up being, is the 4090 strong enough to lock 120fps at 3k*3k*2 VR at ultra with extreme clouds, any mirrors and a sufficiently fast CPU? 

3200x 90fps high clouds and shadows, ultra preset, 2x msaa, low mirrors is 70-85% usage, id guess not unless you dropped msaa. At this rate we need another 5 years of cpu development for 90fps too

3 hours ago, DBCOOPER011 said:

As an update, I put this 13700KF in the system and ran a few benchmarks. Everything is stock except I have two cores at 5800Mhz and allcore at 5500Mhz.

 

CPU Test:
Frames: 8794 - Time: 60000ms - Avg: 146.567 - Min: 129 - Max: 199

 

VR Test 2:
Frames: 4761 - Time: 60000ms - Avg: 79.350 - Min: 60 - Max: 91

 

VR Test 3
Frames: 3141 - Time: 60000ms - Avg: 52.350 - Min: 44 - Max: 58

 

Is this using open composite with/without openxr or steamvr?

Edited October 31, 2022 by MilitantPotato

[DBCOOPER011]

That was with openxr. Been playing around with the project lasso app and got some slight gains moving my background apps to my E-cores and putting Il2 on a higher priority. I believe that the win11 schedular is not as optimized for IL2 as I thought it was before. Still learning how to operate this process lasso app, but this is what I got just recently with the same settings..

 

CPU Test:
Frames: 9001 - Time: 60000ms - Avg: 150.017 - Min: 135 - Max: 199

VR Test 2:
Frames: 4845 - Time: 55750ms - Avg: 86.906 - Min: 54 - Max: 91

VR Test 3:
Frames: 3359 - Time: 60000ms - Avg: 55.983 - Min: 47 - Max: 62

 

Edited November 1, 2022 by DBCOOPER011

[MilitantPotato]

@chiliwili69have you thought about recommending OCAT instead of fraps? It's as simple to use (nearly) but gives much more detailed results, is open source, and free.

 

https://gpuopen.com/ocat/

[chiliwili69]

On 10/28/2022 at 7:08 PM, DBCOOPER011 said:

(Opencomposite) Frames: 4683 - Time: 60000ms - Avg: 78.050 - Min: 62 - Max: 91
(SteamVR) Frames: 4456 - Time: 60000ms - Avg: 74.267 - Min: 61 - Max: 91

Thank you for these comparative tests, it tell a lot. We can clearly see the advantage of OpenXR with the G2.

On 10/28/2022 at 10:02 PM, MilitantPotato said:

Frames: 7223 - Time: 60000ms - Avg: 120.383 - Min: 101 - Max: 166

Thanks for running these test. So your CPU and VR numbers are as expected. CPU is main bottleneck in VR tests.

On 10/29/2022 at 2:49 AM, DBCOOPER011 said:

EVGA 3090 FTW (Stock settings)

Thank you again for this intersting test.

Since most of VR tests are CPU limited the difference between 3090 and 4090 is small.

[-3RW-ghyslain]

5 hours ago, DBCOOPER011 said:

That was with openxr. Been playing around with the project lasso app and got some slight gains moving my background apps to my E-cores and putting Il2 on a higher priority. I believe that the win11 schedular is not as optimized for IL2 as I thought it was before. Still learning how to operate this process lasso app, but this is what I got just recently with the same settings..

 

CPU Test:
Frames: 9001 - Time: 60000ms - Avg: 150.017 - Min: 135 - Max: 199

VR Test 2:
Frames: 4845 - Time: 55750ms - Avg: 86.906 - Min: 54 - Max: 91

VR Test 3:
Frames: 3359 - Time: 60000ms - Avg: 55.983 - Min: 47 - Max: 62

 

 

Damn impressive! I'm about to upgrade from an R9 5900x // RTX 3080ti to an i9 13900kf // RTX 4090. Looking forward to hopefully be able to render that Pimax 8KX at full native resolution 75hz at last. Worse comes to worst, I'll settle for a "normal" FOV to reduce the pixel count by 20% or so.

Edited November 1, 2022 by -3RW-ghyslain

[chiliwili69]

On 10/29/2022 at 5:50 PM, MilitantPotato said:

I wonder if 8x msaa is just saturating the memory bus.

 

In principle the number of pixels handled in VRtest2 and VRtest3 is the same, so the required bus bandwidth is the same.

I think that Clouds to extreme also contribute to constraint the CPU more and then MSAAx8 is a huge load to the GPU

On 10/30/2022 at 9:04 PM, Voyager said:

sufficiently fast CPU?

 no CPU is going to give you 120 fps, unless you decrese to the minimum the graphics settings.

I think the use case of a 4090 for IL-2, given the current CPU constraint, is for VR devices with very high resolution and no resolution cap in SteamVR (Aero, Pimax8KX, perhaps VP2) or G2 with OpenXR if no resolution cap. So, the SS% can be increased as desired with no penalty in fps.

18 hours ago, DBCOOPER011 said:

I put this 13700KF

Many thanks. The first Intel 13th gen tested here. It set a new record.

Did you run with openXR?  was EVGA Dark PC with the 4090?

6 hours ago, DBCOOPER011 said:

vg: 86.906 - Min: 54 - Max: 91

Wow, that´s already pretty close to 90.

Edited November 1, 2022 by chiliwili69

[chiliwili69]

6 hours ago, MilitantPotato said:

ave you thought about recommending OCAT instead of fraps?

 

Hey!, thank you for this suggestion.

I didn´t know OCAT. I have just downloaded and tested it a bit.

It is simple, open source and easy to use for capture.

It can easily provide avg frametimes which are much more valuable to measure VR performance (no cap at 90, no discontinuity at 45, etc).

I will analyze a bit more and probably I will change the instructions to replace Fraps by this OCAT tool.

The tool is also good to compare performance of several runs in a easy way, so it is simpler to evaluate change in settings.

If anyboday else has any suggestion, please let me know.

[102nd-YU-cmirko]

maybe  the dev's read this thread and find out about the rest of the radeon developer tools ?

[MilitantPotato]

9 hours ago, chiliwili69 said:

 

With very high resolution and no resolution cap in SteamVR (Aero, Pimax8KX, perhaps VP2) or G2 with OpenXR if no resolution cap

I spoke with the dev of OpenXR toolkit and the resolution under the system tab of openxr tool tips F2 menu is true super sampling. You can run resolutions as high as you want until it crashes from running out of vram.

 

I still find Openxr has extremely poor motion vectoring, with tons of stutter and distortion. Currently trying to find the cause with the dev, but we've found no reason for it yet.

 

 

Also, if you open startup.cfg it tells you what vr resolution il2 started with last usage under the graphics key for or_height and or_width. Seems like another way to check actual resolution for folks not running steamvr

Edited November 1, 2022 by MilitantPotato

[chiliwili69]

On 11/1/2022 at 11:05 AM, chiliwili69 said:

I will analyze a bit more and probably I will change the instructions to replace Fraps by this OCAT tool.

 

I have been analyzing the use of OCAT instead of Fraps.

 

For the CPU and GPU test it will not offer any significant improvement since the fps in both are unlimited, so both tools will do OK. And OCAT was introducing some spikes which where not present in Fraps.

 

For the VR tests, I have been experimenting with OCAT and I think it is not yet a fully polished tool. I have some issues with the NVIDIA control panel and also the values recorded by OCAT were different from the ones recorded by Fraps.

 

Since the original purpose of the SYN_VAnder benchmark was the CPU test, also GPU test if 4K available, but the VR tests were somehow marginal since they depend on the specfic VR device and software (WMR, Pitools, HTCVive, etc).

 

Basically the purpose of this benchmark is to check new CPUs/RAMs with the CPU test. And compare results peers to peers.

[DBCOOPER011]

Greetings,

 

Did a little testing with the "CPU Test" in regards to the win11 power plan settings in project lasso. The biggest gain I noticed was moving the active processes in win11 to the E-cores, disabling hyperthreading, and keeping the setting on "Probalance" in project lasso. During testing, my processer was maxing out 1 thread in the #2 core, and halfway maxing the #8 core. I do not believe win11 is quite sufficiently optimized at this time for single thread performance in comparison to project lasso for IL2. I think performance increase is possible in win11 if properly tweaked, but for me it was easier to do it with project lasso..

 

I obtained an approximate 9FPS increase in going from the stock microsoft balanced plan to the project lasso tweaked plan. When I increased the frequency/LLC and vcore with the cpu the benefits were as not as significant as I would have hoped for daily use.

 

Cores:58x2,57x4,56x6,58x8. E-cores/Ring stock adaptive voltage 1.288V Most Vdroop:

Frames: 8484 - Time: 60000ms - Avg: 141.400 - Min: 118 - Max: 197 MS Balanced PP

 

Frames: 8539 - Time: 60000ms - Avg: 142.317 - Min: 125 - Max: 195 MS High Performance PP

 

Frames: 8563 - Time: 60000ms - Avg: 142.717 - Min: 124 - Max: 202 Project Lasso Bitsum Highest Performance PP, Probalance

 

Frames: 8897 - Time: 60000ms - Avg: 148.283 - Min: 129 - Max: 204 Project lasso Bitsum Plan Highest, Hyperthreading disabled, Instance Balancer enabled, Probalance enabled

 

Frames: 9019 - Time: 60000ms - Avg: 150.317 - Min: 135 - Max: 198 Project lasso Bitsum Plan Highest, Hyperthreading disabled, Active processes moved to E-cores, Probalance enabled

 

Cores:59x2,58x4,57x6,57x8. E-cores/Ring stock adaptive voltage 1.300V Standard Vdroop:

Frames: 9232 - Time: 60000ms - Avg: 153.867 - Min: 133 - Max: 207 59x,58x4,57x6,56x8 cores- Project lasso Bitsum Highest, HT disabled, Active processes moved to E-cores, Probalance enabled

 

Edited November 7, 2022 by DBCOOPER011

[DBCOOPER011]

As a follow up after further testing, I obtained the folowing results below. It seems that IL2 doesnt like C-States on at all FPS wise for the CPU test. I also obtained better results with a "Per Core" overclock rather then the core ratio overclock that I was using before. Its strange that I also lost FPS on the VR test 3, still trying to figure that out..

 

CPU test:

Frames: 9486 - Time: 60000ms - Avg: 158.100 - Min: 138 - Max: 220 Per Core OC (2 best cores x59, rest x55)+55mv best cores, C-States off
Frames: 9422 - Time: 60000ms - Avg: 157.033 - Min: 137 - Max: 230 Per Core OC (2 best cores x58, rest x55)+40mv best cores,C-States off
Frames: 9285 - Time: 60000ms - Avg: 154.750 - Min: 134 - Max: 214 Core Ratio OC (59x2,57x4,56x6,55x8) C-States off

 

VR Test 2:
Frames: 5316 - Time: 60000ms - Avg: 88.600 - Min: 83 - Max: 92

 

VR Test 3:
Frames: 3219 - Time: 60000ms - Avg: 53.650 - Min: 45 - Max: 60