Multithreading VR Benchmark 4090 + 7950x3D

[DirtyMike0330]

I haven't done testing before so I threw together what I could using FRAPS and running the same low-level path in the Persian Gulf Instant Action Free Flight. This was in the Hornet module.

On every run I turned in, hit the deck and took the same route through the initial big city. Benchmarks ran for 3 minutes each time. The attached picture has my settings. The only settings changed during testing were shadows, clouds, MSAA and will be mentioned below. 

ASUS TUF Gaming 4090oc

Ryzen 9 7950x3D

32gb Corsair DDR5 6000mhz with EXPO enabled

HP Reverb G2 Native Resolution

OpenXR rather than SteamVR

 

The first two tests were pre-update. One was before disabling SMT and the other was after disabling SMT:

Pre-Update:

SMT ON:

2023-03-10 10:59:09 - DCS
Frames: 11935 - Time: 180000ms - Avg: 66.306 - Min: 40 - Max: 80 

SMT OFF:

2023-03-10 11:10:54 - DCS
Frames: 11891 - Time: 180000ms - Avg: 66.061 - Min: 56 - Max: 91 

null

 

Post-Update SMT OFF (Multithreading Launcher):

*** For those that don't know, the 7950x3D can run games using the high frequency cores or using the cores with v-cache, so I ran tests with both.***

Frequency Cores (MT):

Settings Pictured (Med Shadows, Ultra Clouds, MSAAx2):

2023-03-10 13:06:54 - DCS
Frames: 13608 - Time: 180000ms - Avg: 75.600 - Min: 61 - Max: 91

Settings Pictured but FLAT SHADOWS:

2023-03-10 13:11:22 - DCS
Frames: 13666 - Time: 180000ms - Avg: 75.922 - Min: 60 - Max: 91 

Settings Pictured but Clouds HIGH instead of Ultra:

2023-03-10 13:15:28 - DCS
Frames: 13976 - Time: 180000ms - Avg: 77.644 - Min: 62 - Max: 91

V-Cache Cores (MT):

Settings Pictured (Med Shadows, Ultra Clouds, MSAAx2):

2023-03-10 13:22:50 - DCS
Frames: 13428 - Time: 180000ms - Avg: 74.600 - Min: 59 - Max: 91

Settings Pictured but FLAT SHADOWS:

2023-03-10 13:26:39 - DCS
Frames: 13753 - Time: 180000ms - Avg: 76.406 - Min: 60 - Max: 91

Settings Pictured but Clouds HIGH instead of Ultra:

2023-03-10 13:30:39 - DCS
Frames: 14454 - Time: 180000ms - Avg: 80.300 - Min: 64 - Max: 91

 

Random Test:

MSAA Off, All settings High, Clouds Ultra:

2023-03-10 13:35:37 - DCS
Frames: 14282 - Time: 180000ms - Avg: 79.344 - Min: 61 - Max: 91

 

I had the OpenXR Toolkit overlay on while running these. The CPU was rocking during these tests and held the 3ms range admirably. I saw a spike to 4ish in the heart of the city but it was in the 3ms range for the most post. 

Unfortunately, the GPU wasn't getting sub-10ms until I was out of the city or away from a lot of ground objects. It stayed mostly in the 12-15ms range. 

[Panthera_Tigris]

Your conclusion seems to mirror mine that the extra v-cahce doesn't help with DCS much, unlike with other flight sims. 

BTW, the game is run on the frequency cores by default for me. Even though I selected "remember this is a game" in xbox game bar. The only way to run it on v-cahce cores is by using process lasso and selecting affinity. Not that it matters since results are the same for both of us. 

[DirtyMike0330]

8 hours ago, Panthera_Tigris said:

Your conclusion seems to mirror mine that the extra v-cahce doesn't help with DCS much, unlike with other flight sims. 

BTW, the game is run on the frequency cores by default for me. Even though I selected "remember this is a game" in xbox game bar. The only way to run it on v-cahce cores is by using process lasso and selecting affinity. Not that it matters since results are the same for both of us. 

That is really odd that the game bar setting isn't working out for you! I know that I had to do it twice on one of my other games before it actually "stuck" and I could verify that game bar was seeing it as a game. 

On the other note, it certainly does seem that the CPU is no longer the problem! We can all now be GPU bound now and struggle with that  

[Ramstein]

with that bad azz of a computer, video 4090 (I looked at prices and wow!) I would have expected every video setting to be maxed. I have never gotten my VR FPS above (consistent) 40 FPS in everything I fly, DCS and IL-2. YOu are rocking it up 60 to 99 FPS. I also usually have only used Intel CPU all of these decades. No biggie, just different cpu. I hope to build a new computer soon, but had to check out your MT scores in DCS. Thanx for the info.

[DirtyMike0330]

5 minutes ago, Ramstein said:

with that bad azz of a computer, video 4090 (I looked at prices and wow!) I would have expected every video setting to be maxed. I have never gotten my VR FPS above (consistent) 40 FPS in everything I fly, DCS and IL-2. YOu are rocking it up 60 to 99 FPS. I also usually have only used Intel CPU all of these decades. No biggie, just different cpu. I hope to build a new computer soon, but had to check out your MT scores in DCS. Thanx for the info.

It is a little disappointing to not be able to hit that sweet, sweet 90 fps with maxed settings yet I can do it in IL-2 and it is a fantastic experience. Hopefully when Vulkan or some kind of DLSS in VR becomes available, that will be the last little boost people need. I really don't want to buy any new hardware for a long time if possible. 

[Flextremes]

10 hours ago, Panthera_Tigris said:

Your conclusion seems to mirror mine that the extra v-cahce doesn't help with DCS much, unlike with other flight sims. 

BTW, the game is run on the frequency cores by default for me. Even though I selected "remember this is a game" in xbox game bar. The only way to run it on v-cahce cores is by using process lasso and selecting affinity. Not that it matters since results are the same for both of us. 

This is so interesting: when I moved from 5900x to 5800x3D I saw massive improvements.

Maybe try to disable Game Bar and start using Process Lasso?  Or disable a CCD in bios?

 

Edited March 11, 2023 by Flextremes

[DirtyMike0330]

16 minutes ago, Flextremes said:

This is so interesting: when I moved from 5900x to 5800x3D I saw massive improvements.

Maybe try to disable Game Bar and start using Process Lasso?  Or disable a CCD in bios?

 

 

I am wondering if we have just hit a wall in terms of CPU performance with the 5800x3D and now the new Ryzen 9 x3D chips. I am seeing the same results as you with the 3-4ms range with high settings so maybe higher end hardware just isn't going to see as much benefit from the multithreading update as some of the lower hardware specs. I am a little disappointed that the 4090 can't get under 10ms. 

Edited March 11, 2023 by DirtyMike0330

[Flextremes]

5 minutes ago, DirtyMike0330 said:

I am wondering if we have just hit a wall in terms of CPU performance with the 5800x3D and now the new Ryzen 9 x3D chips. I am seeing the same results as you with the 4-3ms range with high settings so maybe higher end hardware just isn't going to see as much benefit from the multithreading update as some of the lower hardware specs. I am a little disappointed that the 4090 can't get under 10ms. 

Yep: I think that is a matter of fact. That is why I am embracing reprojection in openXR. My goal is to stay above 45fps. I have to say that reprojection looks way better now?!!! Clearly not perfect, but less wiggly because so much less dropping to 30fps is happening.

Currently I am running at almost maxed-out settings in DCS and a very high FSR upscaled render resolution (3880x3800). It is so damn smooth and beautiful now.

Edited March 11, 2023 by Flextremes

[Hoirtel]

19 hours ago, DirtyMike0330 said:

I haven't done testing before so I threw together what I could using FRAPS and running the same low-level path in the Persian Gulf Instant Action Free Flight. This was in the Hornet module.

On every run I turned in, hit the deck and took the same route through the initial big city. Benchmarks ran for 3 minutes each time. The attached picture has my settings. The only settings changed during testing were shadows, clouds, MSAA and will be mentioned below. 

ASUS TUF Gaming 4090oc

Ryzen 9 7950x3D

32gb Corsair DDR5 6000mhz with EXPO enabled

HP Reverb G2 Native Resolution

OpenXR rather than SteamVR

 

The first two tests were pre-update. One was before disabling SMT and the other was after disabling SMT:

Pre-Update:

SMT ON:

2023-03-10 10:59:09 - DCS
Frames: 11935 - Time: 180000ms - Avg: 66.306 - Min: 40 - Max: 80 

SMT OFF:

2023-03-10 11:10:54 - DCS
Frames: 11891 - Time: 180000ms - Avg: 66.061 - Min: 56 - Max: 91 

null

 

Post-Update SMT OFF (Multithreading Launcher):

*** For those that don't know, the 7950x3D can run games using the high frequency cores or using the cores with v-cache, so I ran tests with both.***

Frequency Cores (MT):

Settings Pictured (Med Shadows, Ultra Clouds, MSAAx2):

2023-03-10 13:06:54 - DCS
Frames: 13608 - Time: 180000ms - Avg: 75.600 - Min: 61 - Max: 91

Settings Pictured but FLAT SHADOWS:

2023-03-10 13:11:22 - DCS
Frames: 13666 - Time: 180000ms - Avg: 75.922 - Min: 60 - Max: 91 

Settings Pictured but Clouds HIGH instead of Ultra:

2023-03-10 13:15:28 - DCS
Frames: 13976 - Time: 180000ms - Avg: 77.644 - Min: 62 - Max: 91

V-Cache Cores (MT):

Settings Pictured (Med Shadows, Ultra Clouds, MSAAx2):

2023-03-10 13:22:50 - DCS
Frames: 13428 - Time: 180000ms - Avg: 74.600 - Min: 59 - Max: 91

Settings Pictured but FLAT SHADOWS:

2023-03-10 13:26:39 - DCS
Frames: 13753 - Time: 180000ms - Avg: 76.406 - Min: 60 - Max: 91

Settings Pictured but Clouds HIGH instead of Ultra:

2023-03-10 13:30:39 - DCS
Frames: 14454 - Time: 180000ms - Avg: 80.300 - Min: 64 - Max: 91

 

Random Test:

MSAA Off, All settings High, Clouds Ultra:

2023-03-10 13:35:37 - DCS
Frames: 14282 - Time: 180000ms - Avg: 79.344 - Min: 61 - Max: 91

 

I had the OpenXR Toolkit overlay on while running these. The CPU was rocking during these tests and held the 3ms range admirably. I saw a spike to 4ish in the heart of the city but it was in the 3ms range for the most post. 

Unfortunately, the GPU wasn't getting sub-10ms until I was out of the city or away from a lot of ground objects. It stayed mostly in the 12-15ms range. 

Am I right in reading this as it made no difference now to be on vcache CCD?? 

[DirtyMike0330]

6 minutes ago, Hoirtel said:

Am I right in reading this as it made no difference now to be on vcache CCD?? 

The v-cache performance is still better but only marginally so in terms of average FPS. FRAPS doesn’t allow me to benchmark things like the 1% lows so that could be a bigger improvement but overall if you’re on a 7000x3D series CPU you’ll want to use the v-cache cores 

22 minutes ago, Flextremes said:

Yep: I think that is a matter of fact. That is why I am embracing reprojection in openXR. My goal is to stay above 45fps. I have to say that reprojection looks way better now?!!! Clearly not perfect, but less wiggly because so much less dropping to 30fps is happening.

Currently I am running at almost maxed-out settings in DCS and a very high FSR upscaled render resolution (3880x3800). It is so damn smooth and beautiful now.

 

I have everything high with unlocked motion reprojection as well but never thought to try and upscale… I may have to give that a go Are you using the reverb G2 as well? If so I will copy your render resolution first. 

Edited March 11, 2023 by DirtyMike0330

[Flextremes]

15 minutes ago, DirtyMike0330 said:

The v-cache performance is still better but only marginally so in terms of average FPS. FRAPS doesn’t allow me to benchmark things like the 1% lows so that could be a bigger improvement but overall if you’re on a 7000x3D series CPU you’ll want to use the v-cache cores 

I have everything high with unlocked motion reprojection as well but never thought to try and upscale… I may have to give that a go Are you using the reverb G2 as well? If so I will copy your render resolution first. 

 

Cool,

Screenshot my openXR settings:

Edited March 11, 2023 by Flextremes

[DirtyMike0330]

26 minutes ago, Flextremes said:

Cool,

Screenshot my openXR settings:

 

Going to give these a try, thanks! 

[trevoC]

Testing a new build 7900x3D/4090 and I see little to no gains between older single threaded and multi-threaded version. (hard to say if I'm gaining a frame or two in MT or margin of error). note, I'm on G2

I also noticed that I float around specific frames a lot.... Almost like I have vsync on. It sits around 60 fps a lot. If I crank down the settings from ultra/high it will jump to 90 fps but rarely in between. Just when you think its locked it will hit mid 70's for a few seconds and then back to 60. No matter the scenario, marianas in a huey or caucus in an FC3 aircraft.

Before I would see huge fps differences between these two scenarios.

Also CPU bound according to DCS frame times and see the same fps across all 4 scenarios:

single threaded - parked cores (running on high cache cores)
multi threaded - parked cores
single threaded - no parked cores (all cores enabled)
multi threaded - no parked cores (all cores enabled)

All 4 scenarios are within 1 or 2 fps from each other. This is very strange behavior.... vcache seems to have zero effect along with multi threading.

I had no problems getting game bar to make DCS single threaded run on proper cores, but there was some strange behaviour with DCS multi threading and parked cores... It seemed to park random cores rather than the first 12.

I eventually disabled windows from parking the cores and just let all the cores run in DCS MT and that seems to be the best scenario (by 2 frames max) but I'm wondering how it will effect performance in other games (MSFS/iRacing etc...)

What I really can't decide at the moment is if I prefer maxed settings at 50/60 fps, or lesser settings at 90+ fps... oh damn that so smooth. It really is so bloody nice at 90 fps.

[some1]

Measuring FPS is really not the best way to benchmark the hardware in VR. Even if you have motion reprojection disabled, the VR runtime is still able to "lock" fps to a certain number. VR can't have tearing, so the headset display is a monitor with vsync-on always enabled. The synchronization has to happen somewhere.

In OXR, turbo = off will drop your numbers because then OpenXR governs frame timing.

And of course if you're GPU limited, then the faster CPU also won't make much difference on FPS. In VR you can be GPU limited even if GPU does not show 100% utilization, in fact that's what often happens.

The best way to check if the new CPU is working better in VR than the old one is to look at CPU frametimes reported by the runtime. If they are lower, then the CPU is better. But whether that makes a noticeable difference in FPS, that's another story.

Edited March 13, 2023 by some1

[Flextremes]

49 minutes ago, some1 said:

Measuring FPS is really not the best way to benchmark the hardware in VR. Even if you have motion reprojection disabled, the VR runtime is still able to "lock" fps to a certain number. VR can't have tearing, so the headset display is a monitor with vsync-on always enabled. The synchronization has to happen somewhere.

In OXR, turbo = off will drop your numbers because then OpenXR governs frame timing.

And of course if you're GPU limited, then the faster CPU also won't make much difference on FPS. In VR you can be GPU limited even if GPU does not show 100% utilization, in fact that's what often happens.

The best way to check if the new CPU is working better in VR than the old one is to look at CPU frametimes reported by the runtime. If they are lower, then the CPU is better. But whether that makes a noticeable difference in FPS, that's another story.

 

Excellent explanation!

[MoleUK]

On 3/11/2023 at 3:28 AM, Panthera_Tigris said:

Your conclusion seems to mirror mine that the extra v-cahce doesn't help with DCS much, unlike with other flight sims. 

BTW, the game is run on the frequency cores by default for me. Even though I selected "remember this is a game" in xbox game bar. The only way to run it on v-cahce cores is by using process lasso and selecting affinity. Not that it matters since results are the same for both of us. 

I saw huge improvements pre MT update when moving from a 5800X to a 5800X3D. The cache meant I could enable shadows without massive performance loss, so DCS does definitely utilize the cache.

However, since the MT release my performance between ST and MT exe is VERY similar atm. I was mostly GPU bottlenecked after grabbing the X3D, so that is not entirely unexpected. The main simulation/logic thread is still a big potential bottleneck, particularly in busy MP servers I think. It is however a higher bottleneck ceiling for many than before the MT release, which is good.

Edited March 13, 2023 by MoleUK

[DirtyMike0330]

3 hours ago, trevoC said:

Testing a new build 7900x3D/4090 and I see little to no gains between older single threaded and multi-threaded version. (hard to say if I'm gaining a frame or two in MT or margin of error). note, I'm on G2

I also noticed that I float around specific frames a lot.... Almost like I have vsync on. It sits around 60 fps a lot. If I crank down the settings from ultra/high it will jump to 90 fps but rarely in between. Just when you think its locked it will hit mid 70's for a few seconds and then back to 60. No matter the scenario, marianas in a huey or caucus in an FC3 aircraft.

Before I would see huge fps differences between these two scenarios.

Also CPU bound according to DCS frame times and see the same fps across all 4 scenarios:

single threaded - parked cores (running on high cache cores)
multi threaded - parked cores
single threaded - no parked cores (all cores enabled)
multi threaded - no parked cores (all cores enabled)

All 4 scenarios are within 1 or 2 fps from each other. This is very strange behavior.... vcache seems to have zero effect along with multi threading.

I had no problems getting game bar to make DCS single threaded run on proper cores, but there was some strange behaviour with DCS multi threading and parked cores... It seemed to park random cores rather than the first 12.

I eventually disabled windows from parking the cores and just let all the cores run in DCS MT and that seems to be the best scenario (by 2 frames max) but I'm wondering how it will effect performance in other games (MSFS/iRacing etc...)

What I really can't decide at the moment is if I prefer maxed settings at 50/60 fps, or lesser settings at 90+ fps... oh damn that so smooth. It really is so bloody nice at 90 fps.

Yeah, so I just fired both executables up in 2D and confirmed what you're saying. Despite both of them definitely having "this is a game" checked in the game bar, it appears the MT launcher is not parking cores. Which basically renders my whole "benchmark" moot lol

I went back to the original launcher last night anyway. We are GPU bound at this point so I'm not going to tinker with much until DLSS or Vulkan becomes a thing. 

 

[Hoirtel]

Btw, for benchmarking https://www.capframex.com/ is really good. Doesn't seem to record separate CPU/GPU frame times but does get an overall and you can add GPU load etc.

[trevoC]

3 hours ago, DirtyMike0330 said:

Yeah, so I just fired both executables up in 2D and confirmed what you're saying. Despite both of them definitely having "this is a game" checked in the game bar, it appears the MT launcher is not parking cores. Which basically renders my whole "benchmark" moot lol

I went back to the original launcher last night anyway. We are GPU bound at this point so I'm not going to tinker with much until DLSS or Vulkan becomes a thing. 

 

I've sort of come to the same place ATM.

I'm kind of tired of rebooting and OC'ing and messing with DCS.... getting 60+ fps on G2 is not a bad thing. I just find some of these things so strange it bugs me.

Will probably take a brake from it all and just play DCS for a bit until it bugs me again.

Biggest dilema now is what do I prefer.... 60 fps max settings, or 90 fps with some graphic concessions.... man 90 fps is smooth. I really underestimated how much I'd like those buttery frames.

[DirtyMike0330]

34 minutes ago, trevoC said:

I've sort of come to the same place ATM.

I'm kind of tired of rebooting and OC'ing and messing with DCS.... getting 60+ fps on G2 is not a bad thing. I just find some of these things so strange it bugs me.

Will probably take a brake from it all and just play DCS for a bit until it bugs me again.

Biggest dilema now is what do I prefer.... 60 fps max settings, or 90 fps with some graphic concessions.... man 90 fps is smooth. I really underestimated how much I'd like those buttery frames.

Are you opposed to motion reprojection? I can’t stomach anything under 90fps and, to be honest, the unlocked motion reprojection available with openxr toolkit is fantastic. I have all settings on high or ultra as well as msaa x2 and everything is smooth. I have the g2 as well and don’t have any issues with artifacts or the screens looking funny due to motion reprojection. 

[hhhttt]

MT FAQ are talk about E core and P cores, maybe DCS MT is spreading so many thread that is force non 3D core to start? Have anyone test with disable frequency core? The 60fps may be a speed limit when threads from 2 ccd are talking with each other.

In a way, you can think x3d expand cache from 32MB to 96MB, so it would improve a game if its core caluclation/simulation model can not fit in 32mb cache. MSFS definitely see gain there. Maybe with MT they optmized/removed some calculation that it spread out more, so it can fit regular 32mb cache, so x3d improvement is gone now if you are using simple plane. 

I wonder if F15E or Apache or F14 can gain better from x3d.

[trevoC]

6 hours ago, DirtyMike0330 said:

Are you opposed to motion reprojection? I can’t stomach anything under 90fps and, to be honest, the unlocked motion reprojection available with openxr toolkit is fantastic. I have all settings on high or ultra as well as msaa x2 and everything is smooth. I have the g2 as well and don’t have any issues with artifacts or the screens looking funny due to motion reprojection. 

I could never stand it (MR) personally. I do fly a lot of helos and it usually doesn't play well with the blades. I'm also not a fan of the artifacts.

That being said a lot of those were probably because of not providing it with enough frames which I can do now.

I'll give it another go once I get all other items figured out.

I'm pretty tolerant of the headset motion, its the same phenomenon that gets people sick in actual aircraft and in cars. I've built up a high tolerance over 30 years of this motion. I usually don't get ill in headsets above 20 fps. In the teens starts messing with me.

[Kreutzberg]

Based on some of the settings in this thread, I think you guys are not seeing uplift because you're in GPU bound situations. To ensure that you're checking CPU performance, you need to make sure that it is a CPU bound situation. A foolproof way to do that is set VR pixel density to 0.5. But if you're in 2D, then GPU-related settings need to be turned way down and load into a very CPU intensive map.

[trevoC]

I'll address a few things in this post... too many threads on this and don't want to cross post.

 

[DirtyMike0330]

6 hours ago, trevoC said:

I'll address a few things in this post... too many threads on this and don't want to cross post.

 

Interesting read. So if I understand correctly, you've settled on running the MT version with the 7950x3D working stock (no manual core assignments, game bar on, etc etc)? I noticed a reference to something about core parking via the registry?