Benchmark for CPU/RAM performance: Remagen 4.002 to 4.005

[chiliwili69]

5 minutes ago, Gomoto said:

CPU: 3600x RAM: 16 GB 3200Mhz GPU: GTX 1080 Ti

 

2019-12-23 21:50:07 - Il-2
Frames: 23560 - Time: 180000ms - Avg: 130.889 - Min: 88 - Max: 232

 

Thank you for your test.

 

Could you also provide the Mobo, CPU Freq, RAM CL, and STMark as indicated in the instructions?

 

I suppose you have run it in version 4.003. Right?

 

[SvAF/F16_Goblin]

Windows 10 Home build 1909 version 18363.535

Nvidia driver 441.66

Steam VR Beta
Windows MR for Steam Beta

 

Motherboard: Gigabyte Aorus Z370 Ultra Gaming
 CPU:                 i7-8700K HT turned off. 6 cores active
 CPU Freq:        5.0 Ghz
 L3 cache:        12 MB
 RAM type:        DDR4 Dual channel
 RAM size:        32 GB (4x8)

 NB Freq:           4500 MHz
 RAM Freq:        3200 MHz
 RAM Latency: 14
 GPU:                 1080Ti Aorus OC GPU Clock 1725 Mem Clock 11110
 STMark:           3020 Single thread

 

1. Test performed in VR with Reverb. at 90Hz SS:100

Fraps result:
2019-12-24 09:19:24 - Il-2
Frames: 11801 - Time: 180000ms - Avg: 65.561 - Min: 47 - Max: 91

 

2. Test performed in normal mode on monitor

Fraps result:

2019-12-24 09:40:04 - Il-2
Frames: 24608 - Time: 180000ms - Avg: 136.711 - Min: 95 - Max: 228

 

Edited December 24, 2019 by Goblin

[JG1_Greif]

On 12/23/2019 at 9:54 AM, chiliwili69 said:

I don´t know how to do a two variable regression with GoogleDocs (I have other multivariable tools for that but not GoogleDocs). But before introducing a new variable we first need to see a clear influence as an independent variable. So I created another graph in the results table to visually see the tendency.

We can see some tendency, but it is also true that better PCs (equipped with better CPUs and more OC) has better true latency.

 

I really enjoy this thread and study of the numbers, so I thought I'd add my 0.02 as I thought I could help out on running a regression using multiple variables using SAS.

My objective was to see if it is possible to improve the results of the univariate regression in order to get a better impression of the factors that influence average fps. I used the 15 data lines from the Remagen 4.0 tab in the IL2 VR online sheet.xlsx, I skipped the line from Gomoto because of the missing data. I used a stepwise selection for the variable selection with a criterion of significance of 0.05 for adding or removing variables. Because I could reasonably expect the graphics card to have an influence on the average fps, I added a column that represents the Passmark score per GPU from https://www.videocardbenchmark.net/high_end_gpus.html and added that variable to the model.

 

This is the Passmark data translation I used:

GPU	Passmark
2080Ti	16694
2080	15502
2070	14502
1080Ti	14221
1080	12457
1060	9097

 

I ended up with this model:

 

Parameter Estimates
Variable	Label	DF	Parameter	Standard	t Value	Pr > |t|
			Estimate	Error
Intercept	Intercept	1	-77.4523	15.9487	-4.86	0.0005
CPU GHz	CPU GHz	1	22.0593	4.5072	4.89	0.0005
RAM MHz	MHz	1	0.01709	0.0026	6.68	<.0001
Passmark	Passmark	1	0.0037	0.0008	4.76	0.0006

 

In other words: avg fps = -77.45233 + (22.05925 * CPU GHz) + (0.0037 * Passmark) + (0.01709 + RAM MHz)

This results in a very comfortable R^2 of >96%

In order to get an idea of the relative importance per variable, I ran a model using standardized data. This yields a model with the same R^2, but the size of the estimated parameter expresses the relative importance per variable:

Parameter	DF	Estimate	Standard Error	t Value	Pr > |t|
Intercept	1	132.433333	1.276271	103.77	<.0001
CPU GHz	1	8.818447	1.801808	4.89	0.0005
RAM MHz	1	10.529343	1.575338	6.68	<.0001
Passmark	1	7.537976	1.582208	4.76	0.0006

 

I notice that the model favors including the variable of the raw RAM MHz instead of the true latency (which, btw, are obviously correlated) which I found surprising. Also, it suggests that the relative importance of RAM MHz is more important than CPU GHz (though they do not necessarily come at the same cost. Forcing the model to use true latency instead of RAM MHz yielded a model  with a lower performance that does use true latency, suggesting that raw RAM MHz is more important than true latency (R^2 = 0.87, which goes against my intuition).

 

Please share your thoughts on these results. I will be able to run alternative models based on your suggestions and/or when new data will be added tot the data set. However, this will be after Christmas. Happy holidays everyone!

 

 

Edited December 30, 2019 by JG1_G_Greif

[chiliwili69]

19 hours ago, Gomoto said:

Thanks to the systematic benchmarking I moved my installation from my second NVMe drive (Corsair Force MP 510) to my first NVMe drive (Samsung 970 EVO Plus 1TB). While this did not improve my FPS, it removed on the other hand 3 repeatable little sutters during the benchmark replay (For example everytime under the bridge a short FPS hang on the second drive is now gone completly on the NVMe in the first slot. This seems to indicate some stutters are related to loading stuff during play.)

 

Double thanks. Still don´t know your RAM CL and CPU clock freq.

 

On VR, the clear bottleneck is the GPU as shown in the nice fpsVR histogram

 

Regarding the NVMe drive, that´s is surprising. I thought drives would not affect stutters. 

 

[Gomoto]

CPU freq. is volatile 3,8 base 4,4 boost (4,2 to 4,3 most of the time)

RAM freq. is 3200 

[chiliwili69]

8 hours ago, Goblin said:

1. Test performed in VR with Reverb. at 90Hz SS:100

 

Thanks for your tests. Regarding SS100%, What resolution per eye do you have? Do you run 2WMR for SteamVR" or "WMR for SteamVR Beta"?

4 minutes ago, Gomoto said:

CPU freq. is volatile 3,8 base 4,4 boost (4,2 to 4,3 most of the time)

RAM freq. is 3200

 

Thanks, but I was referring to the RAM CAS latency reported by CPU-Z. Look at first post.

[Gomoto]

RAM 1596.8 Mhz * 2 = 3200 Mhz ?

[chiliwili69]

3 hours ago, JG1_G_Greif said:

running a regression using multiple variables using SAS

 

WOW! many thanks for this multivariable regression.

I have created a column in the table with your regression.

 

I think that we are still missing more data from the users to get a more solid based conclusions.

 

The Ryzens CPUs frequencies could change really the regression, thats why it could be more reasonable to include the STMark.

 

Regarding the GPU, I wanted to create a track were GPUs in the line of 2080Ti to 1080 were not bottlenecked, so we could concentrate efforts in determining the CPU/RAM factors.

The problems is that the more capable CPU/RAMs are also equipped with 2080Tis, so this could give wrong statistical conclusions. Again, we need more data.

19 minutes ago, Gomoto said:

RAM 1596.8 Mhz * 2 = 3200 Mhz

 

I mean the CAS latency:

[Gomoto]

CAS Latency 16

[A_radek]

Chili may I suggest also collecting user NB frequency data? Visible under memorytab in cpu-z. Did some very limited testing and would need someone to verify - but back then it had almost as much of an FPS impact as ram frequency.
 

On 12/23/2019 at 9:54 AM, chiliwili69 said:

We can see some tendency, but it is also true that better PCs (equipped with better CPUs and more OC) has better true latency.

Certainly newer architecture will make a difference but wonder how much of it actually is the omitted NB frequency, that's been raised with each intel generation.

Edited December 24, 2019 by SvAF/F16_radek

[Alonzo]

4 hours ago, SvAF/F16_radek said:

Chili may I suggest also collecting user NB frequency data? Visible under memorytab in cpu-z. Did some very limited testing and would need someone to verify - but back then it had almost as much of an FPS impact as ram frequency.
 

Certainly newer architecture will make a difference but wonder how much of it actually is the omitted NB frequency, that's been raised with each intel generation.

 

I certainly think there is something happening with IL2 and something I would call "general memory latency". It seems to like CPU frequency as well as fast access to memory locations (a combination of fast Mhz RAM, low CAS latency, and high speed northbridge / ring ratio).

 

I have an old spreadsheet that I was using for benchmarking. My rig isn't stable at very high ring ratios, and I was changing a lot of settings, but I found two runs where I only changed ring ratio:

Ring ratio 47     Avg FPS 73.775  (old Chili VR benchmark, perfect score would be 90)
Ring ratio 49     Avg FPS 76.267

Does anyone have access to one of those Xeon-ish CPUs with a very large on-chip cache? This behavior looks like IL2 main memory accesses are a big factor in performance, maybe one of those mega cache chips can fit more of the working set into L3 cache.

[SvAF/F16_Goblin]

@chiliwili69 Resolution per eye as reported in Steam VR is 2201x2156 at 100% running Steam VR Beta and Windows MR for Steam Beta.

I think You have mixed the numbers for my VR result in the spreadsheet by the way

Edited December 25, 2019 by Goblin

[chiliwili69]

12 hours ago, SvAF/F16_radek said:

Chili may I suggest also collecting user NB frequency data? Visible under memorytab in cpu-z

 

Thanks for the suggest. I have modified the OP and added a column to the table.

I really don´t know what NB freq is (even after googling it for some minutes) but if you think this could hide the saint grial we need to gather all potential data which influence IL-2.

If fact, you were the man who discovered the RAM influence with the old Balapan track.

 

Please, anyone who has already reported a test, could you please post (or PM me) your NB freq? Thanks.

[dburne]

5 minutes ago, chiliwili69 said:

 

Thanks for the suggest. I have modified the OP and added a column to the table.

I really don´t know what NB freq is (even after googling it for some minutes) but if you think this could hide the saint grial we need to gather all potential data which influence IL-2.

If fact, you were the man who discovered the RAM influence with the old Balapan track.

 

Please, anyone who has already reported a test, could you please post (or PM me) your NB freq? Thanks.

 

According to AIDA 64, my North Bridge frequency is 4701.1 MHz.

[chiliwili69]

I have re-run the test with the 4.003 version and it seems I got about 3 fps less than before:

 

4.002  freq 4.8GHz
2019-12-09 10:41:50 - Il-2
Frames: 22909 - Time: 180000ms - Avg: 127.272 - Min: 86 - Max: 215

2019-12-09 10:45:53 - Il-2
Frames: 22894 - Time: 180000ms - Avg: 127.189 - Min: 85 - Max: 215

2019-12-09 10:49:58 - Il-2
Frames: 22879 - Time: 180000ms - Avg: 127.106 - Min: 85 - Max: 213

4.003 freq 4.8GHz
2019-12-22 18:11:17 - Il-2
Frames: 22502 - Time: 180000ms - Avg: 125.011 - Min: 83 - Max: 214

2019-12-22 18:15:45 - Il-2
Frames: 22267 - Time: 180000ms - Avg: 123.706 - Min: 82 - Max: 210

2019-12-22 18:19:44 - Il-2
Frames: 22470 - Time: 180000ms - Avg: 124.833 - Min: 84 - Max: 209

2019-12-25 12:00:26 - Il-2
Frames: 22362 - Time: 180000ms - Avg: 124.233 - Min: 83 - Max: 213

2019-12-25 12:04:10 - Il-2
Frames: 22328 - Time: 180000ms - Avg: 124.044 - Min: 85 - Max: 213

2019-12-25 12:07:56 - Il-2
Frames: 22360 - Time: 180000ms - Avg: 124.222 - Min: 84 - Max: 211

 

Maybe I will separate the tests by version. Is anyone of the previous tests also obtaining less or just me?

[A_radek]

Thanks Alonzo for posting that. Results rather inline with mine. This was also one of chilis older vr benchmarks with 90fps as a goal.

 

Northbridge @4200 (default)

Frames, Time (ms), Min, Max, Avg
 7284,    120000,  44,  78, 60.704

Northbridge @4600

Frames, Time (ms), Min, Max, Avg
 7766,    120000,  43,  82, 64.717

Thanks Chili, no holy grail but perhaps another small piece of the puzzle. Following your documenting with interest.

[SvAF/F16_Goblin]

I will redo the test with higher NB frequency as soon as possible.
As of now I'm stresstesting with NB at 4.6 GHz and will take it higher to see any difference.

 

I can say already that with 4.6 instead of 4.5 i got 3026 on single thread passmark which is a small increase from 3020.

Edited December 25, 2019 by Goblin

[SvAF/F16_Goblin]

Well, going from 4500MHz to 4700MHz on the NB didn't render any revelations on my configuration.

But on a side note I have no stutters or micro freeze and ghosting is minimal at least that's how I perceive it.
I usually fly with slightly higher settings in VR. In VR the GPU is the culprit I believe, at least on my system as it stands of now.

 

Monitor results a very minor increase:
2019-12-25 14:40:53 - Il-2
Frames: 24803 - Time: 180000ms - Avg: 137.794 - Min: 94 - Max: 232

 

VR results no increase at all

2019-12-25 15:39:39 - Il-2
Frames: 12164 - Time: 180000ms - Avg: 67.578 - Min: 49 - Max: 91

 

Edited December 25, 2019 by Goblin

[chiliwili69]

5 hours ago, Goblin said:

VR results no increase at all

 

Thanks. Well.. you gained 2 fps.. and it is free

[71st_AH_statuskuo]

5 hours ago, Goblin said:

VR results no increase at all

2019-12-25 15:39:39 - Il-2
Frames: 12164 - Time: 180000ms - Avg: 67.578 - Min: 49 - Max: 91

 

Goblin - I have almost the same setup as you with the Reverb, only a 1080 instead of a 1080Ti, and 16GB of RAM.  Yet you're getting 15 more FPS in the VR test, assuming you followed Chili's settings.  I also had the SS@100% (native Reverb resolution) and got:

 

frames: 9372 - Time: 180000ms - Avg: 52.067 - Min: 42 - Max: 91
 

I'd be surprised going from a 1080 to 1080Ti buys you that many more FPS, and the RAM size shouldn't be a factor.   The monitor results seem more in line with my 132 avg fps to your 138 fps.

 

Do you have fpsVR to see what latencies you're getting?

 

Chili - my Nb is 43

Edited December 25, 2019 by 71st_AH_statuskuo

[Alonzo]

8 hours ago, chiliwili69 said:

I really don´t know what NB freq is (even after googling it for some minutes) but if you think this could hide the saint grial we need to gather all potential data which influence IL-2.

 

Please, anyone who has already reported a test, could you please post (or PM me) your NB freq? Thanks.

 

North Bridge is an Intel name, on newer chips it's called "ring ratio" or sometimes "uncore ratio". It's the speed for the non-CPU-core portions of the CPU, such as some of the on-die buses and some of the RAM interface. For chips set to auto, ring ratio is usually a few hundred megahertz lower than the CPU core speed (e.g. 4.7ghz CPU might have 4.2 or 4.3ghz ring ratio). Because IL2 loves fast memory, the ring ratio can have an effect. It may even be that IL2 is doing some kind of locking, and the chip cache speed is affected by ring ratio.

 

44 minutes ago, chiliwili69 said:

Thanks. Well.. you gained 2 fps.. and it is free

 

It's free as long as your system is still stable. If you increase ring ratio it's like anything else -- sometimes the associated voltages need to be increased too. It's worth doing an overnight Prime95 AVX "large FFT" stress test to really check the memory is still working.

 

For my benchmark results, north bridge (= ring ratio) was 4.7ghz.

[SvAF/F16_Goblin]

@71st_AH_statuskuo

I have fpsVR but I've never tried it ? Will test and see what latencies I get in IL-2. I have everything set up as required in procedures.

The increase in NB frequency might not get much fps gain but it feels smoother. I also fly with 150 km horizon draw and high clouds normally.

One interesting thing that I noticed was that decreasing the resolution in IL-2 to 800x600 in the settings actually caused less fps in the benchmark!

And with almost 5 fps as well.

 

Thus I've created a custom resolution in Nvidia panel for the reverb 2160x2160 that I set in IL-2 to test. I will also test whether full screen or not makes a difference.

 

Increasing NB frequency to 4.7 gave a reduction in passmark single thread from 3026 to 3022.

 

@Alonzo
I use OCCT to test stability and it is stable with the voltages i have. I also have done the delid mod to my i7-8700K and have 70 degrees on the CPU with 100% load over time.
The cooler I have is the kraken x52 AIO water cooler.

Edited December 26, 2019 by Goblin

[SvAF/F16_Goblin]

@71st_AH_statuskuo

 

fpsVR Report:
App: Il-2 HMD: HP Reverb VR Headset VR1000-2xxx0 (90.001 Hz, IPD 63.2)
GPU: NVIDIA GeForce GTX 1080 Ti (26.21.14.4166) CPU: Intel(R) Core(TM) i7-8700K CPU @ 5.0  GHz
Delivered fps: 62.77  Duration: 3.6min.
GPU Frametimes:
Median: 13.4 ms
99th percentile: 18.5 ms
99.9th percentile: 26.2 ms
frametime <11.1ms(vsync): 22.8%
CPU frametime:
Median: 10.2 ms
99th percentile: 25.2 ms
99.9th percentile: >30 ms
frametime <11.1ms(vsync): 63.9%
Max. SteamVR SS: 100%
Render resolution per eye: 2201x2156(by SteamVR settings, Max.) (HMD driver recommended: 2201x2156)

 

 

Edited December 26, 2019 by Goblin

[71st_AH_statuskuo]

Goblin - Thanks for doing fpsVR, your GPU latencies  are  quite a bit better (5 ms), and even the CPU is better (2ms) even though we have the same STmark score.  So maybe the I do need to upgrade to a 1080Ti, or more likely a 2080S.   Can you please post your NVIDIA settings.  Below were my fpsVR stats.

 

fpsVR Report:
App: Il-2 HMD: HP Reverb VR Headset VR1000-2xxx (90.0009994506836 Hz, IPD 61.0)
GPU: NVIDIA GeForce GTX 1080 (26.21.14.3630) CPU: Intel(R) Core(TM) i5-9600K CPU @ 3.70GHz
Delivered fps: 35.4  Duration: 3.6min.
GPU Frametimes:
Median: 18.1 ms
99th percentile: 23.4 ms
99.9th percentile: >30 ms
frametime <11.1ms(vsync): 30.1%
CPU frametime:
Median: 12.4 ms
99th percentile: >30 ms
99.9th percentile: >30 ms
frametime <11.1ms(vsync): 31.5%
Max. SteamVR SS: 192%
Render resolution per eye: 2228x2178(by SteamVR settings, Max.) (HMD driver recommended: 1608x1572)

[Alonzo]

8 hours ago, 71st_AH_statuskuo said:

Goblin - Thanks for doing fpsVR, your GPU latencies  are  quite a bit better (5 ms), and even the CPU is better (2ms) even though we have the same STmark score.  So maybe the I do need to upgrade to a 1080Ti, or more likely a 2080S.

 

App: Il-2 HMD: HP Reverb VR Headset VR1000-2xxx (90.0009994506836 Hz, IPD 61.0)
GPU: NVIDIA GeForce GTX 1080 (26.21.14.3630) CPU: Intel(R) Core(TM) i5-9600K CPU @ 3.70GHz
Delivered fps: 35.4  Duration: 3.6min.
GPU Frametimes:
Median: 18.1 ms
99th percentile: 23.4 ms
99.9th percentile: >30 ms
[...]

Max. SteamVR SS: 192%
Render resolution per eye: 2228x2178(by SteamVR settings, Max.) (HMD driver recommended: 1608x1572)

 

I think with the 1080, you're going to need to settle for 45 FPS interpolated on the Reverb. You might want to tune things a little so that your 99th percentile is below 22ms -- that will give you fewer drops below 45. Maybe reducing SteamVR supersample will help a bit with that.

[71st_AH_statuskuo]

5 hours ago, Alonzo said:

 

I think with the 1080, you're going to need to settle for 45 FPS interpolated on the Reverb. You might want to tune things a little so that your 99th percentile is below 22ms -- that will give you fewer drops below 45. Maybe reducing SteamVR supersample will help a bit with that.

That's the mode I'm always in and actually pretty happy with the 1080 and the performance I can get from the Reverb, just seeing if they're was something I was missing to get those extra FPS that Goblin has.   Only time I have trouble is in heavy combat while in MP servers, such as Combat Box (perhaps you've heard of that ?) 

[SvAF/F16_Goblin]

I have re-projection off in WMR for Steam Beta
Also I have “renderTargetScale” : 2.0 activated to get increased clarity OBS!! takes a hit on the fps.

 

Not to forget that both MB and GPU are the same vendor and the gaming branded ones for compatibility.

My GPU is also overclocked to 1.730 GHz and memory is also OC

 

I haven't stopped tweaking and tuning yet ?

Basically my Nvidia settings is:

Turn off G-Sync and set fixed frequency

Set max performance for power

Turn off V-sync

Pre rendered frames to 3

Desired update frequency to programmed controlled.

 

All other settings are the standard ones.

 

The names in English might not be correct here since it's my own translation from my Swedish control panel ?

 

I then change something and do a series of benchmarks and take the mean value to see if it gets better or worse.

As of now I'm experimenting with the latency mode and have it set to Ultra. There are some reports on the DCS forum that it's beneficial.

 

 

 

 

 

Edited December 27, 2019 by Goblin

[71st_AH_statuskuo]

6 hours ago, Goblin said:

I have re-projection off in WMR for Steam Beta
Also I have “renderTargetScale” : 2.0 activated to get increased clarity OBS!! takes a hit on the fps.

 

 

Thanks Goblin, I'll try these settings out.

[SvAF/F16_Goblin]

One more thing to try is to put the WMR mirror to "pause". That stops the extra game rendering in that window.

Just klick on the pause sign at the bottom of the mirror and then minimize the window itself.

 

[Gomoto]

Goblin have you benchmarked your 1080Ti overclock. I get no benefit from overclocking GPU RAM or CORE in IL2 VR, but I get a benefit from increasing Power Limit and Temp Limit to MAX (with an agressive fan profile). (I get an inital benefit from RAM overclocking but as the card gets hot it drops during reruns of the benchmark)

Edited December 28, 2019 by Gomoto

[SvAF/F16_Goblin]

@Gomoto

I have not tried that approach. I've increased power limit a bit but decreased temp limit.
But lets not hijack this benchmarking thread with overclocking discussions ?
We should ask mods for a overclocking thread perhaps?

[chiliwili69]

I have run the benchmark in VR with the Index at 100%SS SteamVR (2016x2240 per eye) and 80Hz mode, no motion smoothing enabled.

These are the results:

2019-12-29 10:19:40 - Il-2
Frames: 8060 - Time: 180000ms - Avg: 44.778 - Min: 31 - Max: 81

 

Pretty bad indeed!

I thought it was going to be around 60 fps avg since my monitor fps was around 125 fps, but it seems the half rule doesn´t apply to me.

 

I have used in a second run the fpsVR tool stats. I think it is mainly CPU limited.

 

 

I have run again the test with fpsVR, but changed the IL-2 graphics (changes in red) settings to something closer to my "usual" more relaxed settings:

 

Now the fpvVR stats improved, the CPU has is more time below 12.5ms and allows GPU to do his job more effectively:

 

 

[71st_AH_statuskuo]

4 hours ago, chiliwili69 said:

I have run the benchmark in VR with the Index at 100%SS SteamVR (2016x2240 per eye) and 80Hz mode, no motion smoothing enabled.

These are the results:

2019-12-29 10:19:40 - Il-2
Frames: 8060 - Time: 180000ms - Avg: 44.778 - Min: 31 - Max: 81

 

Pretty bad indeed!

I thought it was going to be around 60 fps avg since my monitor fps was around 125 fps, but it seems the half rule doesn´t apply to me.

 

I have used in a second run the fpsVR tool stats. I think it is mainly CPU limited.

 

 

 

 

Interesting we (Goblin, you, and I) are all at 2kx2k for SS  yet have pretty different latency distributions patterns (reposted mine below) for your original test settings.   Both your 1080ti have about the same latencies (~5ms better than mine) but your FRAPs numbers are worse than mine (52 vs 44) and much worse than Gob's.    In your fpsVR report what was your delivered fps, I noticed mine was different than what FRAPS reported. 

 

[Alonzo]

How are you both running it in VR? Same as Chili's (old) benchmarking procedure? Set the headset on a non-moving surface, use the 'centre headlook' at start of replay, don't move headset during replay / Fraps?

 

The CPU limit is very interesting. You could both try 'Low' settings, I think that's the one that has the most CPU impact. Some folks on the forums really want a 'true' 80/90 FPS and it would be really useful to know which settings impact this the most. It's a lot of work, though, to go through them all.

 

Also, I personally really dislike the distant buildings "on" setting. I get cities "popping" into view when I look at them. Is that just me, or is that what the setting does in VR? I prefer it off instead of the popping as I look towards and away from a city.

[dburne]

I see them pop in and out occasionally, whether distant is on or not.

[71st_AH_statuskuo]

I ran it like the old procedure with the HMD on the back of my chair with the cockpit on the lower 1/3 of the screen.   Used the settings as specified at the start of the post.  There is no hotkey to start the fpsVR recording so you have to time it properly otherwise you get stats from outside the track.

 

I'm pretty content with the reprojection at 45fps, just looking for improvements during MP battles.  Trying to decide whether to spend money now for the 2080s as GPU is my limiter, and how many fps I might gain over my 1080 in my Reverb setup.

[Jaws2002]

On 12/25/2019 at 7:01 AM, chiliwili69 said:

I have re-run the test with the 4.003 version and it seems I got about 3 fps less than before:

Maybe I will separate the tests by version. Is anyone of the previous tests also obtaining less or just me?

 

 

I think you are right. I got about the same drop between versions on the 3600x, but i didn't do too much testing, just one run.

 

    However, after waiting and trying to find one for a month, few days ago i got the Ryzen 3950X.  Like that was not enough, my month old motherboard blew one of the power stages and dropped dead. I found one across the border in USA and swapped it. 

 It's finally done and i rerun the test. I used just basic auto overclocking in Ryzen master. Two runs in "creator" mode, with 16 cores and 32 theeads running in paralel and four runs in "gaming" mode, that for this chip disables the weaker chiplet and runs on 8 cores and 32 threads.

 

 

Here are the results:

Creator mode 16 cores 32 threads

 

Motherboard:     Gigabyte X570 Aorus Master
CPU:                    AMD Ryzen 9 3950x
cpu freq:             4748MHz
L3 cache:           4x16 MB
Ram type:          DDR4
Ram size           32 GB
Ram speed:      3600MHz
Ram latency:   16
GPU:                 1080Ti
STMark:           3142  

 

 

 Low:               2019-12-28 14:36:34 - Il-2
                        Frames: 28370 - Time: 180000ms - Avg: 157.611 - Min: 104 - Max: 256

 High              2019-12-28 14:31:39 - Il-2
                        Frames: 24897 - Time: 180000ms - Avg: 138.317 - Min: 93 - Max: 238

 

Game mode 8 cores 16 threads

 

 

 Motherboard:     Gigabyte X570 Aorus Master
CPU:                    AMD Ryzen 9 3950x
cpu freq:             4748MHz
L3 cache:           2x16 MB
Ram type:          DDR4
Ram size           32 GB
Ram speed:      3600MHz
Ram latency:   16
GPU:                 1080Ti
STMark:           3264 

 

Low:           2019-12-29 13:55:49 - Il-2
                    Frames: 29045 - Time: 180000ms - Avg: 161.361 - Min: 107 - Max: 270

Balanced:  2019-12-29 13:50:51 - Il-2
                   Frames: 26344 - Time: 180000ms - Avg: 146.356 - Min: 101 - Max: 239

High:         2019-12-29 13:38:49 - Il-2
                  Frames: 25428 - Time: 180000ms - Avg: 141.267 - Min: 93 - Max: 238

Ultra:        2019-12-29 13:45:41 - Il-2
Frames: 24122 - Time: 180000ms - Avg: 134.011 - Min: 92 - Max: 238

 

 

As you can see, at this point, the game can't really tell the difference between my old six core/twelve threads 3600x that boosts up to 4.4Ghz and a 16 core /32 thread CPU that boosts to 4.75Ghz.  The importance of CPU in gaming is overrated. 

 

For everything else this two are a world apart. Even the single thread performance is better with the 3950x. Here are the ST mark difference between this two:

 

Ryzen3600x:

Single core:

 

Multicore:

fallout 4 list

 

Ryzen 3950x:

Single core:

 

 

 

Multicore:

 

 

 

 

 

 

 

Edited December 29, 2019 by Jaws2002

[Loki_1982b_rock1080]

Ran the test on version 4.003. I uninstalled and re-installed IL2 before running the test just to ensure I didn't forget any mods/tweaks. I've ordered some faster RAM (3600mhz CL16), should be here next week then I will re-run the tests so we can see how that influences things.

 

 Motherboard: MSI Z390-A Pro
 CPU:                 9700K
 CPU Freq:        5.0 Ghz
 L3 cache:        12 MB
 RAM type:        DDR4
 RAM size:        16 GB

 NB Freq:         4700 MHz
 RAM Freq:        3000 MHz 
 RAM Latency:     16  
 GPU:             2080 Super
 STMark:          3028

I ran 3 passes all with the same settings:

Frames: 25498 - Time: 180000ms - Avg: 141.656 - Min: 95 - Max: 238

Frames: 25340 - Time: 180000ms - Avg: 140.778 - Min: 95 - Max: 238

Frames: 25940 - Time: 180000ms - Avg: 144.111 - Min: 100 - Max: 239

 

Thanks for your work into this guys! I've been chasing the FPS rabbit for the last month with my Rift S.

[chiliwili69]

5 hours ago, 71st_AH_statuskuo said:

but your FRAPs numbers are worse than mine (52 vs 44) and much worse than Gob's.    In your fpsVR report what was your delivered fps, I noticed mine was different than what FRAPS reported

 

The avg fps reported by the fpsVR stats are 56 fps for the run with same benchmark settings and 66.5 for the one with the modified settings.

But the fps reported by fpsVR are not exactly the same than the one reported by FRAPS. FRAPs just only counts for the 3 minutes of the flight record, whereas fpsVR start to count fps from the moment I launch the game. So fpsVR counts the launching game (enter password), going to track record, launch track record, run track record and exit game. 4 minutes is total.

 

I don´t know how to synchronize fpsVR to start counting just when I press the P to run the test. And stop after 3 minutes.

 

In any case, It is strange that having your GPUs frametimes from 16 to 22 ms your reported fps by fraps is 52. Did you disabled Motion reprojection in WMR? (you have to do it to properly run the test)

5 hours ago, Alonzo said:

How are you both running it in VR? Same as Chili's (old) benchmarking procedure? Set the headset on a non-moving surface, use the 'centre headlook' at start of replay, don't move headset during replay / Fraps?

 

Yes. I didn´t describe that in the post but basically is how you described. I run the test on my head (without moving it) and pressed the recenter button.

5 hours ago, Alonzo said:

it would be really useful to know which settings impact this the most. It's a lot of work, though, to go through them all.

 

Yes, if you see in the google spreadsheet, there is a tab called GainMap. My goal is to determine the impact of every setting in just monitor (for now).

1 hour ago, Jaws2002 said:

As you can see, at this point, the game can't really tell the difference between my old six core/twelve threads 3600x that boosts up to 4.4Ghz and a 16 core /32 thread CPU that boosts to 4.75Ghz.  The importance of CPU in gaming is overrated.

 

Thank you very much for your detailed tests. We know that it might not be worth to go beyond 6 or 8 cores.

As you said, that new CPU (3950X) is a beast in other duties, but for IL-2 in monitor it will not deliver more fps than his little brother 3600X. (most likely to have same results in VR, but need to be proven).

[Jaws2002]

38 minutes ago, chiliwili69 said:

Thank you very much for your detailed tests. We know that it might not be worth to go beyond 6 or 8 cores.

As you said, that new CPU (3950X) is a beast in other duties, but for IL-2 in monitor it will not deliver more fps than his little brother 3600X. (most likely to have same results in VR, but need to be proven).

 

 

Look at the CPU and cores utilization in game:     That should tel you everything about the impact the CPU has on Game performance. 

 

 

 

 

 

Edited December 29, 2019 by Jaws2002