Collect feedback to create a benchmark for IL-2 to measure CPU/RAM and also GPU

[chiliwili69]

I would like to share my view to create a benchmark similar to the previous Remagen track but this time using a mission created by SYN_Vander.

This has been discussed in the Ryzen Zen3 thread and also in the new GPU threads, so maybe it is better to centralize those comments here before we create an easy procedure that everyone can follow.

My ideas are:

 

1. Use the mission way since the variability is really low and it will be enough to capture gains higher than 2fps. So all data and procedure will survive future releases.

 

2. Create 2 runs with the same procedure, one to stress the CPU/RAM only and the other to stress the GPU only. So we can determine best hardware for the CPU/RAM independently of GPU and viceversa. The results will be collected in two different tables in Google Docs as before.

 

3. The test will be primarily focused to be done in monitor, but it could be also be run in VR. So the thread will be in the hardware section, not in VR as before. Everything which is good for monitor is good for VR and it is much easier to measure it in monitor (capped fps at 40-80 or 45-90, SteamVR SS or other, multiple devices with diferent FOVs and specific software, etc, etc).

 

4. The settings that I was thinking to use for the CPU/RAM test run will be this (monitor 1080p, Ultra Preset, all maxed out with no mirrors, low Clouds, no AA, no HDR, No SSAO). With these settings my 1080Ti is at 30%-40% load.

 

5. The settings that I was thinking to use for the GPU test run will be this (monitor 4K, LOW Preset, all to minimun, with Extreme Clouds, MSAAx8, no HDR, no SSAO). With these settings my 1080Ti is at 100% load during all mission.

 

6. The software that I was thinking to use to record fps is Fraps. It is very simple but it did his job in the previous tests.

 

7. As in previous test, I was thinking to run the tests with the hud/icons hidden.

 

8. For the VR test, I think we could run the CPU/RAM test putting 50% SteamVR SS (or equivalent in Oculus) and 150%SteamVR SS for the GPU run. I have tested the mission in VR and it works well (although capped to 40 or 80).

 

9. I don´t think it would be necessary to run an external benchmark (like Passmark in Remagen) since we saw it was not providing a good correlation with performance. For the correlation we would count with Greif when there is enough data.

 

Any other comment about this please post here.

 

The previous discussion about benchmark was here in this thread.

 

 

Edited November 15, 2020 by chiliwili69

[Gambit21]

I will create a Rhineland benchmark/test mission according to my own needs, (using one of the smaller maps defeats the purpose for me and is counterproductive)  I will post it when it’s completed and modify after I get some feedback.

 

Really, one of the best things to test CPU is to include a formation of complex (TC) tanks. 

In any case, it needs to be a large map since the smaller maps don’t cause nearly the time dilation issues.

[Bernard_IV]

VR makes a big difference if you are looking at the ground, sky, instruments, or wherever else so I don't know if we could make that work or even if just looking straight forward would be accurate or not.  CPU gets bogged down by all of the logic stuff involved with other planes and pieces moving on the ground.  Not sure if it is possible to make them all do the same thing.

[Gambit21]

Actually utilizing AI complex tanks on one of the smaller maps would be useful as well - I could do that instead of Rhineland perhaps.  

 

That said I see that Vander created a mission so perhaps I can just leave this alone for now. Figuring out how many units to add to a smaller map to begin inducing TD would be a time consuming  pain in the arse anyway.

[Jaws2002]

What settings do we use in Nvidia control panel? 

 

[chiliwili69]

12 hours ago, Gambit21 said:

don’t cause nearly the time dilation issues

 

As far as I understand time dilation is when the time in the game flows slightly different than in the real world. So for a long flight (more than 15 minutes) there is some differences since the integrated total time is different. (I assume it will be no more than 2-3 seconds in 15 minutes flight).

I don´t understand these:

1.- Why this is an issue in a normal mission play 

2.- Why this is related to performance (fps)

3.- Why it could be dependent on CPU/GPU.

 

Perhaps it has something to do if you are a mission builder to synchronize all events along the mission?

9 hours ago, Bernard_IV said:

VR makes a big difference if you are looking at the ground, sky, instruments, or wherever else

 

My idea for the VR test is as it was doing in the previous Balapan and Samuel tests. Leave the VR device in a fixed position (or keep your head just straight and fix).

The mission camera is moving the view (it is a external view), so all you have to do is enjoy the view.

I have tried that with the v3 version of the SYN_Vander and it works well.

In the view it is captured many ground objects and plane formation, and some smoke as well. I think it is enough dense to stress the CPU enough.

You can test it yourself, just download the mission and put it in the mission folder.

7 hours ago, Gambit21 said:

That said I see that Vander created a mission so perhaps I can just leave this alone for now.

 

I think the SYN_Vander mission include many AI planes and buildings. It stress the CPU a lot since the resulting fps are really low for my humble 4790K at 4.8GHz compared with the previous Remagen test. Take a look of that.

6 hours ago, Jaws2002 said:

What settings do we use in Nvidia control panel?

Here I was thinking to follow the same settings used in the REmagen test. Mostly default except energy mode with "Prefer maximum performance".

You can take a look in the Remagen test procedure.

As far as I can count in the v3 mission you have 6 Pe-2 formation, 6 IL-2 formation, 6 planes ( IL-2+Yak) formation from where the view is taken.

8+8 ground vehicles (I don´t know if they are tanks)

1 train

Many many buildings around the train station.

 

This capture with GPU test settings:

[Gambit21]

12 hours ago, chiliwili69 said:

 So for a long flight (more than 15 minutes) there is some differences since the integrated total time is different. (I assume it will be no more than 2-3 seconds in 15 minutes flight).

 

It can be, or it can be much, much more than that. 

 

Quote

I don´t understand these:

1.- Why this is an issue in a normal mission play 

 

Well...because the time/movement actually slows down in game. If 20 minutes of real time elapses and only 14 minutes in the sim - that’s a problem. Things are actually happening more slowly, your flight/movement - everything.

 

Quote

2.- Why this is related to performance (fps)

 

 It’s not, that’s been my point all along. It can be (and usually is when it happens) a bigger problem as FPS are often not noticeably affected and thus not an indicator of TD. Performance is not just FPS - be clear on that.

 

Quote

3.- Why it could be dependent on CPU/GPU.

 

It’s CPU. I see it under various circumstances - often (but not exclusively) when multiple flights are engaging in ground attacks on the map, even if not within site of the player. 

 

Quote

 

Perhaps it has something to do if you are a mission builder to synchronize all events along the mission?

 

Yes it can make synchronizing voice commands with subtitles impossible, but make no mistake - you’re flying in slow (er) motion when TD is present.

 

 

Quote

 

I think the SYN_Vander mission include many AI planes and buildings. It stress the CPU a lot since the resulting fps are really low for my humble 4790K at 4.8GHz compared with the previous Remagen test. Take a look of that.

 

Low FPS is indicative of being GPU limited in IL2 more than CPU. Not always as it depends on the cause. Often with too much “math” happening you will get TD and still decent FPS. So you already have a performance problem but don’t realize it.

 

If you’re seeing low FPS below your normal baseline, then the reality is that you probably had a TD issue sometime before that, and now things are bad enough that you have both TD AND lower FPS.

 

Buildings do not test  your CPU - yes aircraft will.

 

 

Edited November 9, 2020 by Gambit21

[SCG_Fenris_Wolf]

That is true.

 

 

@Gambit21 is your testrun ready?

 

Where do I find the mission chili mentioned?

 

@chiliwili69 I must mark though, that CPU stress is not linear from desktop -> VR.  VR limits by bottleneck and something AVX in the code, desktop does not. That alone is sufficient reason to state that the comparison is not linear. So we must do the same test in VR.

 

You must state absolute - total resolutions, and 90Hz mode only though, so we can compare.

 

2 x 2160 x 2160 = 9,33Mio pixels for example - that is 2 Displays x horizontal pixels x vertical pixels = total resolution. So, I suggest the following, and will start with it:

 

For CPU test:

3 Mio pixels

 

For GPU test:

10 Mio pixels.

 

Both at 90Hz.

 

The initial post also needs following information: 

- Direct Download to Vader's mission, with manual how to install the mission

- Direct Download to Fraps + how to use it

 

basically similar like the previous thread.

Edited November 9, 2020 by SCG_Fenris_Wolf

[Gambit21]

3 hours ago, SCG_Fenris_Wolf said:

 

@Gambit21 is your testrun ready?

 

No, haven't started it yet.

Not sure it's necessary at this point, but will try to get it done this week.

[Gambit21]

Working on the Rhineland mission - I'll post when it's done.

[Gambit21]

OK, version 1 of my test mission, FPS, CPU/Time Dilation test.

 

At the 1 min sim time I was at 1 min 40 seconds real time, 2 min sim time equals 3 min, 23 seconds of real time.

So @chiliwili69 you can see how quickly this can become an issue.

 

30 FPS on Ultra (down from my normal steady 60 at high settings, medium clouds

So, at present it's overbuilt. I need to back off so that it resembles something plausible that, brings me closer to (but lower than) 60 FPS and with less TD.

 

So I'll delete a few things, retest and when I have something that I feel might be of use I'll post it.

[chiliwili69]

On 11/9/2020 at 10:22 AM, Gambit21 said:

If 20 minutes of real time elapses and only 14 minutes in the sim - that’s a problem. Things are actually happening more slowly, your flight/movement - everything.

 

Wow!,  I didn´t know the effect was so large. Then it is like having a slow motion experience in the sim. I have never piloted a warbird IRL (only glidders) but everyting (speed, vehicles, etc) was very credible to me. I have then to investigate this more.

On 11/9/2020 at 10:22 AM, Gambit21 said:

Performance is not just FPS - be clear on that.

 

Yep. My previous thought was Performance=FPS (at least in VR). But this Time Dilation issue is completely new to me. So then performance is fps and TD. I need to investigate when this is happening and why.

On 11/9/2020 at 10:22 AM, Gambit21 said:

Low FPS is indicative of being GPU limited in IL2 more than CPU.

 

This is not a rule. Low FPS could be due CPU or GPU or both. It depends on what you demand from the game (graph resolution vs maths) and the power of your GPU  and CPU.

That´s why my idea was to run the benchmark in two modes: One to stress the CPU only and another to stress the GPU only

On 11/9/2020 at 10:22 AM, Gambit21 said:

Often with too much “math” happening you will get TD

 

It seems then that when this intense maths occur some IL-2 internal process gets temporarily halted (time stop) until those maths calcs are finished. The more intense maths the larger TD.

On 11/9/2020 at 10:22 AM, Gambit21 said:

Buildings do not test  your CPU - yes aircraft will.

 

I think buildings also load CPU. To prove that you only need to do a low flight (with not planes or vehicles or tanks) in a dense city using a low resolution mode (1080p or less) and no clouds.

You will see that your GPU is only loaded around 20-40% but you still get low fps.

This is because the CPU has to draw the scene (calculate where every poligon is) and then the GPU can do the render.

This affect is even more pronouced in VR since the scene has to be drawn twice (one per eye).

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

CPU stress is not linear from desktop -> VR

 

I ballpark number we guess for fps in 1080p vs VR is divide by two. But this is just a very unacurate rule in many cases. It depends on many factors.

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

total resolutions

 

Yes, you are right. we should compare total resolution after applying a SS factor since the internal hidden SS factor and resolution of each device is different.

[chiliwili69]

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

For CPU test:

3 Mio pixels

 

Yes, this could be an enough good number for must of GPUs.

But devices with larger FOV have more objects to be built in the scene, so CPU will be more stressed in devices with larger FOV.

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

For GPU test:

10 Mio pixels.

 

This could be a bit low if we want to really stress new GPUs below 90fps. Maybe it could be better to put 12Million pixels (this is 134% SteamVR SS with Index) or perhaps 15 millions.

In any case this is something we can adjust once we test the latest GPUs.

[shirazjohn]

I have experienced time dilation in a couple of bodenplatte career missions where frontline activity is dense and hard difficulty.  I would describe the effect as like bullet time from max payne, the clock runs slower and everything around you moves slowly which gives you the illusion of having more time to take aim and shoot but my framerate stays the same as normal. Its a very strange phenomena i did manage to make a track of this but when played back it runs normally. 

[chiliwili69]

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

Both at 90Hz

 

And what do we do with devices with no mode in 90Hz. Like Rift-S?

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

The initial post also needs following information: 

- Direct Download to Vader's mission, with manual how to install the mission

- Direct Download to Fraps + how to use it

 

Sure.

On 11/10/2020 at 2:42 AM, Gambit21 said:

At the 1 min sim time I was at 1 min 40 seconds real time, 2 min sim time equals 3 min, 23 seconds of real time.

So @chiliwili69 you can see how quickly this can become an issue

Wow! it is almost double. I need to test this in my PC. Thanks for providing a mission to test TD.

But how do you measure the sim time while you run the mission?

[dburne]

Yep time dilation has been in this sim as long as I remember.

Easy enough to see, just observe the second hand on cockpit clock and count off seconds in  head to compare.

Not exact science but you can tell when it is.

 

Many PWCG missions with a lot of activity will have it.

Edited November 11, 2020 by dburne

[chiliwili69]

Well, I have just tested the v4 becnhmark mission of  @SYN_Vander and I think it is ready to be used for testing fps performance (no time dilation) for CPU and GPU.

 

I have run 3 runs with the CPU settings indicated above and 3 runs with the GPU settings indicated above. The results are:

CPU tests
Frames: 2690 - Time: 46000ms - Avg: 58.478 - Min: 50 - Max: 78
Frames: 2715 - Time: 46000ms - Avg: 59.022 - Min: 53 - Max: 78
Frames: 2717 - Time: 46000ms - Avg: 59.065 - Min: 53 - Max: 77

GPU tests
Frames: 4030 - Time: 46000ms - Avg: 87.609 - Min: 60 - Max: 114
Frames: 4021 - Time: 46000ms - Avg: 87.413 - Min: 61 - Max: 116
Frames: 4102 - Time: 46000ms - Avg: 89.174 - Min: 60 - Max: 111

 

I think the noise level is low enough to use it for benchmarking. We had a similar noise with the recorded tracks.

 

Plotting the graphs you have this:

 

The low peak for GPU test is exactly when the IL-2 plane crossed the smoke in the middle of the train station and the top peak is when the view is exactly from the right wing and there is less objects in scene to be rendered. The peak also appear in the CPU test since there is less buldings to be drawn.

 

The higher variability on the GPU runs could due to my 1080Ti has not a fixed OC, it is running on his own.

 

During the CPU test the GPU load was really low, around 30% and 40% during the smoke area:

 

During the GPU test the load was always at 100%:

 

Overall I think we have a good candidate for testing.

I will create a new thread with all detailed instructions, so everyone can test his CPU/RAM or GPU. Specially the new ones.

[Gambit21]

I can never get my GPU to get fully loaded (utilized) in game.

I can be at 30 FPS, and 40% or 60% GPU...it's very strange. Always been this way.

 

Similarly I never see high CPU loads regardless of what is going on in the mission and what the FPS or TD situation is.

 

I'll be curious to see what happens with my new system build.

[No_85_Gramps]

2 hours ago, Gambit21 said:

I can never get my GPU to get fully loaded (utilized) in game.

I can be at 30 FPS, and 40% or 60% GPU...it's very strange. Always been this way.

 

 

Using a 1070ti at 1080p, 60 fps, my GPU would never get above 60-70% utilization. About 6 weeks ago I went to a 1440p set-up at 144 FPS and my GPU utilization hits 99-100% easily. I did not change the settings in-game, just the new monitor. Only issue I ran into was GPU was maxing out temperature wise, I ended up undervolting the GPU and temps are again in the low 70's.

[Gambit21]

1 hour ago, No_85_Gramps said:

Using a 1070ti at 1080p, 60 fps, my GPU would never get above 60-70% utilization. About 6 weeks ago I went to a 1440p set-up at 144 FPS and my GPU utilization hits 99-100% easily. I did not change the settings in-game, just the new monitor. Only issue I ran into was GPU was maxing out temperature wise, I ended up undervolting the GPU and temps are again in the low 70's.

 

That makes sense.

Going from 60 FPS, then turning settings to Ultra and going down to 30 FPS and not seeing the GPU spiking up to 100% makes no sense at all.

[chiliwili69]

8 hours ago, Gambit21 said:

I can never get my GPU to get fully loaded (utilized) in game.

I can be at 30 FPS, and 40% or 60% GPU...it's very strange. Always been this way.

What card and what resolution you play?

 

It can be that the reason to be at 30 fps could the CPU constraint. Try to unload your CPU by deselecting mirrors, shadows and put Low Preset. And put a high MSAA. Your GPU will go to 100%.

8 hours ago, Gambit21 said:

Similarly I never see high CPU loads regardless of what is going on in the mission

 

The real CPU load can not be measured. The %CPU reported by most of utilities is meanningless.

The CPU is bottleneck by a heavy thread which is jumping from core to core, so on average the CPU utilization is low but you are constrained by CPU.

On 11/9/2020 at 6:02 PM, SCG_Fenris_Wolf said:

For CPU test:

3 Mio pixels

 

For GPU test:

10 Mio pixels.

 

Both at 90Hz.

 

Guys, and @SYN_Vander,  I have been running the v4 benchmark mission in VR (using also fpsVR) and it is really really heavy for the CPU load (and also GPU).

Even with all settings put at low (also clouds and no MSAA) and the Index running at 80Hz mode, the CPU and GPU frametimes were in the 18-22 ms region during all track. This means being always at 45fps. 

 

The mission is really heavy. It is a really "black death". And will be not useful to benchmark improvements in VR.

 

I think it could be good to relax a bit the density of buildings and planes to unload the CPU, so we get around 90-120 fps in 1080p monitor with the CPU settings.

 

The GPU load can be regulated in VR by putting more or less SS, but the CPU loads depends totally on the mission.

 

SYN Vander, what do you think? (you can play the mission with fpsVR and see what I mean)

[SYN_Vander]

40 minutes ago, chiliwili69 said:

What card and what resolution you play?

 

It can be that the reason to be at 30 fps could the CPU constraint. Try to unload your CPU by deselecting mirrors, shadows and put Low Preset. And put a high MSAA. Your GPU will go to 100%.

 

The real CPU load can not be measured. The %CPU reported by most of utilities is meanningless.

The CPU is bottleneck by a heavy thread which is jumping from core to core, so on average the CPU utilization is low but you are constrained by CPU.

 

Guys, and @SYN_Vander,  I have been running the v4 benchmark mission in VR (using also fpsVR) and it is really really heavy for the CPU load (and also GPU).

Even with all settings put at low (also clouds and no MSAA) and the Index running at 80Hz mode, the CPU and GPU frametimes were in the 18-22 ms region during all track. This means being always at 45fps. 

 

The mission is really heavy. It is a really "black death". And will be not useful to benchmark improvements in VR.

 

I think it could be good to relax a bit the density of buildings and planes to unload the CPU, so we get around 90-120 fps in 1080p monitor with the CPU settings.

 

The GPU load can be regulated in VR by putting more or less SS, but the CPU loads depends totally on the mission.

 

SYN Vander, what do you think? (you can play the mission with fpsVR and see what I mean)

 

We need to find out what is taxing the CPU so much, but I'm guessing the number of planes in that scene. I can cut them in half and see how that turns out?

 

Here is v5 with half of the aircraft and also the 'barrage' (dummy explosions) removed. Also added 3 seconds.

 

Benchmarkv5.zip

Edited November 12, 2020 by SYN_Vander

[chiliwili69]

13 hours ago, SYN_Vander said:

I can cut them in half and see how that turns out?

 

Many thanks SYN_Vander for producing a new benchmark so quickly!

 

I have tested the new benchmark with 3 runs with CPU stress settings and 3 runs with the GPU stress settings:

 

CPU test
Frames: 3768 - Time: 49000ms - Avg: 76.898 - Min: 64 - Max: 105
Frames: 3788 - Time: 49000ms - Avg: 77.306 - Min: 63 - Max: 105
Frames: 3770 - Time: 49000ms - Avg: 76.939 - Min: 64 - Max: 104

 

GPU test
Frames: 4505 - Time: 49000ms - Avg: 91.939 - Min: 73 - Max: 109
Frames: 4549 - Time: 49000ms - Avg: 92.837 - Min: 77 - Max: 109
Frames: 4547 - Time: 49000ms - Avg: 92.796 - Min: 78 - Max: 110

 

The CPU test increases 17 fps and the GPU about 4 fps. That´s good sign. The total lenght is now 49 seconds.

The avg values are quite consintent so we will not need to make 3 runs and calculate averages.

The variability in the CPU test is really low. In the GPU is also quite enough low.

The plots are:

 

During the CPU test the GPU was around 40% and again during the GPU test the GPU was at 100%. So this benchmark will independently identify well the best CPU/RAM and best GPU for IL-2.

 

Now we can talk about VR.

I have run the test multiple times with fpsVR. The mission is incredible demanding for VR in Index (with good FOV). In order to have some values between the measurable band I had to put ALL IL-2 setting to low, put 80Hz mode and put 100%SS in SteamVR.

The the values I got was:

 

VR test - all settings to low -80Hz with Index at 100%SS
Frames: 2743 - Time: 49000ms - Avg: 55.980 - Min: 38 - Max: 81
Frames: 2696 - Time: 49000ms - Avg: 55.020 - Min: 38 - Max: 81
Frames: 2647 - Time: 49000ms - Avg: 54.020 - Min: 38 - Max: 80

 

And this plot:

So, most of the time it is between the 40 to 80 fps, which is what is desirable to measure improvements in VR. In the benchmark the band will be 45-90.

So for VR, I think there will be only one test. Basically all settings to low, 90Hz and %SS to get 9 million pixels. (index at 100% is 9 million pixels).

Since most of the rigs out there are superior to my rig I think they will be in the 45-90 range.

 

I think this v5 is ready to be our new benchmark. I hope new CPU and GPU will be well evaluated in IL-2. And we will know better what tricks we can play.

 

Edited November 12, 2020 by chiliwili69

[Gambit21]

@SYN_Vander 

 

Suggestion (I haven’t run it yet)

Add a TD subtitle group - I can zip my little group up later if you want. Then I can just forget about my own test mission and yours can be the standard (assuming TD is occurring)

 

I have a subtles as follows.

 

5 sec “Time Dilation Test in”

1 sec “3”

1 sec “2”

1 sec “1”

1 sec “Begin” (player starts stopwatch) 

1 min “1 minute”

2 min “2 minutes”

 

60 point font, centered 

 

You can do 30 sec, and 1 min maybe?

A longer mission is better to test gradual creep mind you, but I think your mission might nevertheless be of use in this regard.

[SYN_Vander]

Yes, I thought about that, but since you have to unpause the mission anyway isn’t that a good starting point? 
Or do we need this because you can’t press two buttons at the same time?

Anyway, I’ll add it in.

[chiliwili69]

I think the time dilation effect is not present in this v5 mission since all the test I have done (all CPU tests ad GPU tests) finish exactly at 49 seconds after I press the P.

The IL-2 plane with the external camera reach the same place after the 49 seconds. It is covering the same distance. At a fix speed.

 

But the internal time of IL-2 can not be seen.

[chiliwili69]

I started to produce a draft of the post of the procedure here:

https://forum.il2sturmovik.com/topic/66924-syn_vander-benchmark-v5-to-measure-il-2-performance/

 

[dburne]

10 hours ago, chiliwili69 said:

I think the time dilation effect is not present in this v5 mission since all the test I have done (all CPU tests ad GPU tests) finish exactly at 49 seconds after I press the P.

The IL-2 plane with the external camera reach the same place after the 49 seconds. It is covering the same distance. At a fix speed.

 

But the internal time of IL-2 can not be seen.

 

Odd I think every PWCG mission I flew today had some TD going on.

Edited November 14, 2020 by dburne

[SYN_Vander]

1 hour ago, dburne said:

 

Odd I think every PWCG mission I flew today had some TD going on.

 

I don't think it's odd. The benchmark mission has a fair amount of planes and vehicles in a small area, but it's not a big number in total. Most SP missions (including the one I generate) have many more planes and vehicles on the map.

[chiliwili69]

On 11/13/2020 at 7:48 AM, SYN_Vander said:

Anyway, I’ll add it in.

 

I think the last v5 mission is good enough to be used to measure fps against the CPU, GPU and VR. 

I don´t know if it suffers from TD. Perhaps the easiest way to know it would be to simply add a subtitle with the time in seconds. Maybe 40 point size in the upper right corner, just below the Fraps counter. Starting at 0 before pressing the P.

 

TD is still for me a little bit misterious. I think it would be better to use the longer Gambit benchmark to work and experiment on that. The good thing of this benchmark is that it is quick to run, only 49 seconds.

[SYN_Vander]

I added a counter every 5 seconds. Mission length is now 60 seconds, although it seems the 'mission end' command takes some time to execute. I put a little more distance between first and second flight so the second flight doesn't need to evade in the end; hopefully avoiding more randomness. Also made the train move.

 

Not sure if the counter helps, the duration is probably too short to do TD tests. It's better to have separate benchmark missions for that, with various amounts of AI.

 

Benchmarkv6.zip

[chiliwili69]

Many thanks, I think 60 seconds is a better and perfect length for the track.

I have done 3 runs of each test and the results are a bit higher on CPU now (the added lastest seconds had higher fps):

 

V6
CPU
Frames: 5013 - Time: 60000ms - Avg: 83.550 - Min: 74 - Max: 115
Frames: 4919 - Time: 60000ms - Avg: 81.983 - Min: 73 - Max: 114
Frames: 4933 - Time: 60000ms - Avg: 82.217 - Min: 72 - Max: 112

GPU
Frames: 5559 - Time: 60000ms - Avg: 92.650 - Min: 72 - Max: 107
Frames: 5618 - Time: 60000ms - Avg: 93.633 - Min: 77 - Max: 109
Frames: 5623 - Time: 60000ms - Avg: 93.717 - Min: 76 - Max: 113

VR (Index at 90Hz)
Frames: 2987 - Time: 60000ms - Avg: 49.783 - Min: 44 - Max: 74
Frames: 2911 - Time: 60000ms - Avg: 48.517 - Min: 44 - Max: 68
Frames: 2847 - Time: 60000ms - Avg: 47.450 - Min: 43 - Max: 66

 

 

On the CPU it seems there is more variability, but I had a long download process (SpiderMan of my son) which I was not allow to stop. So in my 4 cores CPU it may have an impact.

 

 

The VR test is now at 90Hz (before was at 80Hz) and clearly it is always sub 90fps, which is what we wanted with this track. There are new CPU beasts out there which for sure will raise the bar higher. And this is exactly what we are looking for.

 

Regarding Time Dilation, I meassured with a stopwatch and when it is pulsed the P, the internal timing is around 1.5 or 2 seconds. This small delay is mantained along the track and the last 60 seconds text shown by the mission appear exactly about 1.5-2 seconds before finishing the fraps counter and the mission. So I think there is no time dilation in this track. We will use the longer Gambit Rheinland benchmark to experiment with that.

 

I will conclude here the feedback collection and will use the v6 as the track to run the test in the dedicated benchmark thread.