[REPORTED]FPS Performance Test Mission

[wilbur81]

Hey, ED Team (and everyone else who's more tech savvy than I).

 

I created a heavy stress test mission in Open Beta after today's (4-17) little 2.5.6 hotfix for the sake of comparing the latest OB with the latest Stable version. Here are my results with the same graphics settings for both builds.

 

When the mission starts in the latest 2.5.6 OB, panning the mouse around the Hornet cockpit before takeoff, I'm doing between 6 and 8 FPS.

When the mission starts in Stable, I'm doing between a smooth 40 and 58 FPS, with dips no lower than 30 FPS.

 

LrgTEST1.miz

 

Keep in mind, my Stable version is installed on a slow HDD and my 2.5.6 OB is on my fast SSD.

 

Please use and dissect this mission, if you'd like, to get an idea of what a heavy (400+ units) mission is like between 2.5.5 and 2.5.6. Again, on my moderately decent system, this mission is VERY playable with the same high settings in 2.5.5 (OB or Stable versions).

 

With minimal units, in 2.5.6 OB, just flying around, frame rates are good to excellent on my rig. However, adding (ground?) units in 2.5.6 just kills it. For what it's worth, my previous stress-test mission worked better in the 2.5.6 before yesterday's update. So the latest update is actually performing worse for me when adding more than a few units.

 

Would love to get some feedback from other users who have the ability to test this mission more in-depth between 2.5.5 and the latest 2.5.6.

It would be great to see if the FPS loss has to do with...

- AI path execution.

- Ground vehicle / soldier weapons firing

- Poorer cpu utilization in 2.5.6

- 2.5.6 lighting enhancements

- etc.

 

 

Again, I'm playing in 2D at 4K.

 

-Cheers!

 

PS - I don't use any mods and I'm an FXO/MS2 folder deleter.

Edited April 17, 2020 by wilbur81

[danvac]

It seems it could be reproducing issue from: https://forums.eagle.ru/showpost.php?p=4290798&postcount=1

[tintifaxl]

I can't imagine the devs don't know about the AI pathfinding eating up all the cpu cycles. They just haven't found a way to fix it yet.

[Distiler]

I've tested the mission and my fps go from 60-80 to 5-10 in a matter of seconds. GPU tax around 5% only! so it must be waiting for the CPU to do something. I tried some things:

 

-I've deleted blue airplanes. No effect.

-I've deleted blue and red airplanes. No effect. So, the air planes and air war is not the issue.

 

Then I've started fresh and only deleted waypoints from the ground units that have waypoints. The number of ground units kept the same, just the moving ones kept at waypoint 0, stopped. The ground battle went on with shots everywehere (and airplanes going on too)...and my fps stayed at 60-80.

 

So i'ts either 1) the pathfinding of ground units and/or 2) the dust created by moving ground units.

 

I'm attaching the edited mission

 

PD: latest Open Beta, 2D 1080p.

LrgTEST1_no_ground_movement.miz

Edited April 17, 2020 by Distiler

[wilbur81]

I've tested the mission and my fps go from 60-80 to 5-10 in a matter of seconds. I tried some things:

 

-I've deleted blue airplanes. No effect.

-I've deleted blue and red airplanes. No effect. So, the air planes and air war is not the issue.

 

Then I've deleted waypoints from the ground units that have waypoints. The number of ground units kept the same, just the moving ones kept at waypoint 0, stopped. The ground battle went on with shots everywehere (and airplanes going on too)...and my fps stayed at 60-80.

 

So i'ts either 1) the pathfinding of ground units and/or 2) the dust created by moving ground units.

 

I'm attaching the edited mission

 

Wow...fantastic! Thanks for testing, Distiler. This is exactly what I was hoping for. :thumbup: Anyone else willing to test?

[Gryzor]

I can confirm the same as you. There´s some serious pathfinding bug, maybe not only related with this one.

[flyingsonofagun]

Confirmed, as soon as I move 3 or 4 tanks its game over, literally. Server crash. Some tanks were even going up in altitude as they were following the way point.

[NineLine]

Thanks for the test mission, looking into it now.

[NineLine]

Here are the results I got with both versions and both missions.

 

Also, I do note that in Stable, its darker and no shadows as well, might be part of it. At least the difference in the versions anyways.

 

[wilbur81]

Thanks for testing, NineLine. :thumbup:

[NineLine]

Thank you for the detailed report, really makes getting these items to the team much faster, I appreciate it.

[VirusAM]

Here are the results I got with both versions and both missions.

 

Also, I do note that in Stable, its darker and no shadows as well, might be part of it. At least the difference in the versions anyways.

 

 

Nineline what do you use to pull fps data from the sim?

I don't think you just do it by hand right?

 

You already told me that you use excel to generate the graphs and that part is ok for me.

But how you pull the FPS data from the sim and import into excel?

Is there some benchmarking tool that export fps data second per second to a csv or any kind of text file?

[VirusAM]

Anyway another mission that is cutting my FPS literally in half with stutters everywhere is the 2nd mission of the stock hornet campaign Task force challenger 1989.

You can find also as a single mission in the Hornet mission tab.

It is called 1989-CAS if i recall correctly.

In VR there if have only 20fps (rift s), in 2d when i am towards the battlefield (wpt2-wpt3) i can go as low as 27fps, while usually i am around 70/80 fps.

Also in the release this mission is heavy, but there i manage to have about 50fps.

That is why i asked Norm for the method he uses to have scientific data about fps from the sim.

I want to generate, and report to ED guys the same kind of report

[NineLine]

I use Fraps Benchmarking, I use a track about 3 mins long, and run the benchmarking for about 2 mins, and compare.

[Eldur]

It's great to have a dedicated test mission. I'd dare to say we even do need an official one from ED with a track replay made within that acts as a benchmark reference. I remember that other sim had a certain track out of the box that literally everyone used for checking how their rig would do with certain settings. Something similar is needed here, so we actually can compare. Doesn't make too much sense to have every user report their frames based on literally a different situation per user where you can't pinpoint the exact reason of a certain frame drop whereas in one official mission / track you know exactly what happens when and see why some rigs act up and others don't. And to go even further I'd appreciate to have that as an actual benchmark function with DCS itself tracking the fps, frametime and other relevant data, outputting something like a csv or something that we could grind though a given excel sheet to have these graphs at hoof, or having that displayed directly inside DCS eventually. I mean it's about easing the troubleshooting process after all.

[VirusAM]

I use Fraps Benchmarking, I use a track about 3 mins long, and run the benchmarking for about 2 mins, and compare.

 

 

 

Of course Fraps...why didn’t i think of it....

This way i can also so test longer then 2 minutes.

You are genius Nineline thanks

[VirusAM]

[REPORTED]FPS Performance Test Mission

 

@NineLine

 

Here you can see my version of the graphs.

The interesting thing about this test is that sure the AI moving has an impact (devastating impact i would say because the sim becomes unplayable with complex/realstic missions)

But the non moving version pf the mission in the openbeta has worse performance then the moving version in the release (if you leave out the first 7/10seconds where probably some computation is ongoing). So probably dcs need still some optimization a part of the ai moving (pathfinding?)

 

 

Edit: the settings are the same between ob and release obviously and in 2d

Edited April 19, 2020 by VirusAM

[NineLine]

Yes, a fix is in for it, I am just waiting on a new build to test. Our development version never showed this issue, so it was something in building the public release. Hope to have something soon on it, thanks for all the help guys... it makes things easier to convey to the team for sure.

[Gryzor]

Yes, a fix is in for it, I am just waiting on a new build to test. Our development version never showed this issue, so it was something in building the public release. Hope to have something soon on it, thanks for all the help guys... it makes things easier to convey to the team for sure.

 

Maybe in public release 2.5.6 there other performance issues like this, due bad mixing in the public version, no so significant like the AI moving trucks huge gap, but it could be related to others.

 

It surprise to me..., mainly, because I remember MANY times Kate´scomments about fps increase planned for the last beta, ...and that fps "fix" didn´t occours (nor significant or insignificant,... none... even there was no mention in changelog .... very extrange, right?).

 

In addition, many of us didn´t see even a 1 or 2% increase in fps, just the opposite, more worse performance,... soy maybe in the editing process of the public beta, and considering that changelog did´nt include any note about fps increase or fixes, those fixes were missed/forgotten in the last patch. So it could be possible, NineLine?

Edited April 21, 2020 by Gryzor

[NineLine]

I have a feeling this bug alone is hiding any positive improvements we would have seen in many cases.

[RealDCSpilot]

Uh, that's not good. If the dev build differs too much from the public build. No wonder that crazy problems pop up after every release. However, there are fps improvements in the last couple of beta releases, but you only see them in an empty free flight map. And in VR i need to have "stereo mode use shared parser" enabled, ED should embrace that option much more and iron out it's current weaknesses.

[VirusAM]

[REPORTED]FPS Performance Test Mission

 

Yes, a fix is in for it, I am just waiting on a new build to test. Our development version never showed this issue, so it was something in building the public release. Hope to have something soon on it, thanks for all the help guys... it makes things easier to convey to the team for sure.

 

 

 

@Nineline

 

i have a track i made in the release version, done to show you the graph for sa page issue.

I discovered an interesting thing.

While reproducing the track on the openbeta, at the start of the track i turn on the internal lights and that immediately drops the fps from 50 to 8. I was not able to view all the track to the point i used the sa page and tried on/off with the cm dispenser.

So probably something is going on with the lights as well.

I can immediately share the track if you want it .

Anyway the mission is a stock f18 one, magnum force in the caucasus theater

And later i will share the graphs for the sa page on the sa page issue thread.

Just let me know how to send to you (I don’t think i will be able to upload to the forum).

[Knock-Knock]

Yes, a fix is in for it, I am just waiting on a new build to test. Our development version never showed this issue, so it was something in building the public release.

 

Copy protection?

[wilbur81]

Regarding this thread and today's line from the change log notes: "Reducing the CPU load when path finding a group of infantry."

 

... Our test mission with moving vehicles improved a little. Instead of grinding to a halt around 6-8 FPS with moving vehicles, it now "only" dips down to about 15 or 16 FPS from steady highs around 40 FPS (with all the same test conditions of course). And the overall FPS performance averages are now, apparently for me, higher. It's still not very playable due to the fairly dramatic stutter of going from 40ish to 18ish FPS... but it is DEFINITELY an improvement for this issue in 2.5.6.

 

Feel free, anyone, to run the test mission again and post results after today's patch.

 

Good steps in the right direction. :thumbup:

Edited April 22, 2020 by wilbur81

[BIGNEWY]

Regarding this thread and today's line from the change log notes: "Reducing the CPU load when path finding a group of infantry."

 

... Our test mission with moving vehicles improved a little. Instead of grinding to a halt around 6-8 FPS with moving vehicles, it now "only" dips down to about 15 or 16 FPS from steady highs around 40 FPS (with all the same test conditions of course). And the overall FPS performance averages are now, apparently for me, higher. It's still not super-playable due to the fairly dramatic stutter of going from 40ish to 18ish FPS... but it is DEFINITELY an improvement for this issue in 2.5.6.

 

Feel free, anyone, to run the test mission again and post results after today's patch.

 

Good steps in the right direction. :thumbup:

 

Good to hear it has improved, we do have more improvements to come.

 

thank you