Jump to content
IckyATLAS

How many bombers in one mission ? What Framerate?

Recommended Posts

I have created a Kuban map scenario and tried to evaluate how may bombers can be used and how playable it can be.

It uses the Kuban autumn map.

Atmospheric conditions are daylight in September at 10:30,  with medium cloud patches.

Light wind and medium rough sea.

Display is 4K resolution 3840x2160 pixels.

My version of IL2 BOK uses many many mods, for grass, trees, smoke, distance visibility, improved explosions, tracers etc. etc.

All graphical game settings are absolutely to the MAX Ultra you name it.

This to say that all of this eats up additional resources, and so the test is in pretty loaded conditions.

 

 

The scenario is made also to have some load in terms of resources so that there is a good combination of air sea land activity to judge the performance and playability.

 

The German side:

1) In the Kuban region the city port of Novorossiysk is occupied by the German troops. The city is heavily defended.

2) They have installed heavy and light AAA defenses all round and in the city and the port. Some AAA are on trucks and on the trains.

3) Atop the peaks of the mountains that are just east of the city they have also installed 88mm Flak guns to intercept incoming air raids incoming from the east.

3) In the city there are cargo loaded trains and AAA equipped fuel tanker trains. These trains move around and go in and out of the city.

4) In the port there are cargo ships and torpedo boats at rest.

5) From the sea a convoy of three cargo ships of which one is equipped with AAA artillery, speed towards the port to bring in supplies.

6) The airport southwest nearby has BF109G2 fighter planes ready.

 

The Russian side:

They have decided to launch a heavy bombing on the city, port installations, fuel depots, railway stations and the incoming cargo fleet.

To that end the have set three bomber groups made each of 9 A20B bombers loaded with internal and external bomb loads.

One bomber group will target the port installation. Another group will target the rail and oil tanks. And finally the third group will attack at sea the incoming cargo ships.

The total number of bombers is 27.

Each group flies in V formation. They are staged at 2000, 2250 and 2500 meters vertical, and the three groups horizontally are in an V shape group formation.

After having terminated their bombing run the remaining bombers fly back and land at Krasnodar-1. Their return route is more dispersed, such as to avoid all of the groups arriving at the same time.

 

As you can see this is a pretty loaded situation.

 

Results with a 27 Bomber scenario:

 

(I ran more tests and refined the values)

 

This scenario is very playable. Framerate never went under 40, and varied between 40 and 110. No stuttering.

In the sky when flying with the bombers and having on the screen the three bomber groups framerate was around 45-50.

During the bombing runs and all the AAA firing tracers, explosions, smoking ruins and fires, again around 40.

But when during the scenario you would look at say the ship convoy then the framerate would jump to over 95.

This means that even if the simulation is running, but what is displayed is not too loaded then the impact on framerate is low.

To put things in perspective, on an empty lapino map with one plane, the framerate is nearly 170.

During the run if you would look at the empty but cloudy sky but without any object then framerate tops at 150. This shows that the CPU side can handle no problem even extremely complex situations and the framerate killer is only depending on graphics and what you see in the viewport. 

 

This result is made with a PC running Win10 Pro 64bit, DirectX12,  Intel i7 7820X 8 core CPU running at 4GHz flat, Asus X299 Deluxe motherboard, 64GB overclocked low latency DDR4 ram, and SSD hardisks. The Graphic Card is an Asus GTX 1080 Ti running stock speed. Probably with a i9 9900K CPU running at 5GHz, and an OC RTX2080 Ti we could get probably 15 FPS more, but with a heavy price to pay for that.

 

I will try adding a 4th Bomber group (total 36 bombers) to this scenario and see what happens.

 

 

 

 

 

 

Edited by IckyATLAS

Share this post


Link to post
Share on other sites

32 is about the max.

I ran tests a while back.

Share this post


Link to post
Share on other sites

So I added a 4th group of A20B bombers. Their objective is to bomb the airfield nearby the city where in total about 12 airplanes are stationed, bombers and fighters, and a group of three fighters is on the runway engine running.

 

So now we have a total of 36 A20B bombers and opposing fighters and bombers. All the aircraft in the airfield are active and so not just static objects. It was amazing to see the HE111 bombers parked on the airfield engaging the bombers that did a low pass with their top fuselage machineguns.

 

Results are excellent. The framerate very occasionally dropped from the minimum of 40 to 35-37, and the averages stayed well over the 50's.

 

All this is very encouraging. So my next test is with a Fifth bomber group, or a total of 45 bombers attacking the port. I have to find new targets.

 

Stay tuned....... 🙂 

Share this post


Link to post
Share on other sites
16 hours ago, IckyATLAS said:

But when during the scenario you would look at say the ship convoy then the framerate would jump to over 95.

This means that even if the simulation is running, but what is displayed is not too loaded then the impact on framerate is low.

This shows that the CPU side can handle no problem even extremely complex situations and the framerate killer is only depending on graphics and what you see in the viewport. 

I did a test a while back where I put 1024 trucks on an airfield with the player plane, and the results were consistent with your observations, but when looking at it a bit more closely, there seems to be something a bit odd: Firstly, I have a pretty old PC running a 1st gen i7 CPU (i7-920 2.8ghz) with a 6gb GTX1060 and 12gb ram, so my pc is on the weak side. Even so, during this test my PC didnt seem to even break a sweat (or at least not in any way I could measure) despite terrible performance in game. These were my results:  The framerate when not looking at the trucks was around 60, which is on the low side of what I get normally, but dropped to a total slide show (3-4fps) when looking at the trucks. Seemed predictable at first, but this is where it gets interesting.. I wanted to try and see what was actually causing the slow down, so I did a bunch of tests: At first I thought it was the CPU not being able to handle all the objects since its the weak link, but when I checked the CPU usage, it was still pretty low (75% max on one core, and about 20%-40% on the others - average CPU usage 30%). If it was the CPU bogging it down, we would expect to see at least one core sitting at 100%, but that was not the case. Next, I checked RAM usage just to rule that out, but the game was only using about 2.1gb of ram even with all those objects, and I had 4 gb free still so it didnt seem to be that either. Finally, I thought it must be the graphics card thats the bottleneck, being unable to render 1024 3d models and slowing everything else down, but the GPU was basically idling with usage sitting at only 30% the whole time, despite only managing 3-4fps,  and GPU memory usage never went above 1.5gb. 

 

So I am pretty confused as to what is going on here - maybe someone else has some ideas, but the slowdown doesnt seem to be hardware related because even my old and relatively slow PC was nowhere near 100% usage on any component that I could see, so there seems to be something in the game itself that slows it down. I have attached the mission I used here (unzip it to  your "\IL-2 Sturmovik Battle of Stalingrad\data\Missions" folder) and it would be interesting if others who have more powerful PC's could test it and see what results you get so we can see if its a game or hardware limitation. Post your fps results here 😂

ObjectTest.zip

Edited by Flashy
  • Upvote 1

Share this post


Link to post
Share on other sites

The data transfer capacity could be a bottleneck between the CPU and GPU. I suppose your old PC did not have the PCIe-3.0 connection with 16 lanes directly devoted between CPU and GPU without going through the chipset.

I will do some additional tests, and thanks for the mission, I will try it.

 

Share this post


Link to post
Share on other sites

My test was with He-111’s.

Netscape came up with roughly the same results I think.

 

Flashy I’ve seen the same slowdown, looking at much fewer vehicles - even if they’re very distant and only dots.  

 

Point the camera away, frames go back up. I’ve reported it several times since I couldn’t make sense of it.

Share this post


Link to post
Share on other sites

Is that 2d on a screen or in VR? VR makes a lot of difference to frame rates and the like even before throwing all those bombers in.

Share this post


Link to post
Share on other sites

On my system I have 10 FPS with all trucks and the plane viewed. When I look only partially to the trucks then I get a little more fps about 15 and then when they are out of view I shoot to over 60 FPS. I am still very puzzled about this. 10 FPS is nearly 3 times more than Flashy but it should be much more if we compare the systems.

Something is going on that is indeed not clear.

Remember I am in 4K all Max Ultra antializing to max, distances max, and I have a very modded system. I will check with mods disabled.

Mods disabled does not change anything 10 FPS no more.

 

Edited by IckyATLAS

Share this post


Link to post
Share on other sites

Now played the same scenario with 5 Bomber groups, total 45  A20B. No change, Framerate never dropped under 37. And with all bomber groups in view you stay at 43-45.

That's pretty awesome to see this bomber stream at different altitudes and positions flying all in the same direction.

I have the feeling that you can add one or two additional bomber groups and the FPS impact  will not change much. 

Somebody asked if one could have a mass of 50 bombers, and I would say YES definitively, if you have a powerful enough system and graphic card.

 

My next test will be flying a rotte of 4 Focke Wulfs into this bomber stream, with all gunners firing, and see the impact, but I feel that it should remain around 40-50 FPS at least and that is really good enough.

Edited by IckyATLAS

Share this post


Link to post
Share on other sites
48 minutes ago, IckyATLAS said:

Now played the same scenario with 5 Bomber groups, total 45  A20B. No change, Framerate never dropped under 37. And with all bomber groups in view you stay at 43-45.

That's pretty awesome to see this bomber stream at different altitudes and positions flying all in the same direction.

 

Icky, any chance you can upload the file once you get the FW-190s in, so that we can benchmark it with different hardware? Would like to know how my i7-2600@4200Mhz and 980TI SLI combo handles this.

Share this post


Link to post
Share on other sites
14 hours ago, IckyATLAS said:

The data transfer capacity could be a bottleneck between the CPU and GPU. I suppose your old PC did not have the PCIe-3.0 connection with 16 lanes directly devoted between CPU and GPU without going through the chipset.

I will do some additional tests, and thanks for the mission, I will try it.

 

 

Yes, I had thought of that as a possibility. Even though my board has the full 16 lanes, its Pcie-2.0 so there could definitely be a restriction there. Hence the reason I only got a 1060 - anything faster would probably be bottlenecked by the board.

 

 

12 hours ago, Gambit21 said:

 

Flashy I’ve seen the same slowdown, looking at much fewer vehicles - even if they’re very distant and only dots.  

 

Point the camera away, frames go back up. I’ve reported it several times since I couldn’t make sense of it.

 

12 hours ago, IckyATLAS said:

On my system I have 10 FPS with all trucks and the plane viewed. When I look only partially to the trucks then I get a little more fps about 15 and then when they are out of view I shoot to over 60 FPS. I am still very puzzled about this. 10 FPS is nearly 3 times more than Flashy but it should be much more if we compare the systems.

Something is going on that is indeed not clear.

Yeah its our old friend the AI routine again I think - the engine is doing something weird when they are all on screen at the same time that doesnt happen when they are out of sight. Just for interests sake, I tried replacing the trucks in my test mission with static trucks and the results are.. well.. bizzare is the only word. Its hard to describe in words, but when you look directly at all the trucks, the game stops rendering them and most of them dissapear! And when you look at them from about 45° angle, they all suddenly re-appear!  But crucially, there doesn't appear to be a performance drop when rendering large amounts of static objects, so that seems to confirm that its not a case of bandwidth to the GPU etc.  I have attached this new mission - try it out its really weird..

StaticObjectsTest.rar

Share this post


Link to post
Share on other sites
9 hours ago, sevenless said:

 

Icky, any chance you can upload the file once you get the FW-190s in, so that we can benchmark it with different hardware? Would like to know how my i7-2600@4200Mhz and 980TI SLI combo handles this.

 

I'll post it when ready. It can be an interesting benchmark indeed.

  • Like 1
  • Thanks 1

Share this post


Link to post
Share on other sites
53 minutes ago, IckyATLAS said:

 

I'll post it when ready. It can be an interesting benchmark indeed.

 

Great! Can´t wait to check it out.

Share this post


Link to post
Share on other sites
On 2/5/2019 at 9:33 AM, sevenless said:

 

Great! Can´t wait to check it out.

I have completed the scenario so as to use even more the capabilities of the ME, and the majority of the available objects of all kinds. I have it made so that you can either run it in pure benchmark mode, or you can jump in the plane and enter the fray, in which case you will also have the possibility to evaluate the perf from your point of view.

I hope to be able to post it by this Sunday at latest. Stay tuned 🙂 

Edited by IckyATLAS
  • Like 2

Share this post


Link to post
Share on other sites

I will use this thread for ME related details, as the Massive Attack V1.0 release was made in the Single Mission Forum, where I will comment only on the mission elements.

 

1) The clock slowdown

I measured precisely the Internal Clock slowdown and the ratio is approx. x 1.8 on my system. This means for an ingame trigger timer set up at 3 minutes or 180 sec the real time will be of 5 minutes and 24 seconds. You need to take care about this kind of differences mainly if you consider synchronizing between vehicles trains ships and planes. I have the feeling that the priorities are not same across all the objects and planes may have higher priority.

 

2) The vehicle convoys

 You have to be careful about the speed capability of each vehicle as well as the starting and acceleration when you build up a convoy. If you mix vehicles with different performances then in the open field no problem. The convoy starts and the faster vehicles will try to move around the slower in front. This is even more visible if you set a max speed to the convoy.

The trouble is if you start in the middle of a village. The faster accelerating vehicles will try to overpass the slower in front of them but doing so they will hit the objects nearby the road like houses. They will go through garden fences no problem. Depending at what speed they hit they will crash and burn. To solve this put the faster vehicles in front and the slow ones behind. It is not logic but it works. The formation dense or loose etc. does not solve the problem.

 

3) The Roads on the Novorossiyrsk city map (Kuban)

I have experimented with convoys across the city roads and here the devs have bugged it when placing the houses and buildings relative to the roads. In the sharp turns at 90degrees  angle in some places the buildings are too near the road and there is not enough clearance for the vehicles to turn and thy will crash on the building, the vehicles behind may go around and crash in an opposite building and then it is a mess. This is because the trucks follow the road but in the turns they have some dispersion in their angle turn, which is nice and very real and makes things less robotic. The only solution, is to move a little the buildings about two three meters to give more space in these right angle road curves.

 

 

Share this post


Link to post
Share on other sites

I haven't had your convoy troubles - but then I use some sense with waypoint speed and vehicle composition.

Share this post


Link to post
Share on other sites
1 hour ago, Gambit21 said:

I haven't had your convoy troubles - but then I use some sense with waypoint speed and vehicle composition.

You are right, as at the beginning I made my convoys not taking these issues into account. Learning by try and fail. I tried to limit the number of waypoints on an already loaded system. I do not know how the WP are handled but I should try to position additional waypoints in the middle of those steep angle turns, with high priority and maybe, they will be forced to pass more precisely in the middle of the road and so avoid colliding with the buildings near the road. I will check it.

 

I also remarked that if the first waypoint is too far from the convoy, but on a point were normally the convoy can go by the road connection, then the convoy trucks start to wander all around. And this even if you have set a Command Formation with OnRoad Column. With trains there is no issue, because they must follow the rail tracks. So you can just have one waypoint very far away, as long as there is a valid track connection even with multiple track crossings and line switches, it works perfectly. Not so with trucks and road convoys. The probable reason is that they are allowed to go offroad. 

An additional difference is that trains are strongly connected elements, the railcars have no individual freedom. Not so with trucks tanks and other vehicles. They are independent and can act on their own.

 

Edited by IckyATLAS

Share this post


Link to post
Share on other sites

I just haven’t seen that behavior after building and testing many convoys.

 

But then I don’t put a half-track or a tank behind a truck and set the waypoint to 80km.

 

Be sensible and you needn’t get all the dither about vehicles type within reason.

 

Also “limiting waypoints” is unnecessary.

Onlt a single waypoint is active at any one time. You need to make a distinction between placed logic and active logic.

Share this post


Link to post
Share on other sites

We experimented in the Massive Attack or the 50 Bomber mission, the differences between the CPU load and the GPU load.
You can have a decent or acceptable framerate if you have a top graphic card, but a very slow overall game caused by the CPU, which kills the playability.
In IL2 it is the CPU performance that will be the main factor and so when designing a scenario that is the parameter to consider.

 

Let me remind you that even if there are 50 Bombers, this scenario involves in reality 65 active planes. All parked planes (10 planes of various kinds) are fully active and not static objects. In addition you have the player plane and four fighters.

 

Two questions here: 
1) How can you measure the CPU performance 
2) What elements of a scenario will impact more or less the CPU performance.

 

Answer to (1)

To answer the FIRST question I have set up in the Massive Attack V1.0 mission an additional internal 30 second timer. This timer is started a minute or so after the mission is launched.
When the timer starts a START message is flashed on the screen, at the same instant I also start a stopwatch, and when the timer stops a STOP message is flashed, and I stop my watch.
I can now compare the two times and have the time slowdown factor (TSF) due to CPU load. A factor of 1 is perfect as 30 seconds ingame are really 30 seconds. Over 1 means that the game is slowed down.
In the version of the game that is published and you can download this factor is approx. 1.89. This is the average of measurements at different places in the mission and you go from 1.86 to 1.93.
I have put also another timer that runs during the air combat between the fighters and A20s, the firing of AAA from ground, ships and trains, oil tanks on Fire etc... This increases slightly the load by 11% and we end up at 2.10. By the way the framerate was a very decent 60 FPS as seen from a FW190 in the midst of combat with a group of A20B.

 

Answer to (2)

To answer the SECOND question I will modify the mission by taking away certain elements and looking at how they impact the time slowdown. I designed the scenario in a very modular way using groups and having each group completely autonomous. You have then above a group management system that will act and coordinate them. By designing this way a mission it becomes very easy to add groups or take them away. Here I will delete groups and measure the impact on the time slowdown factor (TSF). I will take as a reference the 1.89 value.
 
Groups eliminated individually
a) Camera group deleted                     -> TSF = 1.89 -> Improvement 0.0%
b) Trains group deleted                        -> TSF = 1.83 -> Improvement 6.7%
c) Fixed Ground AAA groups deleted -> TSF = 1.88 -> Improvement 0.0%
There are 45 pieces of fixed ground AAA guns. Still remain AAA on trains and on ships or vehicles.

 

d) Truck/Tank convoys deleted           -> TSF = 1.85 -> Improvement 4.5%

 

Combined Groups eliminated
e) Only planes remaining                      -> TSF = 1.73 -> Improvement 18.0%
This means that all active objects have been deleted from the scenario except the 65 planes.
Here we can conclude that trains, ships, truck, and artillery impact the TSF by 20% if extrapolated in full combat.

 

f) (e)+Parked airplanes deleted           -> TSF = 1.50 -> Improvement 43,8%
Here we can extrapolate that the 10 parked planes have contributed alone by 25%.

 

g) Only 50 bombers remaining            -> TSF = 1.36 -> Improvement 59,5%

 

Now we will come back to the original scenario, keep everything, player plane and the four FW190 fighters and only eliminate bomber groups.
To be precise a bomber group comes with all the waypoints, formation commands, attack areas, landing etc...

 

There are 5 bomber groups with 9 planes each and one with five. Lets see:

h) Eliminate 1 group with 9 planes     -> TSF = 1.73 -> Improvement 17,9%
i) Eliminate 1 group+parked planes    -> TSF = 1.50 -> Improvement 43,8%
j) Eliminate 2 groups+parked planes  -> TSF = 1.26 -> Improvement 70,7%
k) Eliminate 3 groups+parked planes -> TSF = 1.00 -> Improvement 100%
This last case has seen the elimination of 37 planes out of 65.

 

Conclusion:
(A) Airplanes take by far the most of the resources. We are at no CPU impact or slowdown with 28 Flying airplanes and all the rest.
(B) If you do not have too many airplanes, you can then add a lot of activity on the ground, animate the cities, rivers, ports and airfields. It will have a limited impact overall.

 

The ME does support 65 but also more planes. The GPU will cope the CPU not, and the impact is more slowdown. 
I have to give credit here to Gambit 21 if we consider playability. He mentioned a limit of about 25 planes which is pretty correct if we want to have a scenario fully playable.
The data above is valid for my system which is already a powerful one. For some of us with less powerful system the threshold maybe around 15 planes.

 

Hope this helps designing future scenarios.

 

  • Thanks 2

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×
×
  • Create New...