Strange FPS behaviour

[TZeer]

Updated: 11.02.2013

 

These FPS tests started mainly due to my curiosity on why my SLI setup was not utilized more then it was.

In this first post I will try to have the latest findings and results at the beginning so people can quickly see what have been done, and maybe help them along to get better frames in this awesome simulation.

 

Some of these results might not be possible to reproduce if you are not using a powerful single GPU config/SLI or have a slow CPU.

 

Issue: Unstable FPS, sudden drops while looking in a certain direction.

If you are experiencing behavior similar to what is written below here together with the screenshots it can be the detail of water. In the screenshots below you can hardly see any difference.

But the FPS speaks for itself. Solution: Reduce water details to low.

 

Example: If I'm using my settings I have FPS at around 80-90 FPS. If I set water detail to high, the FPS can go as low as 35 at very special occasions. Not stable 35, but they jump up and down.

So try and set the water detail to low (You wont see much difference at 12000 feet anyway :smilewink: )

 

Issue: Flying at speeds lower then ~260 knots will cause the CPU and GPU to reduce performance, and therefor loss of FPS.

If you experience your SLI/powerful single card setup not not being used properly this might be the reason. When your A-10C moves slower then ~260 knots it generates a heat blur at the exhaust of the engines.

For some reason when this happens the GPU and CPU cut down in performance. In extreme cases you can go from 120-130 FPS and down to 60 FPS for no other reason then your speed went under ~260 knots. An effect you normally don't look at unless you turn your head 180 degrees. It's not very often the A-10C moves at >260 knots with full tank and payload, so you might not have noticed it. Since it's been there all the time. But it really kills the performance on your SLI setup.

Solution: Take off heat blur effect

 

Issue: Activating flood light in cockpit reduces FPS by ~20-50 FPS

If you turn on the green light that floods the cockpit, it will reduce your FPS with up to ~50 FPS. Depending on your setup.

No solution yet

 

Here's my current settings. With this I'm able to fly over Tiblisi with Kuky's high.lua mod and rocking 80-90FPS in daylight with no green cockpit flood lights :pilotfly:

 

Link to Kuky's mod: http://forums.eagle.ru/showthread.php?t=71639

 

Textures: High

Scenes: High

Civ Traffic: Off

Water: Low

Visible Range: High

Heat Blur: Off

Shadows: High

Resolution: 2560x1440

Aspect Ratio: 16:9

Monitors: 1 Screen

Res. Of Cockpit Displays: 512

MSAA: 2x

HDR: Normal

 

Clutter/Bushes: 0 m (Havent tried increasing it yet)

Trees Visibility: 15000 m

Preload Radius: 150000 m

 

Disable Aero Interface: Check

Tree Shadows: Uncheck

Vsync: Unchecked (For testing, normally on)

Full Screen: Check

Cockpit Shadows: Uncheck (Not tested yet)

TSSAA: Uncheck (Not tested yet)

 

My specs: i7 3820@ 4,6GHz - 580 GTX SLI - SSD - 16GB RAM - Win 7

 

 

I will not remove or edit any of the posts I have made in this thread so you can read along if you fancy. All updates and findings I do I'll put in the beginning of this post.

 

****************************************************************************************************************************************************************************************************************************************************************************************************************

 

 

Been messing around a little bit and trying to find out why the FPS is behaving so strangely in DCS. So far I have no clue, hehe.

 

This is the simple test I did.

 

1 unit in the middle of Tiblisi. Zoom out and pan around and monitor the FPS while having nVidia inspector and task manager up to monitor the behavior of my hardware.

 

For this test I had vsync off.

 

First screenshot. FPS at 57. Everything maxed out. Looks good.

 

 

 

Next screenshot, and we get 46 FPS, a loss of 11 FPS. Whats different? All I did was change the angle a little bit. If I had moved it slightly more the FPS would go up again. So this is just in a very small narrow place I get this behavior. What gives?

 

 

 

Here I reduced view distance to medium. FPS at 134. Really good.

 

 

 

Again, change angle slightly, and we are down to 102 FPS. Thats a loss of 32 FPS just by changing the angle a tiny bit. Same thing here as well, if I moved it a little notch further, and the FPS would go up again.

 

 

 

 

 

During this, the CPU load never changed. Average CPU load was at 15% max. Max load on one single core was maybe 40%.

 

GPU load did change. But when I got those drops in FPS, the GPU load went down. Which would be normal if the CPU was the bottleneck. But I had no change in CPU load. It was constant.

I have done other tests where I have seen CPU load go up while the GPU load have gone down. And it's a nobrainer that it's the CPU that is the bottleneck in those scenarios.

 

But this behavior here, leaves me a little puzzled.

 

Anyone have any good ideas?

Edited February 11, 2013 by TZeer

[TZeer]

Ok, some more info.

 

If I turn my camera 180 degrees and look towards the airfield. My CPU average usage goes up to 20-25% and peaked at 29%, and my GPU usage went up to 100%. FPS @ 120+. All settings on high.

 

So this confirms that both the GPU and CPU usage is going down for some strange reason.

When CPU and GPU usage goes down, I also start noticing random lines, flickering on the screen. Just a fraction of a second.

Which is strange. Cause it goes away when the GPU and CPU usage goes up.

[Wolf Rider]

Here's a fun one - get on the ground at an airport and, on foot (ie first person view), follow the path of the lowest fps in through town....

Edited February 5, 2013 by Wolf Rider

[TZeer]

More testing done :)

 

Scenes, Textures, Visible Range, Shadows....

 

No change in behavior. Got more FPS. But the drop in FPS was still big. With Visible Range at low I actually got a drop of 40 FPS!!! Doing the same thing.

 

Here is the funny part:

 

Setting water to low reduced the drop considerable!!

 

With 2560x1440 resolution everything on high, I got sudden drops at certain angles that took my FPS down to about 40 FPS.

 

Reducing water to low, I'm constantly above 60 FPS. But the effect is still there.

 

Turn my camera 180 degrees toward the airfield. And my CPU AND GPU usage goes up and I have over 120 FPS.

 

Something is messed up when it comes to how water or whatever is handled.

 

Also the flickering lines that suddenly came when the FPS dropped is gone as well.

 

 

 

What hardware resource is used when water is displayed? What bottlenecking the performance?

[TZeer]

More screens and more strange FPS behavior.

 

Flying level, nothing fancy going on. Flat and boring landscape, 87 FPS

 

 

Here, almost the same place, just a few seconds before, 58 FPS

 

 

A new one, nothing fancy going on. Just large space with nothing on. 52 FPS

 

 

And the last one, behold, we actually got a small town/city in right corner. What's this?

96 FPS??

 

 

 

 

Whats up with this? LOL

[Manuel]

I think fps is affected by how much the cpu has to calculate. You can not say what you see is what was calculated. So maybe there is a difference. By the way. Sys Specs?

[TZeer]

I agree with you. If the CPU is no able to keep up, the FPS will be hit.

 

But reason I made this post is that the CPU usage is not going up. It's actually going down, or not moving at all.

 

And when I move my camera, and get those 100+ FPS, the CPU usage goes up as well. The GPU is at 100% usage. Turn the view around again, and BAM!! Average CPU usage goes down with almost 10%, GPU usage is down to 60% and FPS is down to below 50 FPS....

 

 

All the test where done on a clean map in mission editor. So nothing extra "behind the scenes" logic that's being calculated. Everything from OS to drivers and software is a clean install.

 

My specs:

 

2x 580GTX in SLI

i7 3820@ 4,6GHz

16GB Ram

Win 7 on separate SSD drive

DCS on separate SSD drive

 

 

It's like everyone has accepted that DCS is such a monster to run for the CPU, that no one actually questions when you see some sudden FPS drop.

 

"Oh, yeah, it's your CPU being a bottleneck"

 

 

I have monitored everything from CPU average usage, single core usage, temperature on every core, GPU usage, memory usage, VRAM usage. And there is nothing, nothing at all that is pointing in the direction of the CPU getting a heavier load.

 

Is there something deeper in the hardware that is getting clogged?

 

Some really bad coding?

 

Artifacts/bugs in the graphical engine?

 

I would guess ED had some sort of debugging/raw code/monitoring tool to see what is actually going on in the background in scenes like this? What is reducing the performance? What makes the simulator drop by 40 FPS by just changing the angle slightly on how you view the world?

Edited February 5, 2013 by TZeer

[TZeer]

More data incomming!! :)

 

So I disabled my SLI and did the same tests with just 1 x 580 GTX running.

 

While looking straight down and at an area my FPS was ~120 FPS.

When I looked towards the horizon with the city, my FPS went down to ~55 FPS.

 

Now the interesting part. When I was looking straight down, my CPU usage was close to 100%, Meaning my FPS was held back by what the CPU was able handle. As I changed my view making more and more of the horizon and city visible. The CPU usage suddenly took a dump. But the GPU usage was still @ 100%

 

At the same time the FPS started going down as more and more of the city came into view. 100, 90, 80 and at the end ~55 FPS. During that time the CPU usage did not change. That had already dropped like a brick when the FPS stated dropping in the beginning.

 

 

 

Adding a second 580 GTX doesn't do much. It takes my FPS up to roughly above 60 FPS. A gain of less then 10 FPS.

 

The other interesting part is that when running SLI, the GPU's also reduce load during this test (60% on each). So they should in theory have more power to do stuff. I also have not run out of VRAM. And as we can see on the CPU usage, there is plenty of CPU to play with.

 

But adding 2xMSAA drops my FPS with a small amount. Why? There is more GPU to play with, there's CPU and there's plenty of VRAM.

 

 

My thoughts right now:

 

It's actually the lack of power/juice/whatever in the graphic card. Something that's not possible to measure by pure GPU or VRAM usage. I also suspect it's something that's shared in an SLI setup, since adding a second card doesn't do much. Kinda like VRAM. 1,5GB+1,5GB = 1,5GB VRAM

 

 

When some new cards are getting released by AMD or NVIDIA, I'll be able to test this out.

[TZeer]

Update:

 

After the last test I decided to test my theory.

 

My CPU is normally overclocked to 4,6 GHz. If it is my CPU that is holding the performance back, my FPS should drop even more if I took away my overclock. If it's my GPU I should see no difference.

 

I reduced my CPU down to 3,7GHz. Repeated what I have done earlier.

 

And guess what? NO change in FPS. Nothing at all. Horizon with a full city, 55 FPS stable.

 

So a reduction at almost 1GHz on my CPU had no difference at all on my FPS.

[Sov13t]

I just set an FPS cap of 30 for my 5860x1080 set up to avoid noticing huge drops in FPS. Going from 120 to 50 can be rather noticeable.

 

Running a GTX 670 4GB and 5860x1080.

[TZeer]

I just set an FPS cap of 30 for my 5860x1080 set up to avoid noticing huge drops in FPS. Going from 120 to 50 can be rather noticeable.

 

Running a GTX 670 4GB and 5860x1080.

 

I tried that, but it's only ok as long as I look straight. Look to the side and it's slideshow imo.

 

Reason I do this is to find out if it's the CPU or GPU that is the bottleneck for the FPS drops.

 

Normally people are saying it's the CPU. But all my tests says otherwise. Even though my SLI setup is not being used at 100% during the test, there seem to be something in the design of the card that is clogged up and making it impossible to increase the FPS.

 

My goal is to get stuff running at 60 FPS so I can run with vsync on.

 

Been able to do a test with a friend of mine, similar setups. But he had a GTX 690. He used my mission file, and we had identical options selected in the options.

 

And he had over 90 FPS at the lowest. But when FPS dropped like it did for me, his GPU usage went down as well.

 

This just confirms that it's actually the GPU that is whats holding back the performance, even though you wouldn't think it, when you look at the GPU load alone.

[SkateZilla]

shrug... it has me stumped....

[Sov13t]

This just confirms that it's actually the GPU that is whats holding back the performance, even though you wouldn't think it, when you look at the GPU load alone.

 

Don't forget that DCS engine is not the most optimized either. There comes a point where no matter how much you throw at it, it will have diminishing returns.

[Lange_666]

Been able to do a test with a friend of mine, similar setups. But he had a GTX 690. He used my mission file, and we had identical options selected in the options.

 

And he had over 90 FPS at the lowest. But when FPS dropped like it did for me, his GPU usage went down as well.

 

It would be nice if you could drop his GTX690 card into your machine and run the same test.

[TZeer]

Could be hard, since he's living in another country :)

 

But I will replace my SLI setup with a more powerful single card setup when amd/nvidia release them.

 

But so far all the tests I have done points toward that it's actually the GPU that is causing the major FPS drops. Either by some design/component in the cards that get utilized 100% that we can't monitor, or just plain horrible code inside the graphic engine.

 

 

Here's another funny little thing I found as well. I don't now how many of you who run kuky's high lua mod for the rendering distance on districts. Anyway, I do.

 

And when I replaced his high lua file, with the normal high lua file, my FPS didn't change. No increase in FPS. Which is strange, since I basically cut the distance from 20km and down to 10km....

[Lange_666]

Interesting topic this one.

I never did check the load on either the CPU or GPU while in game and took for granted that the sim is CPU hunger. Your test proves this is wrong.

The only thing i always wondered was why the GPU fan hit the roof every time the game is run. I just ran a little program a few times to see what happened to the temp of the GPU and it skyrocketed every time, hence the fan rpm going to max, which means my card (GTX480) runs at max when the game is running.

This made me think of doing a GPU upgrade only but until i'm 100% certain, i won't go that route. I prefer to do a full upgrade at once which i postponed a bit waiting to the next generation of CPU's/GPU's.

[TZeer]

Would be nice if ED could take a peek at this as well.

 

Cause to be honest, this was not what I expected to find. But I'm a little happy as well, since GPU's are evolving faster then the CPU's at the moment. And with AMD and NVIDIA coming with a new generation. I'm gonna jump straight from GTX 580 and up to whatever they gonna cal them and get a nice boost.

[Lange_666]

Another thing I'm interested in is how much difference does the amount of VRam make like between a 2GB card and 4GB card.

Since I'm following this topic, I'm eyeballing a GTX670 card as a replacement for the GTX480 (1,5GB) I have now.

[Vlerkies]

My CPU seems to idle or only run off one core, need to check more. i7 3770k

 

Lange, I have a 4gb 680gtx, most it uses is like 1.3-1.5 on a quick short test. (edit: long night mission now for a few hours 1.8 )

 

GPU load though is the only parameter that pushes up towards 100%

Edited February 9, 2013 by Vlerkies

[Vlerkies]

This just confirms that it's actually the GPU that is whats holding back the performance, even though you wouldn't think it, when you look at the GPU load alone.

 

I agree with that!

[Vlerkies]

Similar discussion with some mod input.

 

http://forums.eagle.ru/showthread.php?p=1654466

[Rotorhead]

While looking straight down and at an area my FPS was ~120 FPS.

When I looked towards the horizon with the city, my FPS went down to ~55 FPS.

 

Now the interesting part. When I was looking straight down, my CPU usage was close to 100%, Meaning my FPS was held back by what the CPU was able handle. As I changed my view making more and more of the horizon and city visible. The CPU usage suddenly took a dump. But the GPU usage was still @ 100%

 

I've experienced similar issue on my old system with a low-end GPU. I was able to achieve around 60 FPS when looking down, but no more than 14 looking towards the horizon, no matter how low settings I used.

 

My GPU load wasn't anywhere near 100% too. Not sure about CPU load though.

 

What is interesting, this issue came with DCSW 1.2.0. In previous beta release, I was able to maintain between 25-30 FPS at all times except huge battles, even with that poor system. AFAIK, nobody came up with some serious explaination why FPS was affected so badly on some systems by introducing 1.2.0.

[TZeer]

I have a few more tests to do. I'm gonna see if I can narrow it down even further.

 

So far I have ruled out the CPU as the reason for my FPS drop.

 

Have a fellow squadron member that have almost identical system as me, i7 3820. But instead of GTX 580, he is running GTX 680. Since this is a Sandy Bridge-E CPU we are still running

PCI-E 2.0 and not PCI-E 3.0 as Ivy Bridge. This way we can at least rule out the PCI-E lanes as the reason.

[FatSlapper]

For over 2 years I have been wondering why the frame rates are higher when You RETURN to Batumi. Keep up the good beta testing.....

[TZeer]

Update:

 

Have not been able to do the PCI-E 2.0 vs 3.0 test. But found some other interesting things.

 

Came back today to continue my testing, just to suddenly not have any GPU drop in usage while tilting my view....??? This was quite interesting. Specially since my FPS was constantly above 90 FPS.

 

Started a SLI vs Non-SLI test at different resolutions:

 

Lowest recorded FPS

 

1024x768 Non-SLI ~100 FPS SLI ~103FPS

1440x900 Non-SLI ~84FPS SLI ~103FPS

1920x1080 Non-SLI ~68FPS SLI ~95FPS

2560x1440 NON-SLI ~55FPS SLI ~95 FPS

 

 

Important to note on the above tests. None of the cores on the CPU where ever taxed to 100%. Maximum average CPU usage was 29%.

I must say, this is a lot better then my previous tests.

 

But I was still curious why my FPS was suddenly so much better then the other day.

 

Decided to fly around for a bit when suddenly out of nowhere my GPU/CPU usage and FPS dropped again. After a while they came back up again.

 

Reason for the FPS drops turned out to be me pulling a tight turn!!!! YES!!

I have not yet found out how to reset whatever I'm doing so the FPS get's back up again. Put I can now decrease FPS/GPU/CPU usage on command while flying by just doing either sudden pull or a right/left bank with my joystick.

 

Every time I do this, the CPU/GPU/FPS goes down. In my case, I drop form comfortable 80+ FPS in cockpit, to 55 FPS. While flying over Tiblisi. Sometimes the effect resets itself after a short while.

After my last attempt I have not been able to reset it. Only thing left to test is a disable/enable SLI again or restart computer.

 

The following pictures are of my GPU's and CPU during the testing.

 

As you can see, the time for it to reset varies. If you pull a gentle turn nothing happens. If you pull a hard-G turn the drop comes. If you just do a sudden movement on the stick, the drop comes. You can be flying straight and do a sudden left/right bank on the stick, and the drop comes....

 

 

 

 

 

 

Edit:

 

Turned out I had to restart my computer for the reduction in GPU/CPU usage to go away after my last attempt. Was unable to provoke the drop by just pulling or sudden moves on the stick this time. But a high-G turn did the trick. And it doesn't matter if I do it above the city, turn toward the city, or just on a plain open boring place with nothing to see.

Edited February 10, 2013 by TZeer