Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[SAS~Storebror]

Hi Chaps,

Long time I've planned to investigate on the impact of different settings concerning the achievable FPS in IL-2 Great Battles Series.
Starting point is the tools, track and setup listed here: https://forum.il2sturmovik.com/topic/29322-measuring-rig-performance-common-baseline

I'm not using VR, the tools, track and setup are valid for comparable results nevertheless.

This is my system for reference:
CPU = Intel Core i5 2500K @4.4GHz
GPU = Nvidia GTX 970 4GB DDR5 (Palit Jetstream), default clocks
Memory = 16GB DDR3-1600 (4x4GB, dual channel)
Display =  2560x1440 144Hz G-Sync (Acer Predator XB271HUbmiprz), display port
HDD = 256GB SSD (Sandisk)

These are the IL-2 Great Battles Settings I'm using per default:
General Setting = Ultra
Screen Resolution = 2560x1440
Full screen = On
Enable VR HMD = Off
Multi GPU Support = Off
VSync = Off
SSAO = On
HDR = On
Sharpen = Off
Shadows Quality = Ultra
Mirrors = Complex
Distant landscape detail = x4
Landscape filter = Off
Grass quality = Ultra
Target FPS = 144
Dynamic Resolution factor = Full
Antialiasing = 4
Gamma correction = 1

These are the Nvidia Control Panel Settings:
Driver Version = 390.65
DSR Factors = Off
Anisotropic filtering = Application-controlled
Antialiasing - FXAA = Off
Antialiasing - Setting = Application-controlled
Antialiasing - Gamma correction = On
Antialiasing - Mode = Application-controlled
Antialiasing - Transparency = Off
Preferred Refresh rate = Highest Available
CUDA - GPUs = All
Triple buffering = Off
Power management mode = Prefer Maximum Performance
Optimize for compute performance = Off
Maximum pre-rendered frames = 2
Monitor-technology = G-SYNC
Multi-Frame Sampled AA (MFAA) = Off
OpenGL-rendering GPU = Geforce GTX 970
Shadercache = On
Texture filtering - Anisotropic sample optimization = Off
Texture filtering - Negative LOD bias = Clamp
Texture filtering - Quality = High Quality
Texture filtering - Trilinear Optimization = On
Threaded optimization = Off
Vertical sync = Fast
Pre-rendered VR-Frames = 1

Now comes the test summary. Details will follow at a later date.
You will find a "Reference" (7th lowest value, therefore dot 7 from the right) which is exactly the above mentioned settings.
Plotted are average framerates during the test (see test link at the beginning of this post).
Each test was done with one single setting changed only. No other applications were active. Track has been played once before (to keep files in cache). Test procedure was start IL-2 -> load track -> wait for object load to finish (HDD idle) -> hide GUI (expect for "GUI On" test) -> run track.
This is the plot (click for full size):


Original Link to the plot here.

Some results were to be expected, others were rather surprising to me.

@edit: Detailed Test results as follows.
Overview of Frame Rates during test:

Direct Link to plot.ly graph here.

Shadows = High vs. Reference


SSAO Off vs. Reference


HDR Off vs. Reference


Antialiasing = 2 vs. Reference


NV CP Anisotropic Sample Optimization = On vs. Reference


NV CP Trilinear Optimization = Off vs. Reference


No Meltdown Patch installed vs. Reference


Grass Quality = Distant vs. Reference


Distant Landscape Detail = x3 vs. Reference


NV CP Threaded Optimization = Auto vs. Reference


NV CP Negative LOD BIAS = Allow vs. Reference


NV CP Transparent AA = Multisample vs. Reference


Power Plan = Maximum Performance vs. Reference


Mirrors = Medium vs. Reference


NV CP Texture Filtering = Quality vs. Reference


GUI On vs. Reference


CPU Clock 3.7 GHz vs. Reference


Sharpen On vs. Reference


NV CP MFAA = On vs. Reference


Landscape Filter = Sharp vs. Reference


NV CP FXAA = On vs. Reference


NV CP Anisotropic Filtering = 16x vs. Reference


...and a few combined results:
Shadows = High + NV CP Anisotropic Sample Optimization = On + NV CP Trilinear Optimization = Off + NV CP Threaded Optimization = Auto + NV CP Negative LOD BIAS = Allow + VSync = On:


NV CP Anisotropic Sample Optimization = On + NV CP Trilinear Optimization = Off + NV CP Threaded Optimization = Auto + NV CP Negative LOD BIAS = Allow + VSync = On:


A few relevant good and bad FPS impact results:


Cheers!
Mike

[Cloyd]

Yes, interesting. Thanks for the comprehensive analysis. I haven't done any benchmarking in BoX for my system, but this helps to focus the effort. I hardly understand what all of the options mean. Seems like if FPS becomes an issue, then start by decreasing AA and shadows. (But I love those smooth edges and sharp shadows.  ) The rest of the options don't look like they have too much impact.

My system is old and mid-range, but still capable:
i7-4770K 3.5GHz Quad-Core Processor
Gigabyte GA-Z87X-UD4H ATX LGA1150 Motherboard
Corsair Vengeance LP 16GB (2 x 8GB) DDR3-1600 Memory
Gigabyte GeForce GTX 1060 6GB Video Card
Sandisk SSD 256 GB
Microsoft Windows 10 Pro - 64-bit

When I get around to testing, I'll report back.

Cloyd

[SAS~Storebror]

Your system is definitely more powerful than mine. The CPU is supposed to be about 10% faster on single threaded programs and the GPU should perform about 15% better.
If you play the game on a FullHD monitor, you should have stable FPS in the 3-digit-range.

Concerning the different options and interesting results, these are a few conclusions:

• Shadows seem to have the biggest impact.
  If you have FPS issues, lower the shadow settings.
  I didn't notice much of a difference between "Ultra" and "High" yet, except for it appears to me that in "High" setting the in-cockpit self-shadowing (e.g. canopy frames) looks slightly jagged instead of crystal-sharp - to me this looks even more realistic.
  This setting alone caused a 17% FPS difference in my test.
• SSAO (10%), HDR (9%) and Antialiasing (8%) are the next "big players", but it's rather interesting to see that Antialising in fact is not the top scorer on this list.
• Using Nvidia Control Panel Settings mostly has a negative effect, as expected.
  Anisotropic filtering setting in NV CP causes 3.5% FPS loss, FXAA another 2.5%. This meets expectations.
• Adding "Multisample" Antialiasing (Nvidia 900/1000 series cards) improves image quality quite a lot, and the FPS loss is negligible (-0.7%).
  This meets the Nvidia Specs, yet I'm still surprised to see how well MFAA performs.
• CPU speed by far doesn't matter as much as expected. Clocking the CPU 15% lower causes the FPS not to change notably (-0.2%, that's within measuring tolerance)
• The GUI itself ("H" key) is no FPS hog. The Minimap is, but that one is not available in Track playback so cannot be measured here.
• Some Myths from the il2sturmovik.com forum have been exposed:
  * Using "Maximum Performance" Power Plan setting doesn't help for Client Games (it does for Dedicated Servers only)
  * Using "Clamp" Setting for Negative LOD bias in Nvidia Control Panel doesn't improve performance, it's vice versa. And even more, with "Allow" the image quality potentially improves.
  * Turning off Threaded Optimization in Nvidia Control Panel doesn't improve performance, it's vice versa.
  * Nvidia's Trilinear Optimization is supposed to improve Performance, but it does the opposite. Turn it off!
• The "Anisotropic Sample Optimization" in Nvidia Control Panel, an option never mentioned anywhere in conjunction with IL-2 so far, causes a considerable 2% FPS gain when turned on. Makes sense to use it.
• Microsoft's "Meltdown" Patch (that's the January Security Patch) for Windows 10 eats about 2% FPS. That's bad, but not as bad as I was afraid of.

Cheers!
Mike

[Mitthrawnuruodo]

One thing to note is that the effects of settings strongly depend on the particular limit that is being hit. In this case it seems that the GPU is affecting your performance.

In some other situations, such as VR, the CPU becomes more important. In these cases, overclocks to around 5 GHz are considered optimal.

It certainly isn't as easy as 1946, which can run above 4K 60 Hz on a potato.

[SAS~Storebror]

Yes of course, if the CPU is the limiting factor, overclocking will have much greater effect.
On a mediocre system without VR though, the CPU will most likely not be that limiting factor.
Let alone other means of overclocking, like RAM OC, which was reported to cause a 25% FPS increase on the il2sturmovik.com boards once - ludicrous.

Cheers!
Mike

[Blaubaer]

Thanks, Mike. This gives a good orientation.

Regards, Michael

[sniperton]

Many thanks, Mike, really valuable info, and should be stickied IMO

[SAS~Storebror]

Detail FPS Test Graphs and link to original Plot have been added to first post.

Cheers!
Mike

[SAS~Storebror]

More Test results (combined settings) added to first post.
It's interesting to note that not all "good" settings for FPS, meaning those which increase FPS when applied standalone, also give the same good results when applied together.
Once again, the biggest impact if you are "GPU limited" like me, is the shadow quality.
Add to that, if you have a G-Sync monitor like me, it appears that the "Fast" VSync setting is not optimal, use "On" instead.

For those who are concerned about shadow quality when you reduce the setting from "Ultra" to "High", this image shows the difference in Cockpit and External view.
Left is "Ultra", right is "High". Click for full size.
If you can spot any difference, be my guest and let me know. I can't.


Maybe the difference between "Ultra" and "High" shadows is something different, e.g. whether or not smokes draw shadows, but it doesn't seem to be a quality thing so for me personally the sweet spot for shadow setting is "high" for the time being.

Cheers!
Mike

[Mitthrawnuruodo]

It will be interesting to see the effects of the upcoming updates.

I suspect that rain on windscreen and increased distant landscape rendering must have some performance impact. Perhaps this will be offset by other optimizations.

Nevertheless, it is an exciting time for FPS analysis. With these crazy graphics card prices, "Just buy a 1070" is no longer a good answer.

[Blaubaer]

Would it be worthwhile to install more RAM? My config:
OS
   Windows 10 Home 64-bit
CPU
   Intel Core i5 4590 @ 3.30GHz
   Haswell 22nm Technologie
RAM
   8,00GB (2x4GB) dual channel DDR3 @ 800MHz (PC3-12800J, 10-10-10-30)
Motherboard
   MSI Z97 PC Mate(MS-7850) (SOCKET 0)
Graphics
   C27FG7x (1920x1080@144Hz)
   4095 MBNVIDIA GeForce GTX 970 (CardExpert Technology)
HDD
   465GB Samsung SSD 850 EVO 500GB (SSD)
Audio
   ASUS Xonar DG Audio Device
---
Possible RAM modules:
   DDR3 1066/1333/1600/1866*/2000*/2133*/2200*/2400*/2600*/2666*/2800*/3000*(*OC) MHz
I think I already have the fastest-possible RAM modules onboard. In this case the question would remain whether a total of 16 GB would have a noticeable effect.

Regards, Michael

[SAS~Storebror]

Would it be worthwhile to install more RAM?
(...)
the question would remain whether a total of 16 GB would have a noticeable effect

No and no.
I've got 16GB myself and have had 8GB before, there is no noticeable difference at all for the games we play.

The difference becomes apparent when you are a video editor or a software developer, otherwise that's just waste of time and money.


Mike