Jump to content
chiliwili69

Measuring rig performance: Common Baseline (for IL-2 v3.010)

Recommended Posts

I have followed this thread with some interest for quite some time now, and whilst my variables in this experiment were different from those outlined in this threads' initial post, I think you all might be interested in the performance my very low end Rig managed to achieve. with ASW at 45fps and the 3DMigoto Mod for getting rid of the prop disc and asociated artefacts - It is possible to have a very enjoyable VR experience.

 

Ok, here are my findings - Newcomers to IL-2 Great Battles with low end Rigs for VR... Read On!

 

So, I used Steam VR and set up the Application Settings specifically for IL-2, in the section Application Resolution Adjustment, I set it to 100%. It defaults to 66% I believe, which means it is actually rendering about 1080 x 1200 per eye, which is the Oculus Rift's native resolution. 100% on the slider gave me 1330 x 1584 rendered.

 

Then I have the very useful 3DMigoto mod which allows for ASW @ 45fps to be used without visual anomalies.

 

I also used the clouds MOD available here to get high quality clouds in all graphics presets, Balanced / High / Ultra / whatever. They look Great but you need to enable mods in game (tickbox) : 

 

And finally, in the games' Graphics settings I used :

 

Settings - Balanced

Screen Res. - 1920x1080

Shadows - Medium

Mirrors - Medium

Distant Landscape Detail - x4

Horizon Draw Distance - 100 Km

Landscape Filter - Blurred

Grass Quality - Normal

Clouds Quality - High

Target FPS - Off

Antialiasing - 2

Gamma - 0.8

Full Screen

Enable VR HMD

Sharpen

Use 4k Textures

 

In game Head Up Display (HUD) - OFF

 

Now I will post my Rig specs:

 

Intel i5 6400 2.7Ghz

RAM 16Gb DDR 2100Mhz

M.2. SSD 120Gb

GeForce GTX 1060 6Gb

Oculus Rift CV1

Thrustmaster T.16000m

Saitek Throttle Quadrant

 

Now, with these settings and my Rig... I got a steady 40-45 fps in game (in the Oculus Rift with fps counter - *Backspace* and Shift + k for disabling propellor disc in the 3DMigoto mod)

 

The Map was Kuban - Summer, 18:30pm, Cloudy weather (the setting below Overcast), Clouds at 2100m and starting altitude of 3000m in Spitfire Mk Vb. even down low over trees and ridges my fps didn't drop below 40 fps and was quite smooth at all times, not juddery.

 

*Disclaimer* I didn't fly with AI Aircraft in the air. Your mileage may vary!  For now I am only interested in leisure flights, not dogfights.

 

So, if, like me, you have a low end Rig, considered barely able to run VR in IL-2 B.O.K, try this setup, programs, Mods and settings and you too can have the WOW! factor of VR Flight

 

Happy Landings!

 

Algy Lacey

Edited by Algy-Lacey
Disclaimer

Share this post


Link to post
Share on other sites

Algy, I've tried running BoX VR on a rig very simillar to yours. And for leisurely flying it does work well. However, as soon as you throw in 3-4 ai aircraft and try a dogfight it starts stuttering. Online multiplayer or with a more complex mission it was far from playable.

 

If you wish you could turn down shadows and mirrors another notch as those in particular are very cpu-hungry.

 

Imho those with low end rigs (below rift min recommendation), that get an hmd to enjoy Box in VR, will be very disappointed unless their just after sightseeing. 

  • Thanks 1

Share this post


Link to post
Share on other sites
2 hours ago, SvAF/F16_radek said:

Algy, I've tried running BoX VR on a rig very simillar to yours. And for leisurely flying it does work well. However, as soon as you throw in 3-4 ai aircraft and try a dogfight it starts stuttering. Online multiplayer or with a more complex mission it was far from playable.

 

If you wish you could turn down shadows and mirrors another notch as those in particular are very cpu-hungry.

 

Imho those with low end rigs (below rift min recommendation), that get an hmd to enjoy Box in VR, will be very disappointed unless their just after sightseeing. 

 

Hey Radek, thanks for the sensible feedback.

 

I will try throwing in some friendly and enemy AI and see what the difference in fps is on the same settings. If it's too low or stuttery, I will try turning down Shadows and Mirrors.

I thought that my settings would be able to handle small amounts of AI aircraft, we shall see...

 

My Rig is just above Oculus' minimum Rift recommendation. I believe that their lowest spec Graphics card for VR is a GTX 960.

 

One question,

Seeing as online: the user's FM, DM etc is modelled on his/her PC, and the aircraft are simulated one per PC, why would playing online in VR hit my framerates? I am guessing that the CPU load should be similar to offline with a couple of AI aircraft unless the server is hosting lots of AI aircraft (eg. Bombers). Can you clarify please?

 

Thanks

Algy

Share this post


Link to post
Share on other sites
2 minutes ago, Algy-Lacey said:

Can you clarify please?


I cant I'm afraid. Taxiing from a busy airfield online can bring down a ~5ghz cpu to 45fps, with dips below that. If that same field is empty I can keep an easy 90. Perhaps what you receive from others is actually controller input that gets interpolated rather than aircraft energy/orientation. Just a guess but would explain some. Like you I don't think the added drawcalls of a few aircraft would have such a huge impact.
Also, for many and including me, multiplayer is a rather competitive environment and loosing track of your opponent due to framedrops can be very frustrating. For offline I tend to turn settings up a notch.

  • Thanks 1

Share this post


Link to post
Share on other sites
45 minutes ago, Algy-Lacey said:

 

Hey Radek, thanks for the sensible feedback.

 

I will try throwing in some friendly and enemy AI and see what the difference in fps is on the same settings. If it's too low or stuttery, I will try turning down Shadows and Mirrors.

I thought that my settings would be able to handle small amounts of AI aircraft, we shall see...

 

My Rig is just above Oculus' minimum Rift recommendation. I believe that their lowest spec Graphics card for VR is a GTX 960.

 

One question,

Seeing as online: the user's FM, DM etc is modelled on his/her PC, and the aircraft are simulated one per PC, why would playing online in VR hit my framerates? I am guessing that the CPU load should be similar to offline with a couple of AI aircraft unless the server is hosting lots of AI aircraft (eg. Bombers). Can you clarify please?

 

Thanks

Algy

Your computer is still drawing all those aircraft, increasing load. Not only that but the missions online are filled with ground objects that are run by AI, it would be pertinent to listen to the people who have experience with this instead of constantly questioning everything people here are telling you. I've played nearly 100 hours in VR in the last few months. As have others. We just might know what we are talking about.

  • Sad 1

Share this post


Link to post
Share on other sites
1 hour ago, 15th_JonRedcorn said:

Your computer is still drawing all those aircraft, increasing load. Not only that but the missions online are filled with ground objects that are run by AI, it would be pertinent to listen to the people who have experience with this instead of constantly questioning everything people here are telling you. I've played nearly 100 hours in VR in the last few months. As have others. We just might know what we are talking about.

 

Hey 15th_JonRedcorn,

 

I am not dismissing other peoples' good advice, I simply want to see it for myself... for one, we all have different Rigs with different Settings in IL-2 and second, I have usually phrased things like "I thought... (such and such)", "I'm guessing..."  "I assume", "I believe", "I expect" etc.

 

I am learning a lot here, and my questioning phrases are a way for me to learn things from more experienced forum members whilst narrowing it down to the issue at hand (for me).

It's clear that people here know much more than myself. "We just might know what we are talking about" - I don't doubt it :salute:

 

1 hour ago, SvAF/F16_radek said:

Like you I don't think the added drawcalls of a few aircraft would have such a huge impact.
Also, for many and including me, multiplayer is a rather competitive environment and loosing track of your opponent due to framedrops can be very frustrating. For offline I tend to turn settings up a notch.

 

Thanks Radek. If I try multiplayer I will turn down some settings and find my own 'sweet spot'.

For now, I am enjoying leisurely pleasure flights over the beautiful Kuban map.

 

Cheers

Algy

Edited by Algy-Lacey

Share this post


Link to post
Share on other sites

I flew my first ever combat quick mission in IL-2 Great Battles last night, with the fps counter on and these settings:

 

Settings - Balanced

Screen Res. - 1920x1080

Shadows - Medium

Mirrors - Medium

Distant Landscape Detail - x4

Horizon Draw Distance - 100 Km

Landscape Filter - Blurred

Grass Quality - Normal

Clouds Quality - High

Target FPS - Off

Antialiasing - 2

Gamma - 0.8

Full Screen

Enable VR HMD

Sharpen

Use 4k Textures

 

In game Head Up Display (HUD) - OFF

 

Mods included the 3DMigoto mod by Lefuneste to get rid of the propellor disc and associated artefacts...

and the High Quality Clouds in Low Preset mod.

 

My Rig:

 

Intel i5 6400 2.7Ghz

RAM 16Gb DDR 2100Mhz

M.2. SSD 120Gb

GeForce GTX 1060 6Gb

Oculus Rift CV1

Thrustmaster T.16000m

Saitek Throttle Quadrant

 

The bombers eluded us, but I shot down both enemy fighters with limited ammo on, first victories for me!

 

I flew in formation with a wingman plus another pair, all Spitfires... vs 4 Ju87's and 2 Me109F2's as escort. Kuban Summer map, starting altitude of 2000 and somethin metres, 7am with cloudy preset. My frame rate didn't drop below 40-45 fps which was the same framerate as I got during leisure flights on the Kuban map. I experienced occasional stutters, but nothing that ruined the experience. So that's 9 AI aircraft and no reduction in fps. Spotting was fairly easy with my settings, kudos to 15th_JonRedcorn and SCG_Fenris_Wolf for their advice, especially in the thread https://forum.il2sturmovik.com/topic/34266-copypaste-settings-for-best-spotting-rift-and-now-odyssey/

 

I am very pleased with the outcomes of my testing and would say it bodes well for those who want to fly VR with a low specced Rig.

 

I want to say a big thankyou to all those who helped and guided me to find the best settings for my Rig, and thanks to those who doubted the performance potential and said it wouldn't work - if they hadn't spoken up I would've been less motivated to find a solution.

 

Cheers,

Algy

Edited by Algy-Lacey
  • Like 1

Share this post


Link to post
Share on other sites

Glad to hear those setting give you a pleasant experience with this low end CPU/GPU combination in the demanding Kuban map.

I assume you were using just 100% supersampling in SteamVR ¿right?

BTW, could you also report your Mobo model?

 

There is a big CPU/GPU requirement from running at 90fps versus 45 fps. So, Oculus ASW and lefuneste Mod give the trick to you.

I have never used the Mod but  with a low end rig I would give it a try.

Thanks for sharing your results.

  • Like 1

Share this post


Link to post
Share on other sites
1 hour ago, chiliwili69 said:

Glad to hear those setting give you a pleasant experience with this low end CPU/GPU combination in the demanding Kuban map.

I assume you were using just 100% supersampling in SteamVR ¿right?

BTW, could you also report your Mobo model?

 

There is a big CPU/GPU requirement from running at 90fps versus 45 fps. So, Oculus ASW and lefuneste Mod give the trick to you.

I have never used the Mod but  with a low end rig I would give it a try.

Thanks for sharing your results.

 

Hi chiliwili69,

I have an Acer Predator G3-710 PC with an upgraded gfx card... but I can't find information on the Motherboard anywhere! Sorry

I know that the CPU uses Intel socket 1151 and that an i5 6700k is a potential upgrade path.

 

I used Steam VR and set up the Application Settings specifically for IL-2, in the section Application Resolution Adjustment, I set it to 100%. It defaults to 66% I believe, which means it is actually rendering about 1080 x 1200 per eye, which is the Oculus Rift's native resolution. 100% on the slider gave me 1330 x 1584 rendered.

 

I agree that there is a big CPU/GPU requirement to run Oculus Rift or any HMD at 90fps, as essentially the graphics card has to process 180fps, which is unacheivable for most systems. But here I have established that even a lowly i5 @ 2.7Ghz and a GTX 1060 6Gb When using the 3DMigoto mod and tweaked clouds is able to run smoothly at 40-45fps with 9 AI Aircraft plus your aircraft, on the Kuban Map, and with game settings which are far higher/more detailed than some would expect.

 

My next test in VR will be to visit an online server when it's busy and see if my setup can cope with that!

 

I hope that the experiments I have done will encourage more customers who are interested in VR to take the plunge, if they have a Rig equal to or better than mine.

 

Glad to share my results 😄

 

Algy

Edited by Algy-Lacey
Online Multiplayer next

Share this post


Link to post
Share on other sites
13 hours ago, Algy-Lacey said:

but I can't find information on the Motherboard anywhere!

 With the free tool CPU-Z you can view all specs and manufacturer, including Mobo, of your PC. With that you can go to the manufacturer and get more info about how to configure the BIOS from the manufacturer page.

Share this post


Link to post
Share on other sites
On 8/17/2018 at 11:14 AM, chiliwili69 said:

 With the free tool CPU-Z you can view all specs and manufacturer, including Mobo, of your PC. With that you can go to the manufacturer and get more info about how to configure the BIOS from the manufacturer page.

 

Thanks, that's some really helpful information! I'll check it out soon :salute:

 

I have checked every option in the BIOS and cannot find a way of increasing Processor Speed from stock 2.7 to 3.3Ghz, perhaps a BIOS update might help. Otherwise, I have heard that Intel offers a program whereby you can switch on Turbo Boost (3.3Ghz). I will report back when I have time to download / use CPU-Z

 

Algy

Edited by Algy-Lacey
BIOS

Share this post


Link to post
Share on other sites

Hey guys, one person told that the test flight record didn't work. I have tried to run it and it doesn't work anymore. Perhaps the last built is not compatible with previous flights.

 

i think it is good to have a common test flight for IL2. Ideally the game could incorporate one test flight just to tweaking performance and for troubleshouting issues.

Meanwhile our only option is to create it by ourselves like we did in this thread.

 

I can ask again for a volunteer to create a test flight with the following requisites.

Duration: no more than 2 minutes. Ideally just 100 seconds.

Scene: Ideally a place that everybody could run (independently of BOS, BOK, BOM, BODP). For example Lapino. The problem with lapino is that it is too flat, ideally there should be mountains

Plane: a common plane  fighter for BOS, BOM, BOK, BODP. 

Wheater: With clouds.

Complexity: about 6-8 planes around. Some bullets and explosions, some fire and smoke but not too much.

 

About the settings, we can use the recommended settings for VR from Fenris for example. Or what other most used settings we typically use in VR.

Share this post


Link to post
Share on other sites

Hey all, wanted to report some experiences. My rig is an 8086k @ 5.0 ghz with AVX offset -2, and an older GTX 1070 Superclocked card.

 

I couldn't use the test recording, it doesn't load in the latest version of BoS unfortunately. I made my own recording with a 6v6 fight between mostly AI planes, and then I used Chilli's method to benchmark with Fraps. I'm still working on tuning the settings how I like them.

 

I had a hell of a time getting the right Rift pixel density setting. After an awful lot of messing around, here's my conclusion: if you use SteamVR to set SS, then change a graphics setting in IL2 and allow it to restart, SteamVR will "double apply" your SS setting. I had started with a 1.3 SS and after changing a graphical setting it jumped to 1.7. Took me a while to figure this out, so I'm now religiously checking using the Oculus Debug Tool "Layer" option. It seem you must fully shut down both IL2 and SteamVR between graphics setting changes.

 

In my benchmark runs on this rig I've noticed the following FPS penalties:

  • About a 10% FPS hit going from "low" to "medium" shadows
  • About another 10% FPS hit going from "medium" to "high" shadows
  • About a 10% FPS hit going from 130% SS to 170% SS
  • About a 15% FPS hit going from "balanced" to "high" graphics
  • Very little change going from "high" to "medium" clouds
  • Very little change going from 100km draw distance to 70km

The CPU in this rig is pretty strong. The main IL2 thread seems to have about 30-40% headroom on the core that it's running on.

 

At the moment I feel like I'm not getting quite enough FPS to be comfortable in online play. I want at least Medium shadows and High clouds as these are both very distracting to me when they're set lower. I also feel like ASW just doesn't cut it for fast moving stuff so I'm really trying to hit 90. If anyone has a recipe for what they use on their 1070 for online play, please let me know.

Share this post


Link to post
Share on other sites

@Alonzo

 

You might give a try setting the SS in Oculus Debug or via Oculus Tray Tool rather than Steam VR. I personally use and prefer the Oculus Tray Tool for this.

Might also try running with ASW set to off, which you can do with the Oculus Tray Tool as well - just to compare. You can also disable it via CTRL+Numpad 1 I believe.

 

 

 

Share this post


Link to post
Share on other sites

Thanks Alonzo for your input. What map/plane do you run? can you share the track?

 

I have just created a 2 minutes track flight that could be ok for the purpose of measuring performance:

 

https://drive.google.com/open?id=122-cL05mZRUEnEPU5sDOGlZ0e7Alaxrr

 

I will update the original post during the weekend since I am not at home right now. I will check the SteamVR doubling SS. that´s quite surprising.

 

Don´t trust the 30-40% headroom on the core. It is a fake number. try to put SS=1.0 to see if it solves your fps problem. if not then the bottleneck is in CPU or other issues,

Share this post


Link to post
Share on other sites

@chiliwili69 Thanks for the suggestions and for your track. I like it because the lighting effects are more pronounced with the low sun. Mine has a bit more aircraft smoke and stuff like that. You can grab my track file here: https://www.dropbox.com/s/876nu40db5hlgef/Alonzo6v6benchmark3minutes.zip?dl=0

 

The two track files run within ~3 fps average, with yours being a bit more demanding.

 

I tried yours down to 1.0 SS. Here are my results (sorry for the bad cut/paste). I feel like this is close enough to 90 to claim I'm not CPU limited, but I'm not sure.

 

Recording
Chilli 2 minute Spitfire fight against bombers
     
BoK.SpitIX.2bombers.trk
                       
CPU settings
8086k @ 5.0ghz, AVX -2
                               
GPU
GTX 1070 SC
                               
Graphic settings
ASW off
SSAO off HDR off Bloom off Grass normal 4k textures on Gamma 0.9 Dyn res ful                  
                              FPS
Steam SS Pixel density Display Window General graphics Shadows Mirrors Distant Landscape Horizon Draw Distance Landscape Filter Clouds Sharpen Anti aliasing GPU Core OC GPU Mem OC GPU Voltage/Heat Unlocked? Min Max Avg
130% 1.3225 1440x900 Low Medium Simple Normal 100km Blurred High Off Off 110 220 No 46 91 85.4
130% 1.3225 1440x900 Low Medium Simple Normal 100km Blurred High Off Off 110 220 Yes 46 91 86.7
100% 1.01 1440x900 Low Medium Simple Normal 100km Blurred High Off Off 110 220 No 49 91 87.5
100% 1.01 1440x900 Low Medium Simple Normal 100km Blurred High Off Off 110 220 Yes 49 91 87.5

Share this post


Link to post
Share on other sites

Further observations. I've heard "IL2 has poor optimization" (which is a bit judgey, I think!) but I tried to quantify it. I tried to remove the GPU as a source of bottleneck by subsampling instead of supersampling using Oculus Debug Tool. If you set Pixel Density to 0.816 (for example) you're modeling a GPU with about 50% more horsepower. In my case I'm pretending that my 1070 is actually a 1080ti.

 

Here's the fun part. Switch everything down to super low settings, load up a replay, then use Oculus Debug Tool to switch on the "Performance" Visible Hud. In the Performance Hud Mode, set that to "App Render Timing". This shows how much time the application (IL2) is spending before issuing "frame finished" calls to the Oculus SDK.

 

On my 8086k @ 5.0ghz with a -2 AVX offset (so the cores are running at 4.8ghz for IL2) with low everything (and I mean low!) and 0.816 pixel density, IL2 still requires 9-10ms of app render CPU time. Watching Chili's Spitfire/Bomber video, every time a physics thing happened (e.g. he blew the wing off a bomber) the CPU render time went up to about 12ms. If we go above 11ms we're missing our 90fps window, without the GPU even getting involved at all.

 

I then tried Ultra everything. Maxed out all my settings, 4x AA, sharpen filter, all that stuff, but again subsampling on the GPU. On this run the CPU was requiring 14-15ms per frame and so we were stuck at 45fps (effectively missing every other frame). In this run the frame time bounces due to physics (or whatever) were up to about 20ms.

 

This basically corroborates the previous discussion and shows why you see very marginal improvements going from a 1080 to a 1080ti, and why the game responds so well to CPU frequency. Anything we can do to get that 9-10ms of CPU render time to be just 1ms less will help.

 

My current theory is that you want to pick a mixture of settings that minimize CPU render time but still allow the GPU to stretch its legs rendering a better quality image. I am guessing that this will be higher AA, sharpening, and super-sampling, but a lower amount of "things to compute and draw" such as reduced polygon counts, reduced render distance, etc. 

  • Upvote 3

Share this post


Link to post
Share on other sites

Further further observations, in my quest for a rock-solid 90 fps. I have been using a Fraps frame analysis tool to visualize the frame times and check things such as the 0.1% low. In VR every frame matters, so average framerate is less important than the times the framerate goes below 90, and for how long.

  • I don't think 90 fps is achievable, fully, 100% of the time. Game physics (bomber gets wings blown off) consistently cause a spike in frame rates.
  • Disabling the 3DMigoto mod improved my average FPS by 3 (82-85), but you lose the enhanced gunsight and improved zoom (this might be a worthwhile trade off, up to you to decide)
  • Overclocking my RAM from 3200 to 3466 mhz and ensuring the timings stay constant made a big difference, with my 0.1% low improving by 5 fps.
  • Switching off hyper threading on my 6-core 8086k improved 0.1% low by another 4 fps.
  • Increasing AVX clock speed by 200mhz from 4.8 to 5.0 ghz (achievable because I had switched off HT) gave me another 4 fps in the 0.1% low.
  • Increasing my Ring Ratio (speed at which the chip cache and other non-core stuff runs) from 45 to 47 (4500mhz to 4700mhz) improved my 0.1% low by a whopping 7 fps

So my advice to anyone running this thing in VR is that you really do need to overclock the snot out of your CPU and RAM to try to help that main IL2 thread stay below the magic 11ms frame time. If you have an 8700k, consider disabling hyper threading if that helps you increase frequency, pay attention to effective AVX core clock, increase your Ring Ratio (or "uncore" ratio) as much as you can, buy fast RAM or overclock what you've got. If you're in the market for a CPU for this game, a highly clocked 8600k will beat out a lower clocked 8700k, and you might want to wait for the new 9-series of Intel chips because they clock even higher. I'm using a Noctua U14S air cooler and I'm at its limits, especially for Prime95 AVX testing. Consider a 240 or 280mm CLC cooler or the absolute top of line Noctua stuff, the D15 (I think).

 

My tests are all performed on a freshly rebooted system with most background stuff not running (certainly no Discord, for example). I still think the cores have capacity for background tasks, but if you do a lot of background processing you might be better off with HT enabled. 

  • Like 2
  • Thanks 3

Share this post


Link to post
Share on other sites

Well, I have just updated the procedure of the first post to run the new benchmark for 3.005 using the track I made. It is just 2 minutes, not very heavy combat but demanding enough in terms of lights, AI, tracers, damage, village etc but not dropping the fps to 45 too much. Ideally I wanted to keep values between 45 and 90, so it will be more effectively used to measure changes in graphics settings or hardware upgrades. This is the fps I achieved with my rig at 4.8 OC:

 

Frames: 7336 - Time: 120000ms - Avg: 61.133 - Min: 43 - Max: 91

 

fps.thumb.jpg.4600a99443fb9d56feb03f8111786a11.jpg

 

I also modified some graphics settings from previous test, just to be closer to actual setting at game, but I relaxed them a little bit in order to obtain more values between 45-90 and not hitting 45 too much. So we can measure better when we increase to High Clouds or use mirrors. The settings of the new benchmark are:

608330812_VRgraphics.jpg.0a306634e8d4deb4df693b6302c68b76.jpg

Share this post


Link to post
Share on other sites

Hey guys, If you follow this thread it would be helpful your vote:

 

On 9/13/2018 at 12:13 AM, Alonzo said:

It seem you must fully shut down both IL2 and SteamVR between graphics setting changes.

Yes, I always shutdown IL2 and SteamVR when I do a change in SS, just to be sure.

On 9/13/2018 at 1:56 AM, Alonzo said:

I tried yours down to 1.0 SS. Here are my results (sorry for the bad cut/paste). I feel like this is close enough to 90 to claim I'm not CPU limited, but I'm not sure.

 

You tests show that you are not GPU limited, since you almost obtain same performance for SS=100% and SS=130%. You can increase SS until you will see a decay in fps. So you will know the SS limit of your GPU.

 

You are almost not limited in CPU since you are very close to 90. Basically because you run at Low graphics preset. The Low/Balanced/High/Ultra is the main variable to load you CPU.

 

I was doing this in the past to see how SS load GPU and presets load CPU:

https://forum.il2sturmovik.com/topic/31307-how-ss-decreases-your-fps-testing-results/

 

Share this post


Link to post
Share on other sites
On 9/13/2018 at 4:30 PM, Alonzo said:

My current theory is that you want to pick a mixture of settings that minimize CPU render time but still allow the GPU to stretch its legs rendering a better quality image. I am guessing that this will be higher AA, sharpening, and super-sampling, but a lower amount of "things to compute and draw" such as reduced polygon counts, reduced render distance, etc. 

 

Yes, but I think (only think) that all the presets and graphics settings load the CPU. And then only the SS load the GPU.

Knowing the performance penalty and visual gain we can decide what to use in every rig.

We can also know how better rigs perform or how other VR devices perform, and help in the buying process.

19 hours ago, Alonzo said:

I have been using a Fraps frame analysis tool to visualize the frame times and check things such as the 0.1% low

 

What is Fraps frame analysis tool?

What is 0.1% low?

19 hours ago, Alonzo said:

Consider a 240 or 280mm CLC cooler or the absolute top of line Noctua stuff, the D15 (I think).

Yeah!, the 8086K reach 5.0 with just the default turbo. I believe that with better cooling you will hit 5.2 or perhaps 5.3.... If not you can also consider deliding.

19 hours ago, Alonzo said:

Increasing my Ring Ratio (speed at which the chip cache and other non-core stuff runs) from 45 to 47 (4500mhz to 4700mhz) improved my 0.1% low by a whopping 7 fps

 

Good to know that!. Thanks for all these valuable inputs. I need to learn how to set that in my BIOS.

Share this post


Link to post
Share on other sites
10 minutes ago, chiliwili69 said:

What is Fraps frame analysis tool?

What is 0.1% low?

 

The tool is Frahfs Bench Viewer: https://sourceforge.net/projects/frafsbenchview/files/

 

It is from 2014 but it works great. Basically you enable "frame time" benchmarking in FRAPS, do a benchmark run, and then drop the .frametimes.csv file into the viewer. It will plot the exact frame times. In VR you want to see lots of frames around the 11ms time, not many slower. The 0.1% low is the worst-case 0.1% of frames, and tells you how bad the 'worst' stutter would be during a benchmark session.

 

Here is a plot from my system before I tuned things:

 

image.thumb.png.eb23530c74f8e001a411e66a77dda0ab.png

 

The big spikey clusters are things like physics happening when your track blows the wings off a bomber. The individual spikes are apparently random.

 

Here is a plot after tuning all the stuff I listed in previous posts:

 

image.thumb.png.c1a653ab64bd62e5b5670f59fe566ba7.png

 

You can see that this is much better, but image quality is only adequate with PD = 1.1 (SteamVR SS equivalent 120%?).

 

I do agree that I am likely not hitting my GPU limit and I could increase my SS, but increasing SS tends to increase the chances of missed frames. So I am seeking to find a balance, I will do some real-world gaming to see. Personally I think if there were a magic 60 fps setting for ASW that would be ideal (but I have no idea if that could even work technically, let alone the difficulty for Oculus to implement).

  • Upvote 1

Share this post


Link to post
Share on other sites
On 9/14/2018 at 2:22 AM, Alonzo said:

Increasing my Ring Ratio (speed at which the chip cache and other non-core stuff runs) from 45 to 47 (4500mhz to 4700mhz) improved my 0.1% low by a whopping 7 fps


Alonzo, as far as I know uncore is still uncharted territory for most of us. While the effects of both cpu clock and memory speed has been well documented in this thread I believe uncore/ring ratio is another bottleneck for us. And perhaps even more so when leaving it at default while superclocking the cpu. 

cpu@4.8, uncore ratio 42(default), mem@3000, cpu-z shows NB @ 4200mhz
Avg: 132.417 - Min: 76 - Max: 220


cpu@4.8, uncore ratio 48, mem@3000, cpu-z shows NB @ 4600mhz
Avg: 136.050 - Min: 76 - Max: 221.33

These are the results of a monitor test. Not VR. At an average of 132fps my 1070 was at it's limit so I believe uncore clocking would yield greater gains in a VR benchmark. Will try to do one tonight.

 

Share this post


Link to post
Share on other sites
13 hours ago, SvAF/F16_radek said:

Alonzo, as far as I know uncore is still uncharted territory for most of us. While the effects of both cpu clock and memory speed has been well documented in this thread I believe uncore/ring ratio is another bottleneck for us. And perhaps even more so when leaving it at default while superclocking the cpu. 

 

Yep, definitely. The reason I looked at it was that memory speed seemed to have such a big effect on 0.1% low framerates, so it implies to me that memory access latency / contention is a big factor for IL2. If memory speed is an issue, maybe cache latency could also be a factor? That's why I messed with the setting, and as you've found it has a significant impact.

Share this post


Link to post
Share on other sites

my latest test with 3.005 Spit track.

 

i7 8700k @5gh (4.8 avx)  (42 uncore ratio)

16gb ddr4 3200mhz

gtx 1080

 

Frames: 8619 - Time: 120000ms - Avg: 71.825 - Min: 44 - Max: 90

 

I tried OCing the uncore to 48 and bumped avx to 4.9ghz which was verified by cpuz, however the fps went DOWN(?)

i7 8700k @5gh (4.9 avx)  (48 uncore ratio)

 

Frames: 8099 - Time: 120000ms - Avg: 67.492 - Min: 44 - Max: 83

 

so i tried i7 8700k @5gh (4.9 avx)  (46 uncore ratio)

 

Frames: 7784 - Time: 120000ms - Avg: 64.867 - Min: 44 - Max: 77

 

and then reverted back to my standard OC

i7 8700k @5gh (4.8 avx)  (42 uncore ratio)

 

Frames: 9075 - Time: 120000ms - Avg: 75.625 - Min: 44 - Max: 90

Frames: 8832 - Time: 120000ms - Avg: 73.600 - Min: 44 - Max: 90 (ran it twice to be sure.)

 

Anyone have some ideas why this might be?

i watched the cpu core frequency during the benches and didnt see them throttle (didnt watch the FSB frequency though.). Temps stayed normal too.

 

 

 

 

 

Edited by Wraithzlt

Share this post


Link to post
Share on other sites

Maybe try changing just one thing per run. It could be that raising uncore too high caused some kind of problem. You could also try a bit of Prime95 (recent, with AVX) doing a mostly-memory test just to check that things are ok on the CPU at whatever settings you're using. For your setup I would try specifically:

  • Baseline: 8700k @ 5.0, AVX -2, uncore 42
  • Uncore 45 (any difference? better/worse? stable in Prime95?)
  • AVX -1 (any difference?)
  • Memory set to manual 3200 timings (16-18-18-56 or whatever)
  • Memory set to 3333 16-18-18-56 (or whatever -- you must manually set the timings because if you just specify 3333 your mobo will loosen the timings to hit the frequency)
  • Memory set to 3466 16-18-18-56

The other thing to look at is Frahfsbench to look at the 1% low and 0.1% low numbers. These tell you how bad the worst stutters will be, sometimes a better indication than average framerate.

 

The other other thing is that in-game settings have a lot of impact. Balanced is a bit less CPU intensive than High, and then sub-settings will also have a CPU + GPU impact. 

  • Thanks 1

Share this post


Link to post
Share on other sites

i have run some more benches. All these tests are run using this threads VR bench settings and the Spit track.

 

i7 8700k @5gh (47 avx)  (42 uncore ratio)

16gb ddr4 3200mhz

gtx 1080

 

Frames: 8832 - Time: 120000ms - Avg: 73.600 - Min: 44 - Max: 90

avg time - 13.6 

1% time - 23.1

0.1% time - 26.3

 

i7 8700k @5gh 49 AVX 42 Uncore

Frames: 9331 - Time: 120000ms - Avg: 77.758 - Min: 44 - Max: 90

avg time - 12.9 

1% time - 22.7

0.1% time - 25.1

 

i7 8700k @5gh 49AVX 44 Uncore

Frames: 9461 - Time: 120000ms - Avg: 78.842 - Min: 44 - Max: 90

avg time - 12.7 

1% time - 22.6

0.1% time - 25.3

 

I will add more of just uncore tweaks  without AVX increases later and maybe RAM increases if i have time.

 

i7 8700k @5gh 47AVX 45 Uncore

Frames: 7592 - Time: 120000ms - Avg: 63.267 - Min: 44 - Max: 78

avg time - 15.8

1% time - 23.4

0.1% time - 30.7

 

Well something blows it out with this setting?! i guess there is some throttling going on by the motherboard that im unaware of.

Edited by Wraithzlt

Share this post


Link to post
Share on other sites
3 hours ago, Wraithzlt said:

Well something blows it out with this setting?! i guess there is some throttling going on by the motherboard that im unaware of.

 

Those were some pretty good results, right up until the setting that was worse! Maybe try sticking with uncore at the lower speed, or increasing it only very slightly? Apparently uncore likes CPU core voltage, but that will increase heat too, so it's a balance.

Share this post


Link to post
Share on other sites

Thanks @chiliwili69 for the new track! I tried it out on my all AMD system with Samsung Odyssey.

 

Settings were as prescribed, with SS 1.28 in Steam VR.

 

My PC:

AMD Ryzen 2 2600X @ 4.2GHz (single core boost)

16GB DDR4 3000MHz RAM

AMD Vega 56 with balanced settings

Samsung Odyssey

 

PASSMARK CPU Mark 14072

PASSMARK CPU Single Thread 2161

 

Did three runs with similar results each run. I think the FPS is capping at 45 thanks to WMR. However mins were lmited to 33/34.

 

Frames: 5054 - Time: 120000ms - Avg: 42.117 - Min: 33 - Max: 46

Frames: 5114 - Time: 120000ms - Avg: 42.617 - Min: 34 - Max: 46

Frames: 5107 - Time: 120000ms - Avg: 42.558 - Min: 34 - Max: 46

 

 

FRAPSLOG.zip

Edited by guidom

Share this post


Link to post
Share on other sites

That's rough.

 

Anyways for the guys getting lower scores with the uncore oc and higher cpu oc's thats because the chip isn't stable and is probably throwing errors. If you ran prime on it you would most likely see failing cores. I'd bet money on it. I don't mean failing as in going bad just cores that aren't doing number crunching properly and throwing out bad numbers which in turn slows stuff down.

Edited by 15th_JonRedcorn

Share this post


Link to post
Share on other sites
On 9/22/2018 at 12:48 AM, Wraithzlt said:

i7 8700k @5gh 49AVX 44 Uncore

Frames: 9461 - Time: 120000ms - Avg: 78.842 - Min: 44 - Max: 90

avg time - 12.7 

1% time - 22.6

0.1% time - 25.3

 

Thanks, I have updated your top result to the table. I have not your STMArk.

Have you tried with AVX=0?

 

You also don´t need to report avg time, 1%time or 0.1 %time since it is some how redundant:

 

avg time 12.7 is just 120000/9461 or also 1000/78.842

1%time 22.6 is basically 1000/44

0.1%time are just ramdom spikes

 

Share this post


Link to post
Share on other sites
On 9/22/2018 at 3:32 PM, guidom said:

Did three runs with similar results each run. I think the FPS is capping at 45 thanks to WMR. However mins were lmited to 33/34.

 

Frames: 5054 - Time: 120000ms - Avg: 42.117 - Min: 33 - Max: 46

Frames: 5114 - Time: 120000ms - Avg: 42.617 - Min: 34 - Max: 46

Frames: 5107 - Time: 120000ms - Avg: 42.558 - Min: 34 - Max: 46

 

Thank for reporting this test. I am not familiar with WMR, but as per other comments, it seems that those devices uses a mechanism similar to Oculus ASW, so any value from 90 to 45 is equal to 45. I don´t know if this can be switched off, otherwise your avg values can not be compared with other tests.

Share this post


Link to post
Share on other sites
On ‎9‎/‎22‎/‎2018 at 9:32 AM, guidom said:

 

 

Did three runs with similar results each run. I think the FPS is capping at 45 thanks to WMR. However mins were lmited to 33/34.

 

Frames: 5054 - Time: 120000ms - Avg: 42.117 - Min: 33 - Max: 46

Frames: 5114 - Time: 120000ms - Avg: 42.617 - Min: 34 - Max: 46

Frames: 5107 - Time: 120000ms - Avg: 42.558 - Min: 34 - Max: 46

 

 

FRAPSLOG.zip

 

You might have a peek at the below thread regarding reprojection in the Odyssey.

https://forums.eagle.ru/showthread.php?t=220815

 

I don't have a WMR device so can't really comment on how this might help or not. Just happened to remember reading it a little earlier today.

Edited by dburne
  • Thanks 1

Share this post


Link to post
Share on other sites
16 hours ago, chiliwili69 said:

Thanks, I have updated your top result to the table. I have not your STMArk.

 

Cpu mark 17672

SingleThread Mark - 3009

 

 

16 hours ago, chiliwili69 said:

You also don´t need to report avg time, 1%time or 0.1 %time since it is some how redundant:

Was putting those in for Alonzo and anyone else who was interested in the Frafs graphing tool.

  • Like 1

Share this post


Link to post
Share on other sites

Testing uncore according to latest benchmark settings. I'm not quite positive I managed to uninstall the Migoto mod correctly, but if not it was equally uninstalled for both tests.

Constants are:
i7 7700k @4.8
3000mhz ddr4
gtx1070
0 avx offset
Passmark: 13481 single thread: 2818

Here, the uncore is at default 42. And this is the average result of two runs:
Frames, Time (ms), Min, Max, Avg
 7284,    120000,  44,  78, 60.704

Next, uncore bumped to 48. Nothing other than uncore and cpu-voltage were changed. Again the result is an average of two runs:

Frames, Time (ms), Min, Max, Avg
 7766,    120000,  43,  82, 64.717

Notes.
When uncore is left at default 42, Cpu-z shows northbridge frequency at 4200mhz. When I set uncore to 48 and bios confirms my cache target frequency is set to 4800mhz. Cpu-z says northbridge is at 4600mhz. I don't know which to trust.

Edited by SvAF/F16_radek
  • Like 1

Share this post


Link to post
Share on other sites
2 hours ago, SvAF/F16_radek said:

Frames, Time (ms), Min, Max, Avg
 7766,    120000,  43,  82, 64.717

 

Thanks radek.

You are lucky  by changing your ring ratio. In my BIOS I have not the option to do it. It is in Auto and I can not change it. It is always at default 4000Mhz independently if I OC to 4.4, 4.6 or 4.8.

I spend some time in my BIOS but I don´t know how can I remove my Auto.

Share this post


Link to post
Share on other sites

Chili, I'll see if I can google something up regarding your bios/ring ratio. As you already are running a higher than stock cpu-voltage you might have a stable ring-ratio increase already on the table. 

Is the "VR expected" still based on single thread passmark number? Or did you ad further variables?

Share this post


Link to post
Share on other sites
46 minutes ago, SvAF/F16_radek said:

Chili, I'll see if I can google something up regarding your bios/ring ratio. As you already are running a higher than stock cpu-voltage you might have a stable ring-ratio increase already on the table. 

Is the "VR expected" still based on single thread passmark number? Or did you ad further variables?

Thanks. Those BIOS are like a big monsters for the inexperienced users like me.

 

The VR expected is as always, just based in STMark and the lineal correlation of all test in the page.

Share this post


Link to post
Share on other sites

I ran the benchmark these are my results. Followed the instructions to a t. I am going to try it with my settings and see how it goes.

 

2018-09-25 17:01:26 - Il-2
Frames: 8028 - Time: 120000ms - Avg: 66.900 - Min: 43 - Max: 91

 

Surprisingly got a worse result but the tracking started bugging out at the end, it was twitching the camera all over don't know if that effected it.

 

2018-09-25 17:08:55 - Il-2
Frames: 7602 - Time: 120000ms - Avg: 63.350 - Min: 42 - Max: 91

 

The game certainly looks far better on these settings I posted though. I think I posted them in the wrong thread.

Edited by 15th_JonRedcorn

Share this post


Link to post
Share on other sites
22 hours ago, 15th_JonRedcorn said:

Surprisingly got a worse result but the tracking started bugging out at the end, it was twitching the camera all over don't know if that effected it.

 

Thanks for the results, but in order to be sure about the result, you can run the test 2 or 3 times in order to see if you obtain almost identical results. So, I can write your result to the table.

Please, report also the Passmark and Single-thread passmark.

 

The settings of the test are not the best, but just medium values for most of them. IT is just a benchmark.

This weekend I will try your settings. I think you posted them in this post:

https://forum.il2sturmovik.com/topic/39799-2080-and-2080-ti/?do=findComment&comment=673485

 

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

×