[Official] DX9 vs DX11 performance difference

[demerzel]

VERY BASIC - it is more technical than this and im not over the intracies to the level needed to code directly - but

You are rendering polygons.. you do math, convert 3d points to a 2d image, you do some more maths to work out where light is to determine where / how triangle is shaded. you shade triangle with texture map bump map spec map.. Dx9 to DX11.. almost the same.. (Same math same process). (So should take identical time and hence identical frames)

However DX11 then offers advantages with tesselation (RF doesnt use it).. Post processing effects and the ability to handle more complex shaders

So you should find (Post processing off) Dx9 - dx11 very similar frames in all instances. You will find with PP on any level higher than off you will be a little worse off..

The equation changes if you start using some custom tailored shaders for DX11.. then you may see performance gains.. however none of the current content uses them. (as it wouldnt work on people using dx9 version).

Put simply, if they abandon DX9, Tailor it only for DX11 and everyone reworks content with new more complex shaders, you could get better performance improvements - but you will allienate any DX9 users or lose them altogether.

So best case dx9 to dx11 with no effects = framerate the same. Custom content built with custom dx11 shaders (And will only work on dx11) will be faster - Of which there is none.

At one point. They will stop updating dx9. @Marcel Offermans said on the other day on discord.

 

[stonec]

So i should change it to MEDIUM right? The image quality loss is high? Will test it later, thank you very much!

You should put post FX to None if you want to compare with DX9. DX9 had no post effects.

VERY BASIC - it is more technical than this and im not over the intracies to the level needed to code directly - but

You are rendering polygons.. you do math, convert 3d points to a 2d image, you do some more maths to work out where light is to determine where / how triangle is shaded. you shade triangle with texture map bump map spec map.. Dx9 to DX11.. almost the same.. (Same math same process). (So should take identical time and hence identical frames)

However DX11 then offers advantages with tesselation (RF doesnt use it).. Post processing effects and the ability to handle more complex shaders

So you should find (Post processing off) Dx9 - dx11 very similar frames in all instances. You will find with PP on any level higher than off you will be a little worse off..

The equation changes if you start using some custom tailored shaders for DX11.. then you may see performance gains.. however none of the current content uses them. (as it wouldnt work on people using dx9 version).

Put simply, if they abandon DX9, Tailor it only for DX11 and everyone reworks content with new more complex shaders, you could get better performance improvements - but you will allienate any DX9 users or lose them altogether.

So best case dx9 to dx11 with no effects = framerate the same. Custom content built with custom dx11 shaders (And will only work on dx11) will be faster - Of which there is none.

Well explained. I may add that there are certain other improvements in DX11 version that costs FPS, such as ambient occlusion and somewhat better shadows. I think at this point we can't expect better performance from DX11 than DX9, unless the shaders are re-written. Hence I decided to buy a new GPU, I think performance improvements from here on will be limited.

 

[MaD_King]

So i guess DX11 it isn't very well optimized yet, usually DX11 offers much better graphics with better performance, we have other examples i won't mention here, so we'll have to wait for future updates...

Thanks!

When I read this (and you are not the first to tell that), I'm not sure the things done by S397 are clear for all.
In the DX11 beta build, S397 introduced the DX11 but also added new graphical features:
- Improved MSAA rendering
- Improved Clouds resolution and depth
- Static AO;
- Dynamic AO.
- Shadows Normal renders shadows, taking into account the shadow receiver normal map.
- New Car Ambient Shadow

See here for more details.
https://www.studio-397.com/2017/05/open-beta-with-directx-11-and-virtual-reality-support/

So if you have same (or quite close) FPS between DX9 & DX11 with Post Effects = None, then you can be happy of that, because, you get more feature in DX11 than the DX9 version with no impact. And visually, the results is a huge better.

And the bonus (visually speaking, but not in FPS) is the addition of Post Effects, that add Bloom and DOF blur, and that is hungry in FPS.

So I hope I will not read again guy saying, DX11 is worse in FPS than DX9, because for me it's not the case, I have quite same FPS if I test with Post Effects = None, and I have a better rendering of the shadows/cloud, lights, ... in DX11 (and the texture look better for me in DX11)

 

[Guimengo]

At one point. They will stop updating dx9. @Marcel Offermans said on the other day on discord.

If they move to Vulkan then this whole DX issue will be forfeit, and people who bought on the promise of not being discontinued will not have a problem. [/Perfect World]

 

[suryaprihadi]

The multi GPU improvement is not "up there" yet. Nevertheless, it is a progress.

Parameters :
- Setting at best/full even for shadows, except Anti Aliasing which is at Level 4 to match DX9
- 20 AI + me* ( *AI driving too ) in race mode, so FPS will be slower
- Cloudy sky, so FPS will be slower
- SLI bits for DX11 is NONE ( all zeros )

===================================================

Resolution 7680x1440 Multiview. 2 x GT1080Ti

Silverstone, I do not consider this as an SLI gain ...........

.
Nola C, did get some SLI gain ....at least something

====================================================

Resolution 2560x1440 . 2 x GT1080Ti . Post Effect MEDIUM.
Capped on purpose to 120 FPS. Single GPU result look promising, so I was hoping by capping and using 2 way SLI at post effect medium, I can get near flat 120 FPS, but that is not the case unfortunately.

==============================================
Different cars different FPS ?
Indeed

Clio is "light" and GTR 500 2013 is "heavy", FPS wise. Silverstone GT.

NOLA-C

================================================================

Even at 2560x1440, different cars, different FPS.
Parameters the same, setting at best/full even for shadows, except Anti Aliasing which is at Level 4 to match DX9

Clio at Nola-C and SilverstoneGT, capped to 120FPS can maintain near flat 120FPS, but not GTR 500 2013

These results are actually much better than 1 build ago, which in 2way SLI is actually like 5% lower FPS, but GPU loads both are 85-95%...weird.

I think in 2 or 3 more DX11 multi GPU update, some good result will happen.

End.

 

[Dady Cairo]

Sorry,fore disturbing the discussion...but Admin means to move it here better.
As i bought RF2 for my Laptop i January 2017,it was great. i hoped with updates it will even get better..but .....

Dx11 content mixed with Dx9 content give unstable performance.
Example: F1 1975 cars on a Dx11 Track like NOLA -worst FPS and stuttering.Same track with EVE F1 Cars 58-60 fps.Crahes of the AI cause stuttering.
Settings: PP Low, No AA no FXAA,Res 1280x720, in game Grafix set to medium Shadows low,no refelections,no Blur,visible cars 15,opponents 14.
It will End like GPL,the game was top,the Grafix less and you got to buy hardware for 3000 Euro,and then it will not run to satisfying yourself.
60 fps is needed,i can pull down settings but on low,no,no,no...
Sure RF2 has all,Dynamic weather,Nigth Race etc,but you can't play that on a medium computer.
I prefer Automobilista,Assetto Corsa (also@1280x720 no AA DX11 title),and RaceRoom,wich all run on DX9.
I wish the Developers hard luck,but i do not buy a brand new Laptop,wich i did in January even to play that Fu...RF2!
And Automobilista looks GREAT!Even Raceroom on my machine.....Good they left their fingers from things,that they won't get right.
My stand.
A Sim ,written for DX9 and now in Dx11 and the 3rd pary tracks and cars in DX9,oh my GOD,maybe Christmas-half ready.
PS.I speak of the unfinished DX11 Beta and i can't make a "real" test in cause of the resolution issue with fullscreen.Had to set resolution to nativ 1280x720p otherwiese no Fullscreen in this resolution.If someone has a similar Machine with better results,let me know what is possible.
DC

Specs:
Lenovo Ideapad 700
Pocessor:Intel i5 - 2.3 up to 3.2 MhzTurbo
GFX:Nvidia 950 M (mobile) 4 GB
Ram: 16 GB DDR-3
HD:250 GB Samsung SSD
System:Win 10 64 bit Home

Just to compare...
Thanx
If you find misspelling Keep it by yourself....lol

 

[Emery]

The reduction in latency was confirmed & much appreciated! Confirmed triple screen tool (Ctrl+=) is working with build 1944770. Fps are still not doing so well on the tougher track.

settings: Level 4 AA, PPE=off, resolution 5760x1080 multiview, no sync, aggressive threads=off
system: single GTX 1080, 8 GB VRAM; i7-4770 @3.4 GHz, 32 GB RAM
view: cockpit camera
weather: overcast

Atlanta Motorsports Park, Cobra
DX9, Avg 88 fps, Min 75 fps, Max 101 fps
Build 1941092, Avg 82 fps (-7%), Min 72 fps (-4%), Max 94 fps (-7%)
Build 1944770, Avg 83 fps (-6%), Min 72 fps (-4%), Max 95 fps (-6%)

Palm Beach, Cobra
DX9, Avg 109 fps, Min 100 fps, Max 122 fps
Build 1941092, Avg 107 fps (-2%), Min 99 fps (-1%), Max 121 fps (-1%)
Build 1944770, Avg 109 fps (-0%), Min 101 fps (+1%), Max 123 fps (+1%)

 

[Lazza]

@suryaprihadi The latest update mentions multi-GPU, but I don't think it means SLI. It's more about having a primary and secondary GPU, and being able to select which one you want in the game. Or something like that. If they were targeting an SLI improvement I'm sure they'd say SLI.

@Dady Cairo There isn't really much difference between dx9 and dx11 content, as far as I know. 'Optimised for dx11' is mainly just about albedo textures (avoiding excessive bloom, better overall balance). There were also guidelines for texture sizing but I'm sure that would benefit both. There's definitely nothing you can do in dx11 (shaders etc) that makes it dx11 and not dx9 (which might make for some performance difference), at least with the standard shaders (?). But different content will perform differently, sometimes because of different authors, different development era, or just the nature of different content. Your CPU base speed is also potentially a concern, though again different content might have an influence there.

 

[Daniele Vidimari]

Could be an excessive bloom a reason for low performace with not optimized contents? I've noticed some stuttering with some not optimized tracks like Sebring or Adelaide.

 

[Lazza]

Could be an excessive bloom a reason for low performace with not optimized contents? I've noticed some stuttering with some not optimized tracks like Sebring or Adelaide.

I don't know enough about how bloom is calculated/done, but I suppose it's possible. And maybe that's why this thread instructs people to turn post processing off when comparing, so it's not an issue

 

[suryaprihadi]

@suryaprihadi The latest update mentions multi-GPU, but I don't think it means SLI. It's more about having a primary and secondary GPU, and being able to select which one you want in the game. Or something like that. If they were targeting an SLI improvement I'm sure they'd say SLI.

aha...make sense.. thank you. No wonder there is only AUTO and my two cards only recognize as 1 model and quantity.

 

[Rony1984]

breakthrough!! (for me at least) succesfull reproduction of the 'fullscreen-issue':

https://forum.studio-397.com/index.php?threads/succesfull-reproduction-of-the-fullscreen-isue.56754/

 

[RS500]

GPU: AMD R9 270X
OS: Win7 64bit

 

[Findla22]

GPU: R9 280
OS: Win 10 Pro x64
RAM: 16 GB

DX9 vs DX11: DX11 is having more or less good fps, but they are little lower than DX9. The main difference is the post processing, I think it needs to be optimized because I loose about 30-40 fps from going to None to Low... I can't enable it and it hurts because post processing is the main graphical difference with DX11 build.

 

[Wand Villela]

For me this latest dx11 is near identical with dx9 at 69-70fps average in Medium setting.
First time ever dx11 is not slower , but circuit is Nordschleife Tourist Autumn v2.01 which is not by ISI/S397.
7680 x 1440 in angled multiview. GTX980 2 way SLI 100-95% GPU load, DX9 SLI compatibility Bits 0X02402004 as recommended by Ari. DX11 sli compatibility bits as per Nvidia and based on driver 382.05

DX9 on the left, DX11 on the right
View attachment 8180
After messing around with SLI compatibility bits for weeks, unless you get one which can really throttle up your GPU both in DX9 and in DX11 to 90%+, I think your result will be more influenced by under utilization of the GPU and not DX9 vs DX11 per se.

Dang, this graphic thingy is complex for me

What profile are you using for DX11?

My GPU2 is 10% always.

 

[suryaprihadi]

Wand,
It was all bit zeros ( none ) for the DX11. Anyhow I already change GPU to GTX1080Ti, not the 980 anymore.

 

[Jon]

I personally believe the best visual improvement would come from simply adjusting the colour pallette and lighting. It also comes with an added benefit: next to no performance impact. Bloom and DoF are really just icing on the cake, and really not needed during the race considering their impact on performance, although the DoF does really look amazing in replays, they've certainly nailed that especially when the car is blurred.

Which reminds me, hopefully in future builds there'll add the ability to set up graphics independently; a graphics set up for replays and a separate one for races. If I remember correctly iRacing has this implemented.

 

[stonec]

I had a chance to borrow a GTX 1000 series GPU to compare the results with my older 600 series card and had a free day to perform various tests. Here are the observation I made

Observation 1: PCIe Bus Usage % (reported by MSI Afterburner) is very high. This is same regardless of DX9 or DX11. Bus usage in most other games I tried hovers around 10%, in rF2 it approaches 50%. This tells me that a lot of traffic goes between the GPU and main RAM, which is a bit odd, as new GPU's have plenty of inbuilt VRAM and should be very good at minimizing the PCIe traffic, which can become a bottleneck.

Before today's tests, I had accidentally plugged my old GPU in the wrong PCIe slot, which only gave 8X speed. This cut rF2 FPS up to 30%, even though the Bus usage was "only" 60%. So anyone using PCIe 2.0, beware that if it's operating at reduced speed, it will cost massively. Same with PCIe 3.0 on more powerful cards if it's stuck on 8X.

Observation 2: DX11 deficit is much bigger on Kepler generation GPU than on Pascal. Below are the charts comparing both cards with DX9 & DX11 on Silverstone with 20 other cars. DX11 deficit with Kepler is around 25% here, with Pascal it's less than 10%. The performance difference on the modern architecture I would call acceptable considering that image quality is better, but the 25% is a lot for anyone that owns an older GPU. Not sure what the status is with AMD, but traditionally AMD GPU's age less slowly than Nvidia, so maybe the situation is better there.

 

[suryaprihadi]

Nice one Stonec, on the Kepler vs Pascal architecture
I googled to find more similar test and this is good reading too. Kepler vs Maxwell and that BUS use thingy.
http://www.tomshardware.com/reviews/graphics-performance-myths-debunked,3739-4.html#4

I don't understand how one GPU architecture , over other architecture , prefer ( or do well ) for DX9 or DX11 games, but my
VegasPro video editor has a nasty preference for old Kepler GPU, GTX580/570 for MP4 codec by MainConcept.
Since 2012, I need to use GTX580 and not newer model, if I want to get fast rendering and high GPU load.
2 way SLI dont work, even dual Kepler GPU like GTX690 wont work, only 1 GPU will get the load.
However, Adobe works better with newer GPU, so I heard.

So I think game developer themselves play a big role in how they develop their games to optimize DX9 or DX11 and/or newer GPUs.

 

[hitm4k3r]

I made a performance comparison regarding my slowdowns with the McLaren 650s GT3. All information about system, game and race settings are within the zips. I used the 650s vs the Fia F2 in DX11 and tested the 650s both in DX9 and DX11 allways with the same settings. There is definetly something with that car in DX11, that doesn't like my CPU.