Comparison and evaluation of the perfomance and resource usage of AA and AF in AC

[Georg Siebert]

valid message

 

[Bram Hengeveld]

What a top quality post

 

[Blkout]

George, first of all, thanks for spending so much time running tests and typing this up. Much respect for that. But, here's where I have to add some exceptions that can paint a different picture compared to your data but as you mentioned there are variables which can change the results.

I guess I can start with your AF numbers. You're still getting a far bigger hit than I am on a single 780 or SLI GTX 780 setup. Is it because the 780 has more memory? Probably not. Is it because the 780 is more powerful? Maybe. Is it due to a different bench test? Possibly.

Secondly, the AA results from 4x to 8x will have far more impact with more trees and cars on screen. Monza 66 is a killer because of the amount of trees. When you go through the first few turns, the framerate will take a bigger hit with 8xAA due to the volume of trees. You can easily record a reply and then watch it after changing each setting. Its a repeatable test without any variables. I get a 12fps drop going from 4xAA to 8xAA at Monza 66 and that is with 0 AI cars on screen, just myself, and those are minimum framerates.

All that being said, there's no perfect way to test due to so many variables even though you tried to eliminate as many as possible. Kudos.

 

[Stuart Whitehead]

Great post George

Thanks for taking the time to do all that testing and for writing up the results in such a detailed and professional manner.

 

[Martymcfly]

Very useful read George. Thanks
Being a Bit of a PC enthusiast and hobbyist I agree with Your scientific conclusion of further optimizations are needed to get the FPS performance up a bit. Looking at your PC specs I think the scientist needs a little bit quicker CPU to do his scientific research a bit quicker......

 

[Pedro Cunha]

Thanks, nice work!
I just hope one day i can let you guys know it is possible to max out this game at 1080p w/fullgrid above 60fps.

 

[Blkout]

Thanks, nice work!
I just hope one day i can let you guys know it is possible to max out this game at 1080p w/fullgrid above 60fps.

If anyone can, it will be you Pedro.

 

[Chris]

Terrific post!

You should seriously consider posting this on the AC forums, because this is exactly the type of feedback that AC developers will find invaluable.
Because 90% of the feedback developers get is along the lines of:

'Fix this"
"The graphics are sh*t"
"I can't run it"
"The cars handle terribly"

So when they finally get something of this quality, then it helps them immensely.

Great stuff mate!

 

[vee10]

Great work Georg obviously Kunos need to do some optimisation and with information like this they know exactly what needs working on.

As for your report as my old French teacher used to say marking homework (not on mine of course) A+ formidable.

 

[Eric Bergeron]

Great work Georg !
You will be a great scientist, that's obvious.
thanks, it's appreciated

 

[Joseph Maxx]

I like your methodology.

The only thing I can add is that the results do vary depending on which track you're running. As I think has already been mentioned, tracks with a lot of trees trackside have a greater impact on performance. I hope that gets sorted out by the time we reach 1.0.

 

[Mogster]

My only performance issue is that the in game AA just doesn't work....

I understand that its probably Nvidia's fault rather than Kunos but the shimmering at Monza even with 8xAA in game is just horrendous. The game desperately needs 4XSGSS but anything more than 2XSGSS means the HDR overexposes everything

 

[Paul Lock]

4XSGSS but anything more than 2XSGSS means the HDR overexposes everything

Yeah the 4xSGSSAA via inspector does create somewhat unwanted over bloomed/exposure as well as smoothing out all those jaggies.

Which is why you just edit the config file for Bloom and HDR pass levels.

I found bloom @ 0.6 and hdr pass at 0.4 were pretty sweet and got rid of that 'aaaarrrrggghhh my eyes' effect when you hit most tracks

 

[Blkout]

My only performance issue is that the in game AA just doesn't work....

I understand that its probably Nvidia's fault rather than Kunos but the shimmering at Monza even with 8xAA in game is just horrendous. The game desperately needs 4XSGSS but anything more than 2XSGSS means the HDR overexposes everything

I don't think its Nvidia's fault on the antialiasing. I think it has to do with the type of rendering they're using in the graphics engine which is why standard MSAA can't fix it all and only SGSSAA can. Don't quote on me on this because I can't swear to know how AC's rendering is done, but it could be a form of deferred rendering. Before you ask how MSAA and deferred rendering play well with each other, I'll quote a post on another forum from someone that can explain it much better than I can. The reason I specualte it could be deferred rednering is for two reasons, without V-sync, AC heats my video cards up like no other game which I've only seen in deferred engines and deferred engines tend to do real time lighting better than other engines which is why DICE chose to use it in Battlefield 3 and 4, for better lighting effect which AC also seems to have.

"If MSAA is properly implemented in a deferred engine, it is equal to a proper MSAA implementation within a forward renderer.
If a scene isn't fully anti-aliased, then either, the developers are incredibly sloppy and don't really care about accurate MSAA, or it is done to increase performance.
Frostbite 2 uses compute shaders to calculate the lightning on each pixel, if the contrast is low MSAA takes a single sample per-pixel for the shading, if contrast is high, it uses multisampling for that pixel's shading. This get's merged with the fully multisampled geometry in a final pass. This significantly lowers the memory footprint, and thus costs less bandwidth. Aliasing might be theoretically less accurate, compared to 'full scene' AA, but this should not be noticable at all.
Sadly however, there are a ton of examples of awful MSAA implementations, forward engines just as well as deferred engines. Well known forward-rendering engines with awful MSAA implementations are Unreal Engine 3 and the EGO engine.
And if you're not aware, MSAA only affects geometry. Games with tons of sharp specular maps and shader effects (EVE Online comes to mind) will still have a lot of aliasing, even with 8x MSAA. There are techniques to significantly reduce such aliasing without reaching for blurring post-process-AA, but nobody has cared enough to implement those outside of techdemo's."


And another quote from another forum in regards to deferred rendering pros and cons.

"The main thing with deferred rendering is that you fill in info in a giant buffer about every point until you've gotten through all the geometry. Then you just render a full-screen flat polygon using that buffer in the pixel shader so it can get info about every point.
The biggest advantage for a lot of games is how well it scales with lighting complexity, especially since you never light a pixel that isn't visible.
Biggest disadvantages include transparency rendering (which is really a pain, but not impossible), and inability to use hardware AA (and even software AA cheats don't work out so smoothly)."

 

[Toni Anicic]

Great great great post man.I was searching for this

 

[Leonardo Ratafiá]

thanks a lot for this information! it's really helpfull!

 

[Georg Siebert]

Thank you for your feedback, it's appreciated.

[...] You can easily record a reply and then watch it after changing each setting. Its a repeatable test without any variables. [...]

Testing AF on a track with lots of trees, like Monza '66, is a good idea. However, benchmarking with a moving car is to open another can of worms. Gonna have to figure out first how to properly do that.

[...] You should seriously consider posting this on the AC forums, because this is exactly the type of feedback that AC developers will find invaluable. [...]

Yes, I will consider it. Probably after multiplayer has been released and the official forums quiet down a little. Have to make an account first, too.

 

[Blkout]

Thank you for your feedback, it's appreciated.


Testing AF on a track with lots of trees, like Monza '66, is a good idea. However, benchmarking with a moving car is to open another can of worms. Gonna have to figure out first how to properly do that.


Yes, I will consider it. Probably after multiplayer has been released and the official forums quiet down a little. Have to make an account first, too.

George, you can record a replay of your game using the in game save replay function, then you can simply playback the replay and watch each graphics setting change in real time and see how it impacts framerate. Since its a replay, its the same test every time, or were you thinking of some other added variable in testing this way?

 

[Georg Siebert]

[...] or were you thinking of some other added variable in testing this way?

Yes, they are. Time of day, shadows, camera angles, liveries, complexity of car textures and probably some few more I'll come across.
Then there is the method of recording and plotting fps: logfile - min, max, avg - graphical visualisation - homogenisation, comparison and differentialisation of graphs - compressing time in graphs - ...
That a few times, because the fps counter is moving.

First I want to re-release my setup guide. Edit: done
Comprehensive Setup Guide for Novice and Advanced Drivers

 

[Rafa91]

Hello everyone, I'm sorry to bring up this old thread but I think it is very interesting.
Unfortunately all I can see in the first post is "valid message"
Is there anywhere where I can find this information?

Thanks and cheers