I have just made some testing of the FPS performance, as follows http://acstuff.ru/s/3xCY:
Cannot test skin format as again I don't have DDS tools installed for Photoshop to be able to open and save them, somehow I always have to install DDS support almost every time I want to work with DDS as I only work with it rarely. Skins seems to be DXT1, Kunos seems to use DXT5 which as far as I remember might have higher compression and what I would normally use, though DXT overall is a lossy format, the advantage of it is that it can be straight copied and used by the GPU and doesn't need converting such as PNG and JPG textures do. I do not see the need for the number plate be this large, could be half the size but I doubt that's an issue anyway.
As you can see GPU usage is very low around 30-40% during the test and CPU single thread as reported by AC is around 95% no matter what car is used. The total CPU usage is around 60-65% for me on all cars again. FPS varies but I have not been able to find out why.
IMHO it may do with car paint differences, also clamshell has car paint reflections (didn't test that one, brb, added), small possibility would be textures, most likely it's the detail offered on Caterhams both external AND internal (dash, seats, everything in driver view), even Kunos cars can vary greatly.
If I had to guess I would say it has to do with car paint reflectivity and reflections in general as most likely Kunos is still computing those on CPU, at least that's how it was done in the past in graphics engines when it came to reflections and other similar lighting stuff as far as I remember. It's definitely being CPU bottlenecked in the AC render thread by something, reflections or drawcalls etc.
The hitbox should not be crazy, you can check that in CM. Collisions again are CPU processed as far I'm aware in every game.
I do not know or have tools right now to go debug AC graphics if it would even allow me, only to see how many polygons are being used by each car and total in the scene. Ben will have to compare and answer that if the Cats have similar poly count - complexity in geometry compared to other stock AC cars. To me it doesn't seem crazy poly count and LODs work well. Even with LODs put low so that quality is poor the gain in FPS is minimal. IMHO it is not a poly count issue.
Unfortunately AC doesn't offer much when it comes to dev apps to see where the game is bottlenecked beside the never ending CPU single thread bottleneck.
BTW the RD CODE block is not working as a proper CODE block it doesn't keep tabs correctly.
There is always variance in CPU MT usage so expect +-2% no problem, in the end for all cars the usage is about equal. ST usage is always 94.x - 95.x% on all cars. There is no fps limit, it should run as fast as it can but AC is poor at that.
4690K 4.5GHz + 1060 6GB 2.0/8.9GHz + 16GB 2.4GHz CL11 has been used, AC runs from SSD same as OS.
Test preset for CM used: http://acstuff.ru/s/3xCY
Code:
Mugello, mid clear, optimum, Pro+blankets, 23 opponents 99+-1% aggression 100% starting last, penalties + jump start to pits, temp 26C, wind 0km/h.
All skins available and used.
Many AC apps open as I normally use plus GPU app, plus MSI afterburner with rivatuner overlay, plus older ReShade. Fullscreen exclusive 1080p, custom graphics and all other settings as I normally use.
Values read out during start lights countdown from driver eyes view until cars start moving and smoke ensues from tyres hence the variance noted.
Car FPS GPU CPU ST/MT
-------------------------------------
1700 53-1 33% 95% 62%
Clam. 52-1 31% 95% 64-1%
165 59-7 36% 95% 66%
SF70H 63-2 39% 94% 62%
311 63-1 37% 95% 64%
Cobra 74 42% 94% 60%
XbowR 67-5 35% 95% 60%
5cup 67-4 39% 95% 62%
-------------------------------------
1700 test mule, aka 1700 with supercharger in it's physics:
1700 53-2 33% 95% 64% 5cup LOD distances
1700 59-7 34% 95% 67-1% low LOD
1700 53-2 33% 95% 63% power+throttle LUTs minimized to poor resolution
1700 54-2 33% 95% 64% power+throttle+aero+tyre LUTs all only 2 values
Cannot test skin format as again I don't have DDS tools installed for Photoshop to be able to open and save them, somehow I always have to install DDS support almost every time I want to work with DDS as I only work with it rarely. Skins seems to be DXT1, Kunos seems to use DXT5 which as far as I remember might have higher compression and what I would normally use, though DXT overall is a lossy format, the advantage of it is that it can be straight copied and used by the GPU and doesn't need converting such as PNG and JPG textures do. I do not see the need for the number plate be this large, could be half the size but I doubt that's an issue anyway.
As you can see GPU usage is very low around 30-40% during the test and CPU single thread as reported by AC is around 95% no matter what car is used. The total CPU usage is around 60-65% for me on all cars again. FPS varies but I have not been able to find out why.
IMHO it may do with car paint differences, also clamshell has car paint reflections (didn't test that one, brb, added), small possibility would be textures, most likely it's the detail offered on Caterhams both external AND internal (dash, seats, everything in driver view), even Kunos cars can vary greatly.
If I had to guess I would say it has to do with car paint reflectivity and reflections in general as most likely Kunos is still computing those on CPU, at least that's how it was done in the past in graphics engines when it came to reflections and other similar lighting stuff as far as I remember. It's definitely being CPU bottlenecked in the AC render thread by something, reflections or drawcalls etc.
The hitbox should not be crazy, you can check that in CM. Collisions again are CPU processed as far I'm aware in every game.
I do not know or have tools right now to go debug AC graphics if it would even allow me, only to see how many polygons are being used by each car and total in the scene. Ben will have to compare and answer that if the Cats have similar poly count - complexity in geometry compared to other stock AC cars. To me it doesn't seem crazy poly count and LODs work well. Even with LODs put low so that quality is poor the gain in FPS is minimal. IMHO it is not a poly count issue.
Unfortunately AC doesn't offer much when it comes to dev apps to see where the game is bottlenecked beside the never ending CPU single thread bottleneck.
BTW the RD CODE block is not working as a proper CODE block it doesn't keep tabs correctly.
There is always variance in CPU MT usage so expect +-2% no problem, in the end for all cars the usage is about equal. ST usage is always 94.x - 95.x% on all cars. There is no fps limit, it should run as fast as it can but AC is poor at that.
4690K 4.5GHz + 1060 6GB 2.0/8.9GHz + 16GB 2.4GHz CL11 has been used, AC runs from SSD same as OS.
Test preset for CM used: http://acstuff.ru/s/3xCY
Last edited: