While waiting for N-KART demo, here it is a little fun for all. Don’t expect much for this demo, still in very raw mode, so no menus, the in-race position indicator doesn’t work (the lap counter does), etc.
Be careful, when you reach 125 Mph, the suspension is lowered and the engine gain momentum, the speed raises, but you can loose the control of the car easily.
I’m interested here mainly in performance, there are 17 cars racing at once, so post your fps stats if you can, I’m interested primarly in the start when all cars are in view.
On my G5+X800, I get about 24fps at the very beginning, then when all cars start it drops down to about 6-8fps and stays there, whether they’re visible or not. This seems much, much too slow…
I was wondering that the performance issue is related to having many particles and particle emitters at the same time, but I’m not sure.
Because I’m working in a Mac Pro with eight cores and a 8800GT, sometimes I cannot appreciate differences with or without some effects, so I updated the demo with particles off and you can activate it pressing “P”.
Please tell me if this make the difference, or I’ll must to look for in any other direction.
My goal is to put at the same time as much cars as I can and I’m trying to get the 100 figure, but I don’t know if the scripts are the problem, they are all made in Javascript.
I was wondering that the performance issue is related to having many particles and particle emitters at the same time, but I’m not sure.
Because I’m working in a Mac Pro with eight cores and a 8800GT, sometimes I cannot appreciate differences with or without some effects, so I updated the demo with particles off and you can activate it pressing “P”.
Also I changed the opponent car’s shader to “vertexlit”.
Please tell me if this make the difference, or I’ll must to look for in any other direction.
My goal is to put at the same time as much cars as I can and I’m trying to get the 100 figure, but I don’t know if the scripts are the problem, they are all made in Javascript.
for dreamora,
I made some other setups with large models and that don’t seem to be a problem, but I will test with a scenery made of separate parts.
About car reflections, they are not real reflections, only a premade cubemap, and yes, they were all active all the time (not anymore as I use now the “vertexlit” shader).
Unfortunately, not really…that bumped it up a little to 10fps, but that’s still at around 10X slower than it should be running on my machine, based on what I can see is happening.
What language they are isn’t relevant, but whether they are efficiently written is relevant.
Steady 45FPS now with and without particles
a start but I expect 300 FPS+ for this scene on my hardware.
As for you not seeing a difference.
After the long night I spent on working on our unity driven server and its serious CPU time burnthrough problem, I might have a very simple reason for this:
You have 8 cores and osx → unity uses 8 cores
I /others are using Windows → Unity on windows forces itself into a single core. so while your system is able to distribute physics, mesh handling, particle handling onto potentially 8 cores, we have only a single one to handle that. (the webplayer shows this behavior as well. On osx, the browser goes up to 85% cpu time, while my desktop remains <= 13% so a single core)
See the attached pic, don’t seem to be using the 8 cores, this is a capture with Unity and the game running (not in background).
Can you show me any Unity scene involving physics at 300 FPS+? Really I must be making things horribly bad, or I have some undocumented problem in my computer.
See the second attached pic, a simple scene, one cube and one camera, no scripts, no fx, no physics, no nothing and I only get 100 fps (?).
That indeed is interesting that it does not go up above that single core. But might be related to the editor, did you test it outside as standalone for example or in FireFox?
As for the 300FPS: My graphic card is seriously much more powerfull than yours (NVIDIA GTX280 1GB, comparable to Quad SLI+ of yours).
I tested the webplayer build once again, but this time in fullscreen.
Result:
With enabled particles it is 40-50 FPS
With disabled particles, it goes up to 130 - 220 FPS, average around 170
Thats more like what I expect with the offered visual quality actually but makes me wonder on what is going wrong with the particles that they kill the performance that effectively.
Perhaps you have to disable the particle emitters yourself, that unity does not disable emitters that are out of view (behind the camera).
Well, now I’m getting mad. I made a webplayer of the test scene (the one with only the cube in it), figures are:
Safari - 50 fps normal mode on open, 2200 fps in full screen, 3300 fps when “esc” from full screen.
Firefox - 57 - 220 fps on open, again 2200 fps in full screen and 3300 fps when escape from fullscreen.
dreamora wrote:
Yes, I look for CPU usage in the webplayer, and again only one core is working. Maybe there is some difference when in full screen mode, but I only have one monitor and cannot see the Activity Monitor at the same time.
Eric5h5 wrote
I supossed something like that, but thank you for the confirmation.
Off topic: I must confess I’m something like a half-brained sometimes, but How the hell can I put the name in the quote instead of the “quote” word?
I like the car physics, though it felt like after a minute or two of racing I dropped behind and absolutely couldn’t catch up.
On OS X, Unity still primarily uses 1 core…if I disable my second CPU, I only get a framerate reduction of about 15-20%.
