Issue with Polycounts/Frame rate
by Rapid Fire T · in Torque X 3D · 09/13/2009 (5:51 am) · 4 replies
Currently working on a project, and I ran into an issue with the frame rate.
With only 2 custom models, the framerate was marked as 9 fps by the frame rate function.
We removed all components, but the DTSRenderComponent. It was still at 9 fps. We added 5 models, and it went to 9 fps. That's when we learned the framerate counter has an artificial lower limit of 9.
After doing some testing, I found that so long as the models were off screen, the framerate would climb, up to 17 for 1 model.
I did a bit more testing with models both ours and Torques, and came up with these numbers.
Note: this was tested on the Xbox 360 in Release mode. Debug numbers were 25% Slower.
box.dts: 12 Polys (assumed, as it's a box; GG's Model)
If the framerate reads as 9 or less, I will just mark it as <9 because we cannot get an accurate frame rate at this time. It also seems that 249 is some sort of artificial cap.
Empty Project = 0 Polys FPS: >249
10 box.dts = 120 Polys FPS: >249
100 box.dts = 1200 Polys FPS: 200
500 box.dts = 6000 Polys FPS: 41
750 box.dts = 9000 Polys FPS: 28
1000 box.dts = 12000 Polys FPS: 20
2000 box.dts = 24000 Polys FPS: 10
Keeping in mind, this is with 0 components outside the DTSRenderComponent, no terrain, and no skybox. There is no game logic, and this is with all models on the screen, but with no textures using only simple material.
This means that before factoring in time for game logic, which is going to be quite significant, the maximum number of polygons on the screen before dipping below 30 FPS on the Xbox 360 is approxmiately 9000 polygons. This number seems somewhat low. I'm curious if others are finding these sorts of numbers or if we are possibly doing something wrong (though there is not much in the scene).
For such an empty project, why is it so low?
With only 2 custom models, the framerate was marked as 9 fps by the frame rate function.
We removed all components, but the DTSRenderComponent. It was still at 9 fps. We added 5 models, and it went to 9 fps. That's when we learned the framerate counter has an artificial lower limit of 9.
After doing some testing, I found that so long as the models were off screen, the framerate would climb, up to 17 for 1 model.
I did a bit more testing with models both ours and Torques, and came up with these numbers.
Note: this was tested on the Xbox 360 in Release mode. Debug numbers were 25% Slower.
box.dts: 12 Polys (assumed, as it's a box; GG's Model)
If the framerate reads as 9 or less, I will just mark it as <9 because we cannot get an accurate frame rate at this time. It also seems that 249 is some sort of artificial cap.
Empty Project = 0 Polys FPS: >249
10 box.dts = 120 Polys FPS: >249
100 box.dts = 1200 Polys FPS: 200
500 box.dts = 6000 Polys FPS: 41
750 box.dts = 9000 Polys FPS: 28
1000 box.dts = 12000 Polys FPS: 20
2000 box.dts = 24000 Polys FPS: 10
Keeping in mind, this is with 0 components outside the DTSRenderComponent, no terrain, and no skybox. There is no game logic, and this is with all models on the screen, but with no textures using only simple material.
This means that before factoring in time for game logic, which is going to be quite significant, the maximum number of polygons on the screen before dipping below 30 FPS on the Xbox 360 is approxmiately 9000 polygons. This number seems somewhat low. I'm curious if others are finding these sorts of numbers or if we are possibly doing something wrong (though there is not much in the scene).
For such an empty project, why is it so low?
About the author
#2
John K.
www.envygames.com
10/13/2009 (6:11 pm)
Tilan, very interesting findings. The next release made a lot of performance improvements, including how vertex and index buffers are managed. There's also a big difference between Release and Debug builds. Can you please email me? My email address is in my profile page.John K.
www.envygames.com
#3
10/15/2009 (3:15 pm)
I also noticed that the performance is not very good overall on Xbox as well as on PC. I really hope the performance improvements are significant.
#4
As far as number of objects, I had a high poly model that was at about 5000 Polys, and 2 of them produced similar numbers to the 750 Boxes. I originially wrote it up with multiple models, but the other models I was unsure of their exact poly counts.
And this was done in a Release Build, as well as being not through the XNA Game Studio Connect, as I found those reduced the speed even further, as you stated.
I'm currently working on a few other things at the moment, but If I get the time I'll test .X and .FBX, though I'm more keen on seeing how the upcoming performance improvements turn out.
10/23/2009 (1:09 am)
I didn't do any testing with .FBX or .X files yet, though that could help.As far as number of objects, I had a high poly model that was at about 5000 Polys, and 2 of them produced similar numbers to the 750 Boxes. I originially wrote it up with multiple models, but the other models I was unsure of their exact poly counts.
And this was done in a Release Build, as well as being not through the XNA Game Studio Connect, as I found those reduced the speed even further, as you stated.
I'm currently working on a few other things at the moment, but If I get the time I'll test .X and .FBX, though I'm more keen on seeing how the upcoming performance improvements turn out.
Torque Owner Ezra
have you tried using FBX or X format models and the T3DXNARenderComponent ?