1  |   Order by index

Apr 26, 2003, 10:26pm

Recent experimentation has shown that a massive performance drop related to
collision detection/sliding ONLY takes place (on my system) while in D3D
(T&L or no).

Tests with D3D indicate that colliding with an object, even if no movement
is being made (face a flat surface at a 90 degree angle), FPS can often be
cut in half (60 drops to 30)
The same test with OpenGL and software indicate that no more than 1 FPS is
ever lost while colliding. I also noticed that sliding is generally much
smoother in OpenGL, even though OGL is mostly useless as far as speed and
accuracy of rendering.

This seems to indicate that the glitch which causes the FPS to plummet is
not actually in the collision code but (somehow?) related to the rendering.
Anyone able to confirm or deny these results?

UberMonkey

Apr 26, 2003, 11:10pm

How did you conduct these "tests"?

I don't think rendering can cause anything of the sort. The rendering
simply shows your view and the objects and everything whereas the
programming determines if you can or can't move forward any longer and such.
I don't think the rendering engine really gives any ouput to the programming
code, either, except if there is an error and for the frame rate.

-Dion

[View Quote] Tests with D3D indicate that colliding with an object, even if no movement
is being made (face a flat surface at a 90 degree angle), FPS can often be
cut in half (60 drops to 30)
The same test with OpenGL and software indicate that no more than 1 FPS is
ever lost while colliding. I also noticed that sliding is generally much
smoother in OpenGL, even though OGL is mostly useless as far as speed and
accuracy of rendering.

This seems to indicate that the glitch which causes the FPS to plummet is
not actually in the collision code but (somehow?) related to the rendering.
Anyone able to confirm or deny these results?

UberMonkey

Apr 27, 2003, 5:45am

[View Quote] I'll tell you exactly how. I teleported myself directly in front of a flat
object (in this case table2.rwx) facing at an exact 90 degree angle. I then
proceeded to walk directly towards the table until colliding with it. I then
stopped holding the forward key and waited for the FPS meter to stabilize.
standing FPS: 60
I then held the move forward key, which resulted in absolutely no movement
since I was at a 90 degree angle to the object, and waited for the FPS meter
to stabilize.
standing FPS while holding move forward: 30
Then I tried it with other objects. Always similar results.

I then repeated the test in all other video modes.
I found that it occurs more severely in D3D (regardless of T&L). While in
OpenGL, colliding with an object does not appear to affect FPS to as severe
a degree (often losing no more than 20% of fps) (though the overall FPS is
lower and there are a number of horrible z-buffering problems). The same
appears true in software, though an accurate determination here is difficult
considering the low FPS (6). I'm not necessarily implying that the rendering
has caused this glitch, however there must be some reason why the OpenGL
engine does not display this issue, yes? Or else the percent lost decreases
as the FPS decreases, due to the program taking less samples per second and
therefore doing less collision calculations overall. ... confirmed, D3D also
loses less at lower FPS. However testing 60fps in OpenGL drops to 45, while
D3D still drops to 30.

I dunno.


Monkey

Apr 27, 2003, 5:49am

Oi!
Here's an interesting one.
In D3D, starting at 93FPS, still drops directly to 30 and sits there.
Tried others between 60 and 100, still goes to exactly 30 fps.

Apr 27, 2003, 9:06am

"ubermonkey" <henry at wcalliance.com> wrote in news:3eab8b6f$1
at server1.Activeworlds.com:

> Oi!
> Here's an interesting one.
> In D3D, starting at 93FPS, still drops directly to 30 and sits there.
> Tried others between 60 and 100, still goes to exactly 30 fps.

Probably because of the setting to keep framerate above $foo FPS.

KAH

Apr 27, 2003, 3:27pm

However I was under the impression that this setting only affects the
visibility, which in all tests I have locked to 100m.

I just now tried setting it to 10 fps, however the browser's FPS still drops
to exactly 30 as long as it's over 30 to begin with (otherwise, it displays
similar behaviour to OpenGL, losing about 25% fps on collision)

[View Quote]

Apr 27, 2003, 3:35pm

A quick intall of 3.3 and a repeat of the same tests gives the same results.
IOW, this has nothing to do with 3.4 specifically.. whatever the problem is,
it most likely comes from an original botched sliding implementation that
has not been altered in 3.4.

(as a note to those complaining about sliding in 3.4 being unreliable on
terrain, try 3.3... it's even worse.)

Apr 27, 2003, 3:54pm

Negative. On my system, the collision detection lag occurs in all video
modes except software. It's quite noticeable for me, since I almost
never get framerates above 20FPS unless I'm out in the open.

Goober Specs
--
Pentium III 700MHz
384MB RAM at 66MHz
GeForce2 MX 32MB VRAM
56K Dialup modem
Windows 98 SE
DirectX 9.0
DirectX w/ T&L mode

[View Quote]
--
Goober King
Lagging like a mosquito in amber...
gooberking at utn.cjb.net

Apr 27, 2003, 7:06pm

I just did a repeat of these tests, and I lose no more than .1FPS (From a
starting point of 64.1) in D3d/T&L both while sliding along the object at
varying angles and pushing up against it perpendicularly. See the chart for
my full results. Sliding count is the LOWEST that I see, not constant.

Mode Standing Still Pushing Sliding
D3D/T&L 64.1 64.1 64.0
D3D 64.1 64.1 64.0
OpenGL 64.1 64.1 49.9
Software 12.1 11.8 11.8

Vis locked at 200 - pushing up against boardh4m rotated 90 degrees around
the X axis to act like a "wall". looking towards directional light source
over user defined light sources, over animated water, under a 3-layer sky
and pushing upwards on a hill.

No performance hit to speak of. What's your system configuration?

Test platform:
WinXP Pro w/ SP1
GeForce 4 Ti4200 128mb
Asus A7n8x Deluxe motherboard
AMD Athlon XP 1900+ (1.6GHz)
256mb DDR single channel at 133MHz
56k Winmodem
DirectX 8.1 (WinXP stock)

Running software:
WinXP
StyleXP service w/ custom WingCorp skin
mIRC
PC-cillin 2002
WinZip Quick Pick
PowerStrip
InterVideo WinCinema Manager
Winamp Agent
Daemon Tools w/ 4 virtual drives enabled
CloneCD w/ 8 virtual drives enabled
3Com network diagnostic tool
nForce Tray Options
Winamp 2.90 (Playing a variety of songs - framerate fluctuated during track
change, so those results were thrown out)
Outlook Express
AOL Instant Messenger
Windows Messenger not running - removed from system.

[View Quote]

Apr 28, 2003, 5:18pm

And Shamus just made a post in Beta saying that the slowdowns may be caused
by debug logging being turned on - which I do not have enabled.

[View Quote]

1  |   Order by index