New instancing demo

ninelven said:
Would it be possible to alter the demo to make it stress the gpu more?

I'd probably want to write a whole new demo then based on something with a bit larger instances. When dealing with this many objects, even the simplest computations will easily make you CPU bound.
 
using Cats 4.9 Beta, DX9c, WinXP Sp2:
1.290
2.255
3.275
4.88
5.42
Application crashes if I try to see the desktop.
 
Humus said:
There's no difference between NV40 instancing and R300/R420 instancing, except that MS refused to add a caps bit for it and arbitrarily linked it to the entirely unrelated VS3.0, and therefore it must be queried through a FOURCC format on VS2.0 hardware.

Do you know for sure that the implementation in hardware is close to equal? Just asking! 8)
 
6800GT @ 446/1.15

1-335
2-415
3-357
4-182
5-94

P4 2.6 @3.35, 2X512 mem ASUS P4C800E, memory 248(5/4) CPU & GPU water cooled

9800Pro @ 412/712(356)

1-370
2-308
3-345
4-103
5-50

Athlon XPM 2400+ at 2.3 (200X 11.5), 2X512 mem, Shuttle AN35 Ultra, VGA Silencer
 
Alstrong said:
Isn't that just a physics demo?
Yes, but (IIRC - I've only read about it on gdalgorithms) it can
a) compute the physics and store them on disk to be played back later (making it entirely GPU bound), and
b) support instancing (at least for NV40, not sure about ATI's cards)
 
Chalnoth said:
pocketmoon66 said:
You could at least describe what this is, pocketmoon.

Sorry :oops:

From the Author:

"An excellent test for instancing is 'building explode'

Select this demo, it will run very slowly because there are 17k rigid
bodies in the scene. Hit the 'P' key to pause physics. Now you will
just be rendering bound"

Might be interesting to compare FPS across vendors using this rather than Humus's demo which does some extra tricks (packs 4 particles per instance) which may or may not work as will on NV hardware.
 
Select this demo, it will run very slowly because there are 17k rigid
bodies in the scene. Hit the 'P' key to pause physics. Now you will
just be rendering bound"

Might be interesting to compare FPS across vendors using this rather than Humus's demo which does some extra tricks (packs 4 particles per instance) which may or may not work as will on NV hardware.

Using that demo and pressing P right at the beginning before any action happens:

Resolution: 1024x768

w/ instancing: 89fps (My refresh rate is 85hz, I found no way to disable vsync, so I'm not sure if that's a usable number...)
wo/ instancing: 19.8 fps

Resolution: 1280x1024
w/ instancing: 56fps
wo/ instancing: 18fps

My machine:
Athlon 2500XP @ 2600XP speeds
MSI Radeon 9800pro stock
1024GB RAM
WinXP SP2

edit: I used default settings for quality/performance in the ATI CP[/img]
 
Resolution: 1024x768
w/ instancing: 130 fps (hp) - 130 fps (hq)
wo/ instancing: 7 fps

Resolution: 1280x1024
w/ instancing: 130 fps (hp) - 118 fps (hq)
wo/ instancing: 7 fps

Resolution: 1600x1200
w/ instancing: 112 fps (hp) - 105 fps (hq)
wo/ instancing: 7 fps


My machine:
Athlon 2500XP @ 2GHz
Gainward 6800U 400/1100
512 MB RAM
WinXP SP1
hp = high performance
hq = high quality
AA & AF by Application
 
I'm asking this to try and learn something and I know the demo was ment to show off instancing. But couldn't you have used a index buffer to reduce the amount verts sent over by 1/3. Also point sprites sence these quads are always facing the screen. I'm looking to incorparate particles into my first engine and this demo has me worried about the performance hit.
 
flick556 said:
I'm asking this to try and learn something and I know the demo was ment to show off instancing. But couldn't you have used a index buffer to reduce the amount verts sent over by 1/3. Also point sprites sence these quads are always facing the screen. I'm looking to incorparate particles into my first engine and this demo has me worried about the performance hit.

I'm using an index buffer already. Point sprites could be used if particles never are going to be larger than a certain implementation specific size. If point sprites didn't have that limitation, I'd be using it in all my demos with particle systems.
 
Humus said:
flick556 said:
I'm asking this to try and learn something and I know the demo was ment to show off instancing. But couldn't you have used a index buffer to reduce the amount verts sent over by 1/3. Also point sprites sence these quads are always facing the screen. I'm looking to incorparate particles into my first engine and this demo has me worried about the performance hit.

I'm using an index buffer already. Point sprites could be used if particles never are going to be larger than a certain implementation specific size. If point sprites didn't have that limitation, I'd be using it in all my demos with particle systems.

This is sad particles must be faster :( thanks for the reply
 
Back
Top