New dynamic branching demo

Humus

Crazy coder
Veteran
There's a new demo on my site showing the benefits of dynamic branching.

dynamicbranching3.jpg


Download here
 
but performance improvements by enabling dynamic branching is only expected to happen on Radeon X1300 and up and possibly also GeForce 6 series and up (haven't tested) since earlier chips do not support this feature.

Er...I'm preeetty sure that the Xx00 series supports dynamic branching too.
 
Cool demo :p I got about 25-33FPS on my laptop with mobility 9600 64MB. Turn on AA at 4x or turn off is no different (or maybe it does not work on my card). Running around the demo the frame rate is still in 25-33FPS. Setting DB or not, it is no different as my hardware not support it. And I keep getting this error prompt while change setting on F1 menu

Error
Varying IVec read in fragment shader, but not written to in vertex shader.
Link failed.

And then it exit program. Also, I cannot change resolution to other than 640x480.
 
Without branching it runs a tad faster on the 7800GTX here; when I use shadows it turns out so dark that I can only see the animated lights. W/o shadows it's fine. ForceWare 84.25.
 
X1900XT @ 690/790 here, my CPU is only P4 3G though. Driver is Chuck.

1680x1050 4xAA 16xHQAF
w/ branching 145-165 fps
w/o branching 140 fps

1280x1024 4xAA 16xHQAF
w/ branching 210-230 fps
w/o branching 200 fps
 
Ailuros said:
Without branching it runs a tad faster on the 7800GTX here; when I use shadows it turns out so dark that I can only see the animated lights. W/o shadows it's fine. ForceWare 84.25.

Numbers please? (I have no nV card at the moment)

EDIT: would just like to know if X1x00 really does much better than the somewhat comparable nV counterpart with branching
 
_xxx_ said:
Numbers please? (I have no nV card at the moment)
6600GE (450/1000) , iP4-3.0, 1024x768, noAA (lots of background apps including running VMware client)

With Shadows ON, everything is very dark, only flying lights visible... something seems broken
Branching On, Single Pass On, Shadows On is + + +, off is - - -
+ + + 78fps
+ + - 88fps (shadows off make the picture ok, 10fps boos)
+ - - 69 (sinlge pass IS good! drops from 88 to 69)
- - - 72 (no branching is faster)
- + - 107 (no branching + single pass is best)
So, in optimal case (Single pass, no shadows), no branching is much faster (107 vs 88)
With more than 1 pass the difference is much smaller (72 vs 69)
 
at home testing, getting 105-150 fps with branching on a 7800 gt, without branching 105-150 fps

drivers 84.25

settings 1280x1024

single pass on

single pass off it goes down to 50-80 fps

Shadows don't seem to be working though when turned on, well they are working, but the lights die is what it seems, same thing that happens in the ogl demos in the ATi sdk.

84.45 drivers
getting 120-170 fps with branching on a 7800 gt, without branching 115-150 fps
 
Last edited by a moderator:
16,4% average performance hit when Branching is enabled on my 6800GS (both for AA and no-AA modes) at 1280*960.
Still "playable", though. :D
 
Last edited by a moderator:
Sorry to snap-off the topic, but I just notice that most people with NV card seem to have problem with shadow on the demo, but the ATi user are not. While this seems opposite in the games (like FEAR, Oblivion, etc) which indicates NV users having no problem with shadow things but ATi users do. Is this possibly mean that the method to implement shadow on NV and ATi is different so that it cause problem?

On my laptop (Mobility 9600 64MB) when single pass off... FPS drops to 11 at 640x480 :cry:
 
So is it safe to assume that nV's branching in 79xx is not all that bad as thought (looking at pharma's results)?

pharma, the first set of numbers shows it to be faster with branching? You sure you didn't switch the numbers?
 
_xxx_,

I redid the first set and the results seem to be the same whether branching is enabled/disabled (on average). I'm getting:

1920x1200 4xAA 16xAF
Shadows enabled, single pass enabled, animate lights enabled:
w/ branching 495 - 535 fps
w/o branching 493 - 540 fps

A nice option would be to export the average fps to a text file, instead of trying to eyeball the range of fps.

Pharma
 
I did not run the tests with "Full Screen" enabled, as the third digit (fps) was cropped and could not get an accurate reading. A quick test at a lower resolution showed much lower fps when "Full Screen" was enable. When I get the time, I will rerun the tests and revise the posted results above.

Pharma
 
pharma said:
I did not run the tests with "Full Screen" enabled, as the third digit (fps) was cropped and could not get an accurate reading. A quick test at a lower resolution showed much lower fps when "Full Screen" was enable. When I get the time, I will rerun the tests and revise the posted results above.

Pharma
afais when not in full-screen the app can run only at 640x480
 
chavvdarrr said:
afais when not in full-screen the app can run only at 640x480

I don't think this is correct -- the objects & textures within the app window are too highly detailed. To test, I changed my monitor resolution to 640x480 and ran the app -- textures were horrible and lacked detail. Could not make any adjustments on the menu (F1) as it was 1/2 off screen.

I'm sure Humus will enlighten us. When I get back later will rerun the results.

Pharma
 
Back
Top