Test a DX8.1 1.1 Vertex Shader?

F

flipper

Newcomer
hi all. first time poster, sometimes lurker here.

a co-worker and i are writing a DX8.1 hobbyist "game" that features "life-like" fish movement. we posted a request on Gamedev.net for some help in testing one of the app's shaders on various hardware so we could see what mods, if any, we would need to make. the guys at Gamedev gave us a good set of numbers but unfortunately many didn't post their "shader off" numbers, which is when the app manipulates the vertices itself and constantly streams a new set to the video card for every frame.

so we thought maybe a somewhat benchmark-oriented site might be interested in helping us out? if you are, the ~400KB app's available for download here. there's been a couple reports of failure and a known issue with the bounding box display that causes the app to crash with some drivers, but that's all the bad news that's been reported. what we're looking for are the number of fish it takes to just drop below 30fps, with the shader on and then again with the shader off. INS adds fish, cap A toggles the shader mode. cap H displays keys help.

thanks.

oh, and if you're interested, here's our current set of numbers, shader off sorted first then shader on sorted:
Code: 
CPU/CARD                     Shader Off    Shader On
2.4ghz/9700pro:              53            175
unknown:                     46            134
2.1ghz/geforce4ti4600:       20             77
1.7ghz/geforce4mx440:        17             27
1.0ghz/geforce2go:            7             10
2.4ghz/9500:                  ?             42
2.0ghz/geforce4ti4400:        ?             67
2.2ghz/8500:                  ?             54

CPU/CARD                     Shader Off    Shader On
2.4ghz/9700pro:              53            175
unknown:                     46            134
2.1ghz/geforce4ti4600:       20             77
2.0ghz/geforce4ti4400:        ?             67
2.2ghz/8500:                  ?             54
2.4ghz/9500:                  ?             42
1.7ghz/geforce4mx440:        17             27
1.0ghz/geforce2go:            7             10
 
hehe, probably not terribly interesting to add to the list, but i'm waiting for a phone call at work...

Numbers for Win2k, ATI Rage Mobility (M3), Mobile P3-850, 800x600
Shader Off:
3

Shader On:
3

:)
 
CPU: Athlon (Tbird) 900Mhz
MB: Abit KT7raid(kt133), 1GB 133mhz cl3 sdram, via 4in1 4.46
gfx: Geforce2 GTS 64MB, Detonator 40.71, DX9
OS: Windows XP pro, no service pack.

Screen resolution: 1400x1052x32@85Hz, 800x600 window (framerate was not affected by size of window)

Code: 
# of fish   1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20
Shader off 96 89 48 47 33 32 29 24 23 20 19 18 16 15 14 13 12 12 11 11
Shader on  96 94 49 48 47 40 32 31 28 24 24 22 19 19 17 16 16 15 14 13

The PC had over 7 days of up time when I ran this test, maybe I should have restarted it first... :?
 
thanks for the numbers guys.

Ichneumon, we think we can reduce the model's vertex/face count significantly w/o too much loss in quality, so it's good to see some numbers on the mobility. these are fairly high poly models compared to what's out there currently in commercial products. the models we're using have about 15 to 20 times more vertices/faces in them.

Thowllly, thanks for the breakdown by fish. seeing that halving of the fps for the 3rd fish just might help us write a better vb sizing algo.
 
flipper said:
thanks for the numbers guys.

Ichneumon, we think we can reduce the model's vertex/face count significantly w/o too much loss in quality, so it's good to see some numbers on the mobility. these are fairly high poly models compared to what's out there currently in commercial products. the models we're using have about 15 to 20 times more vertices/faces in them.

Thowllly, thanks for the breakdown by fish. seeing that halving of the fps for the 3rd fish just might help us write a better vb sizing algo.
Click to expand...

I can corroborate the extra drop with the 3rd fish. There was almost no drop in frame rate from one fish to two (heh, like 33fps to 32 fps)... but adding the 3rd fish dropped it by about 8 to around 24, and then in would come back up a couple in a few moments. then going to 4 fish only dropped it 1 or 2 fps again.
 
System: AMD XP+ 2.2Ghz actual [11*200], Epox 8RDA+, 1Gig PC3500-C2 (4-2-2-2) in Dual-Channel, WinXP+SP1, 1600x1200x32bit@75Hz desktop, 8X AGP, FastWrites On
Video: Radeon 9700Pro (324x310.5 default clock)
ATI Control Panel: VSync Off, MipMap MAX, Texture MAX, 8X-AF Qual, 6x AA, TrueForm OFF, 75Hz DX Override

800x600x32 Window

Shader On (Major Transition Points):
1 = 64 fps
78 = 64 fps
79 = 63.02/62.03 fps [355421/552684]
80 = 62.03 fps [359920/559680]
81 = 61.05 fps [364419/566676]
83 = 60.06 fps [373417/580668]
171 = 29.61/30.06 fps [769329/1196316]
173 = 29.54 fps [778327/1210308]

Shader Off
1 = 64 fps
30 = 64 fps [134970/209880]
32 = 48/54 fps
57 = 32 fps [256443/398772]
59 = 28.55/31.03 fps [265441/412764]
62 = 21.33 fps [278938/433752]


I'll try running with vsync enabled later if these numbers seem wonky to anyone. Let me know if you'd like me to keep adding more fish or try overclocking the card some or the cpu (some more).
 
Tried various modes, with application AA and AF settings - 800x600x32 windows - 1600x1200x32 full screen ... the numbers were pretty identical to 6xAA Q8xAF (given the randomness of the app).
 
TB2100+, GB VAX1394 (KT400), 512MB DDR@266MHz, Win2K SP3
R9000Pro, 128MB 275/250, Cat 3.2, DX9.0a, 4x AGP, FW off, Vsync off
1152x876x95Hz
texture&mipmap - Quality
Aniso - Application, Smoothvision - off

800X600X32 (d16)
number shader ON shader OFF
1 330 255
2 250 225
3 200 200
4 165 165
5 140 140
6 120 125
7 110 115
8 95 100
9 88 90
10 80 83
25 27-33
30 29-30
 
On startup:

Generic application error. Enable debug output for detailed information.
Click to expand...

Asus A7V266-E
XP 1800+
R9700Pro w/ Catalyst 3.2
Win2k + SP3
DX9.0 debug runtime
 
Ichneumon, thanks man. interesting pattern that, huh? we're going to re-run on the geforce2go we have here to see if the problem shows up with it as well.

BRiT, thanks for the thorough research. i should have passed along some info re: the setup. we tried to make the test so only the shader was a factor, i.e., it was independent of as many other things as possible. your testing sure helps us feel better that maybe we did okay.

hey chavvdarrr, that's another new card for us. thanks! those numbers look really interesting. notice how they're almost the same after the 3rd fish? what in the world is up with that 3rd fish!!!! guess we'll have to dig deeper than we thought.

thanks again all.
 
hi Hyp-X. the only thing i know of to cause that error would be if the fish model was not in the fish sub-directory? sorry if that's the case. i know my explanation of what to do was not very thorough.

if that's not the problem then i'm sort of at a loss since your setup is very similar to one we're using. and we used the DX 8.1 common framework for init to try and make sure we mucked up as little as possible.
 
hey BRiT. i just had a thought. those numbers start out as if the "sleep" mode was set. would you mind re-running with it off? it's the cap L key. no problem if not since the 30 fps numbers are the ones we're most interested in.
 
flipper said:
hi Hyp-X. the only thing i know of to cause that error would be if the fish model was not in the fish sub-directory? sorry if that's the case. i know my explanation of what to do was not very thorough.

if that's not the problem then i'm sort of at a loss since your setup is very similar to one we're using. and we used the DX 8.1 common framework for init to try and make sure we mucked up as little as possible.
Click to expand...

It works when I disable the debug version of Direct3D from the control panel.
 
I also had the sleep mode set. Here are the numbers with it off:
Code: 
# of fish    1   2   3   4   6   8  10  12  16  20
Shader off 212 107  74  56  38  28  23  19  14  11
Shader on  239 128  90  69  48  36  29  24  18  14
 
Heh... it appears I had sleep mode On as well.
whups.

Performance is much different with Sleep mode off on this system. (the mP3-850/ATI mobility m3). With shader off, the FPS fluctates by as much as 10fps (35-45) depending on the fish movement, where the performance is pretty steady with shader On. The numbers are rough averages of any fps fluctations...

Code: 
# of fish    1  2  3  4    
Shader off 40 35 28 23   
Shader on 46 38 31 26

Seems that incremental drop in fps after 2nd fish is gone with sleep off... though on this system the fluctiations are significant as individual fish movement has a lot more effect with a couple fish rather than with a couple dozen...
 
CPU: Celeron 566@850 (100 MHz bus)
MB: Abit BX6r2, 452MB PC133
gfx: Geforce2 GTS 32MB, Detonator 43.45, DX9
OS: Windows 2000 pro, SP2.

Screen resolution: 1280x1024x32 full screen, sleep mode off

Code: 
# of fish    1  2  3  4  5  6  7
Shader off 116 63 43 31 26 21 19 
Shader on   48 23 16 12 10  8  7
 
System: AMD XP+ 2.2Ghz actual [11*200], Epox 8RDA+, 1Gig PC3500-C2 (4-2-2-2) in Dual-Channel, WinXP+SP1, 1600x1200x32bit@75Hz desktop, 8X AGP, FastWrites On
Video: Radeon 9700Pro (324x310.5 default clock)
ATI Control Panel: VSync Off, MipMap MAX, Texture MAX, 8X-AF Qual, 6x AA, TrueForm OFF, 75Hz DX Override

800x600x32 Window, Sleep Mode Off

Listings broken up per 10 fish (1-10, 11-20, 21-27). Towards the end there, individual fish adding had very slight effect on fps.

Code: 
On : 1210, 964, 805, 705, 612, 550, 496, 451, 416, 384
Off: 1098, 670, 510, 400, 336, 296, 250, 222, 200, 187

On : 359, 335, 313, 295, 279, 265, 253, 240, 228, 219
Off: 172, 159, 143, 137, 126, 121, 112, 105, 100,  96

On : 210, 202, 194, 187, 180, 175, 168
Off:  91,  87,  84,  82,  77,  75,  73.85
 
athlonxp 2000+, 64mb 8500 (overclocked a little), 512mb pc2700, 1152x864 full screen.

Code: 
Fish        on         off
1           704        504
2           501        355
3           388        284
4           317        239
5           268        205
10          159        119
15          105         84
20           80         64
50         33.7       26.4
100        17.9      11.98
 
Hyp-X, thanks for the heads-up regarding the DX debug version. we have been "flying blind" here since we installed DX9 over DX8 and haven't bothered to install the DX9 SDK.

and thanks to everyone who re-ran with sleep mode off. and to Bambers and Crusher for the new set of numbers.

and i just want to say thanks again for all of this data. you've given us much more detail than we hoped to get. and after making some plots, which we did on a per fish basis thanks to you guys, we now have a pretty good idea where to concentrate our efforts.

although i have to admit that Crusher's numbers are confusing me. :)
 
I get the same results every time I run it, with nothing else running in the background. Actually the "shader off" numbers are probably lower than they shoudl be, since you're using software vertex processing, instead of re-initializing the device for hardware T&L when you toggle the shader off. IIRC vertex shader 1.1 on the GTS are actually done by the drivers in software on the CPU, so "shader on" mode should be pretty slow on this system.

What what confuses me is that Thowllly's Athlon can do so much better when it's only running 50 MHz faster and on a 33 MHz faster bus... and also the fact that he somehow was getting 200+ FPS with sleep mode off... which I can't even come close to. I turned off the 2x anisotropic filtering and set the mip-map selection to best performance in the drivers, but it didn't make any noticable difference. I don't have FSAA on, and that would be the only other thing I could think of that would slow it down.
 
You must log in or register to reply here.

Similar threads

D
101 Amazing earth facts! (Must read - test your IQ :))
Replies
0
Views
4K
Deepak
D
Back
Top