Benchmarking OpenCL by raytraycing on ATI+Nvidia(OpenCL)

[mitch_de]

smallluxGPU (Snow Leo only- NV9400M...GTX2xx, ATI 48xx)

Does raytraycing by GPU, GPU+CPU or CPU only

Very complex (real life) computing, so less advantage for weak GPU than running more low level OpenCL Demos.

Does much better hybrid (CPU+GPU) than Galaxies - even an NV9400M make sense by helping the CPU to an 15-20% speedboost

Uses ALL openCL GPUs (up to 4) parallel which it find.

Also works with ATI 48XX GPUs (most OpenCL Demos had lot of problems running on ATI 48XX)

For GPU only benches, select use GPU only , otherwise it will use CPU + all GPUs

Speed: k samles/sec (higher is better) , use key h to switch help screen on/of

DL Newest 1.5.2 Version here !

http://www.macupdate.com/info.php/id/33632/smallluxgpu

 

posting results here.

 

Known results:

GTX 260 GS 16 sec

8800GTX 28 sec

9600GT 38 sec

9400M (Macbook) 156 sec

[mrheat]

mitch

 

please clear up your inbox

[mitch_de]

Done

[mitch_de]

31.01 : Updated to 1.2B2 and also as an normal (easier to run) app avialable ( terminal app now second download link)

 

http://www.insanelymac.com/forum/index.php?showtopic=182874

[mitch_de]

15.02 : NEWer smalllux with GUI !

Highly optimized OpenCL code now, 2 scenes selectable,..

Goal of that smalllux is to be later/in future an add on to the luxrender (Mac OS X also, raytracing app)

 

DL : as always from Post #1 in the OpenCL Bench/Demo Thread

http://www.insanelymac.com/forum/index.php?showtopic=182874

[C3sound]

I have read in video game audio they have a technique called ray-tracing. It has to do with the ability to compute reverb volumes with convolution and other high quality sounding reverbs on the fly. Usually it is done with volumes, rectangular boxes with audio defining characteristics, such as "metal hallway" or "open area concrete ground" etc. This is done away with, and replaced with on the fly computation based on the actual game distance away from buildings/objects/etc and object material. It was first used in the xbox 360 title "Crackdown"

 

Just thought I would throw it out there since the word came up.

[mitch_de]

Now new version 1.4B1 is out (DL 1. Post) which has enhanced Material support, even better lights support .

Could now also handle materials which are glas / gold ...

Because of changes in the scene files there are 2 new scenes : classroom and Little Buddha

 

 

Maybe also interesting for tests if OpenCL 10.6.3 is faster than 10.6.2 , which will available next 2 weeks.

[mitch_de]

Updated to V 1.4.3 !

[mitch_de]

updated to 1.4.7 : now easy handable Benchmark function = shows time in seconds how fast luxball scene was rendered (without shown renderwindow).

[mitch_de]

updated to 1.5.2 : Now shows GPUs used in the new benchmark mode (time used for an fixed scene).

as always

http://www.macupdate.com/info.php/id/33632/smallluxgpu

[kozlek]

For Gainward GeForce GTX 260 GS, 896MB:

 

[mitch_de]

Thanks !

Is that GTX 260 GS that with less shaders/units (196 insted of 216) ?

Or what means ..GS ?

[mikoffski]

HD4850

[mitch_de]

GREAT !

Your time will be included in the next version as reference.

Funny that driver detects it as 4870

Different Units MHz will be shown in real ATI 4870 i guess.

[kozlek]

Thanks !

Is that GTX 260 GS that with less shaders/units (196 insted of 216) ?

Or what means ..GS ?

 

It's 216 units. GS - Golden Sample, don't know what does it mean for Gainward

[morfy]

My 8800gt. Only GPU test.

[render^]

XFX 1gb HD 4890 on 10.6.3

 

highest: 10.7 sec

lowest: 9.5 sec

[bbeepp]

Mine 4870 on 10.6.2 scores between 8.9-10.2 sec

[Goron]

Scored under 10.6.2 ... 10.6.3 to follow:

 

 

EDIT:

 

29,1 sec. in 10.6.3

[mitch_de]

Thanks for the new 4870 + 4890 rates.

How much MHz (for the gpu units) are shown ?

Also , the faster the gpu itself the more the other times vary - (compiling the OpenCL on the fly, loading the scene stuff, ...). My 8800GTX give real stable time sec from 27,5 - 28,5 sec , less % diff than the fast ATI 48xx (also 1 sec diff between runs).

Therefore next version will have an Highend Benchmark GPU Mode, which does double the scene size = doubles the time needed = the overhead (around 1 Sec) for running get less % diff between runs which will be twice as long = more excat measurement of the highspeed cards 48xx and GTX 285 .

[bbeepp]

How much MHz (for the gpu units) are shown ?

 

Mine shows 750 Mhz.

[mitch_de]

Now updated to 1.5.3

users with very fast GPUs lile ATI 48xx and GTX 285 may run again the new Highend Benchmark Mode which takes 2 times of the normal (now called Midrange Benchmark Mode).

So differences will be seen better than with the 8-10 sec results.

[kerr]

 

1.5.3 highend benchmark

[mikoffski]

1.5.3 highend benchmark

 

HD4850 on 10.6.3 64bit

[mitch_de]

Thanks for new High Restults (needs calulated exact twice the time as Midrange/old Benchmarkmode).

Overhead time is now not as much in % runtime as before.

ATI 48xx performs really good since 10.6.2 !

Would be interesting in which area the 4850M (320 shaders) "play" , and of course the GTX 285.

In the win/linux world which also try smallluxGPU (beside as usage as Blender plugin for all platforms) get extreme fast rendering using ATI 5770 /58xx gpus ! Also Fermi will be much faster - NV had very long the benefit of much better OpenCL drivers (faster, less bugs) + CUDA (vs ATI STEAM) and so beside their less GigaFlops (less shaers/units) got as fast as ATI (hw faster, drivers for OpenCL slower).