OpenCL Oceanwave Bench and (new) CompuBench CL

[mitch_de]

Ok thanks for the help you gave to me... but the luck is not with me =( another error:

I added the RunMe script (MANY THANKS to the developer STLVNUB) into the zip (Download end of site 1).

Redownload that OpenCL_NO... and unzip it. Then double click on the RunMe.

Your error 4:

I found out that the RunMe Script fails if you have spaces or special characters in the path to that folder.

Copy the folder somewere else (Desktop) wre you dont have subfolders (path) with spaces in the pathname.

 

@jaymz60:

Thanks for your results. You posted 2 different fps results. Does this mean its 2 cards (your profile GTX 580 & GTX 285) so the lower fps is GTX 285 ?

[oSxFr33k]

How is everyone getting on average just under 3ms and just under 400 fps? When I first start it up I get just around 6.5-7ms and just under 200 fps. Are you doing something right after it starts to get the higher frame rate and lower milliseconds?

 

ESC or pause maybe? There is no way it starts up with those values. My scores are very high in all the other benchmark applications such as Luxmark Luxball, Geekbench, Cinebench, Novabench and Open GL viewer with an average score of 2400-2700 fps.

[jaymz60]

New test GTX580 other one was also GTX580

[mitch_de]

For me your script works.

Can you run ./OpenCL_OceanWave without the script or maybe same error, whoch means its not an problem of the script, OpenCL fails with your gpu for that bench in general.

Also it can be an problem of intructions for the cpu (used also of course) by some cpu features uses which your (older cpu type) didnt habve = illegal intruction. I will upload an other OpenCL_OceanWave compiled with less modern cpu features compile options.

[mitch_de]

OK , then its the Oceanwave app itself in conjunczion with your cpu type features (SSE3, SSE4,....).

I will compile an other version soon.

[mitch_de]

Try out that new build of mine (no SSE4, SSE3Ext features anymore) , also disabled multisamling (the only change to orig. Apple source).

OpenCL_OceanWave.zip

[mitch_de]

Ok, can you upload your unmodified build to use for all. Perhaps disabling multisampling has some negative side effect on some gpus. I have no freezes when i move the window.

[mitch_de]

I first used an version with no changes in the Apple Source.

This versuin runs at my 9600GT much slower, 79 fps (orig. Apple) vs 102 fps (no multisampling / glFlush).

I think you mean with freeze some display corruption and display problems for the fps stat text when moving the window around with the no multisampling version.

I add an glFlush OpenGL command in the sourcepart of the rendering (window with text) , so you can move the window now without or much less problems AND fps stays much higher than with orig. Apple source.

If you let run more than 1 OpenCL App, it is normal that the result can be bad or give errors because the gpu gets overloaded in in some cases.

 

I uploaded my new version (no multsamling, older cpu work, glflush for less problems) at page 1.

OceanWave_NO_Multisampling_GlFlush.zip

[quattrofx]

Hi Mitch_de and OS-X Family!

 

GTX 285 Results:

 

Pause:

 

Cheers

QuattroFX

[mitch_de]

Thanks !

The 150 fps are not well readable in the rendering window - Title fps not affected.

Had you used lastest Version from today (always on end of page 1) ? Sometimes screenshoot slows FPS and/or make some disorted fps text.

Please try lastest (Page1) version with glFlush, which gives much less problems moving the window & i show FPS stats only in the window title.

 

For my tests, fullscreen screenshoot works much better than using area selectable (Mouse selectable) screenshoot, which does reduce FPS significant -at least on my C2D system.

 

For me the FPS rised from 101 FPS to 105 FPS (removed OpenGL text rendering in the window).

Compared to orig. Apple source an huge fps gain : 79 fps orig. vs 105 FPS no muiltisamling no fps Opengl rendering.

The slower the card the more diff between org. Apple source and no multisamling / no fps rendering, because more gpu power left to do the OpenCL (beside the OpenGL rendering of the waves & co).

The latest version also removes the irritating OpenCL compiler warnings from the terminal, only errors will be shown.

OceanWave_NO_Multisampling_GlFlush.zip

[jaymz60]

Ran your test could not be any easier good work

[mitch_de]

WOW - 546 FPS now with GTX 580 (no multi, no fps stats in the rendering window) - NR 1

 

to all : If you have time please update (post) your fps with new version, if the new version (DL page 1 end) gives more than 10% diff to your "old" fps result.

[quattrofx]

Hi Mitch_de,

Here are the new GTX 285 results:

 

Pause:

 

slightly difference, as you sad.

 

Cheers

QuattroFX

[Badeendje]

Tested again with the updated drivers from the MBP package, going from 7.12.9 to 7.14.5, I got 1 fps more with my HIS 6870 with the old version. Here some results with the new version, a bit lower.

 

Normal mode:

 

Pause:

 

Old version for comparison with old screenshot:

[mitch_de]

Thanks. I think the AMD/ATI OpenCL Update is an bugfix and not effect speed.

Some AMD/ATI gpus had, even with 10.7.2 problems with some OpenCL functions, like the now waves bug with OceanWave. Maybe that updates helps to avoid such bugs. Nvidia OpenCL drivers seems to be some steps "in front" - less bugs and didnt got updated by this update.

[np_wski]

Hmm.. Don't know, maybe something wrong? I get this messages at terminal window:

np$ ./OpenCL_OceanWave

-------------------------------------------------------------------------------

Multiple GPUs found

Using device[0] for compute: NVIDIA GeForce GTX 580 OpenCL 1.1

Using device[0] for rendering: NVIDIA GeForce GTX 580 OpenCL 1.1

-------------------------------------------------------------------------------

Connecting to NVIDIA GeForce GTX 580...

-------------------------------------------------------------------------------

Building compute program...

-------------------------------------------------------------------------------

Building compute kernels...

OpenCL Build Warning : Compiler build log:

<program source>:213:5: warning: unused variable 'c'

fftKernel8(a+0, dir);

^

<program source>:61:9: note: instantiated from:

float2 c; \

^

<program source>:276:5: warning: unused variable 'c'

fftKernel8(a+0, dir);

[.. &ct]

 

With "OceanWave_NO_Multisampling_GlFlush" FPS varies from 400 to 600. It's shot of highest value.

 

It's "OceanWave_NO_Multisampling", about 500fps:

 

slightly overclocked GTX-580 (MSI "Lightning")

[mitch_de]

Thanks!

FPS vary may happen by using screenshoot. Normally (without any other programm running / no mouse move) should be stable fps (+/- 5 fps in your case of 600).

 

The compiler warnings doesnt matter - i had removed them in the lasted version, should not shown anymore in the terminal window.

Please check first posting / DL.

 

PS:

Macs benched by robart / http://www.barefeats.com/index.html

On the 2010 Mac Pro hex-core 3.33GHz with Radeon HD 5870, I get 246 fps.

On the 2011 iMac quad-core 3.4GHz with Radeon HD 6970, I get 315 fps.

On the 'late 2011' MacBook Pro 2.5GHz with Radeon HD 6770, I get 143 fps.

 

iMac with 6970M performes very well !

[mitch_de]

NEW:

I made an GUI app for oceanwave. DL first page .

- uses latest ocenwave console app

- has version info - more easy to see if you use latest version - now 1.0

- shows CPU+ GPU information

- can submit fps results (including CPU + GPU info) by mail after benchmarking

- shows max reached FPS after benchmarking

- has an buildin log and results tab (filled with already known FPS results)

[mitch_de]

Got

HACKINTOSH MacPRO 4,1 GeForce GTX 260 1296 MHzCPU Intel® Core™2 Quad CPU Q9400 @ 2.66GHz 2666 MHz RESULT: MAX= 42.08

by new GUI version. THANKS

Really low OpenCL speed for that gpu type. 8800 GT performs at least twice as fast, with lower hw gpu computing speed.

Maybe here some AGPM problem (GPU runs not an fastest gpu GHz at load). IN this case also OpenGL speed would be low.

[fwood]

Nvidia Quadro FX 4600 on Q9550@3,4ghz = 140,58 FPS

[mitch_de]

New V1.1 ( DL Posting 1)

- speed SAME (fps results stay valide / same to V1.0)

- added gpu vendor pic

- updated results list

- added ESC to finish in the benchmark window (some waited for ever to finish )

- added max fps shown on the main screen after finish (ESC) benchmark

- changed app icon to official opencl icon

[mindlessmissy]

It's funny how I manage to get way higher values when more than one display is connected ...

[mitch_de]

It's funny how I manage to get way higher values when more than one display is connected ...

Normally should not happen, only the normal measure diff between runs on same system is OK (1-3% diff betwenn runs).

Maybe your GPU / drivers have some problem with the gpu powermanagement - 2 dsiplays get full Ghz, 1 not.

What card do you have and whats the diff between 1 Disp and 2 Disp?

The only scenario which can happen to get more FPS is, when you have more than 1 OpenCL gpu - then both gpus are used, independen of how many displays you have. But that also can be different by gpu type and the drivers.

[Gotier]

With Lion 10.7.1, I have 139 FPS with Sapphire ATI 4870 Toxic 780MHz / Intel E8400 3.6GHz

[mitch_de]

With Lion 10.7.1, I have 139 FPS with Sapphire ATI 4870 Toxic 780MHz / Intel E8400 3.6GHz

Thanks for all fps result submits - i updated fps lsit (Posting 1) and will do that in the app also for next version.