Jump to content

Luxmark v4.0alpha1 (Nov 2020) - OpenCL Benchmark


mitch_de
 Share

211 posts in this topic

Recommended Posts

  • 1 month later...

Which Scene with 135 ? Scenes (3 available) have much different result count! Perhaps you benched different scenes on Win/OS X?

I looked in the Luxmark resultbase and that 515 of AMD 5850 / WIN64 can NOT be the most complex room scene - maybe it was the scala (middle complex) scene?

AMD 5850 / WIN64 are around 300 (295-320) in that most complex room scene - no way to get 515 in that scene with that card.

But in middle complex scene (sala) AMD 5850 / WIN64 getting around 600.

 

Normally OpenCL Luxmark runs same or often a bit faster than WIN64 Bit native.

I get 754 with the simple Luxball scene and Nvidia 9600 GT - your 5850 should perform much better.

In middle complex scene (sala) i get 74 and most complex scene 41 (room) - you see much different results 754 - 74 - 41 , depend on scene you benched. An AMD 4830 gets 93 in room (9600 GT 41), yours should perform >= 130.

Also - newer OS X Version should perform better OpenCL on some AMD gpus. For Nvidia there is near no speed diff between 10.7.2 and 10.7.4 dev .

 

GREAT : With 10.8 DP3+update Luxmark again working - with ML DP1+DP2 LM 2.0 wasnt working correct (render preview damaged/wrong).

Link to comment
Share on other sites

Failed to give full specs. Sorry!

I am using a Laptop.

HP Probook 4530s

i7-2630QM

Intel HD3000. (no Open CL support)

 

eGPU HD5850 Extreme connected @x1 2.0

 

Well when running LuxMark v2.0 Scene=Sala Mode=OpenCL GPUs only

W7 64bit gives me around ~500

While 10.7.2 gives me ~169 (best score after several tries)

 

Is it the x1 2.0 bandwidth limitation.

Link to comment
Share on other sites

After I updated to 10.7.4 I have this eror:

OpenCL ERROR: clBuildProgram

Before 10.7.4 everything works fine

 

Same here..i'm investigating :worried_anim:

 

 

edit: well i went back to 10.7.3 cause i got tired.It runs fine now

ASUS 6850 sala 389-390

6850.jpg

Link to comment
Share on other sites

  • 1 month later...

Hi,

 

I have problems regarding OpenCL support. I use OSX 10.6.8 and a ATI HD5570 with 1GB amount of RAM, Core2Quad 2,5, 4GB Ram.

 

There are four problems regarding the test in Luxmark:

 

1. The amount of usable graphics RAM is wrong, only 512MB are usable. How can I change this? Do I need to use another ATI personality or bin patch the kext?

 

2. The processing on most scenes of Luxmark never ends, it shows these pixelated results. This happens also with CPU only. What is wrong here?

 

3. How do I enable OpenCL 1.1?

 

4. Also the amount of system RAM seems to be wrong.

 

Thanks for help.

 

luxmark.jpg

 

 

BTW: I cannot upload any image to this forum anymore. Why?

Link to comment
Share on other sites

1. Amount of OpenCL VRAM amount of VRAM on the card, thats normal and is limited by gpu type

2. OpenCL 1.1 is available only with 10.7, OpenCL 1.2 with 10.8+

 

Luxmark is an benchmark and runs all available scenes exact 120 sec long. So you must wait that time.

Because of speed diffs between different gpu+cpu systems the rendered image of the scene vary much in case of clean+fine vs pixels!

With single gpu card (not highend AMD 79xx) you will not get an clean image within 120 Sec benchtime. That would take much longer for an cleaner image (10 Min... hours) depends on gpu power + scene.

 

If you want to let it render longer (than 120 sec) time for an more cleane+fine image, you can set Mode to interactive ( gpu only).

 

Beside that benchmark, Luxmark uses slg to compute OpenCL .

Its used by Luxrender and Blender (with slg) in real life to render gpu accell raytraycing.

Visit http://www.luxrender.net/en_GB/index to get more informations.

Luxmark has focus on benching, not raytracing for production!

Link to comment
Share on other sites

Mitch, thanks. Is the openCL-renderer also included in Blender 2.63? Where do I get a version with SLG included?

 

I read that since OSX 10.7.4, there is a huge OpenCL performance boost for all ATI cards - is this correct? So it would be worth a thought to upgrade to 10.7.4 now?

Link to comment
Share on other sites

OpenCL speed improvemnts depends on gpu tyoe. For lowend / midrange (like yours) there is no speed diffs but some bugfixes.

I get same OpenCL speed since 10.6 (now 10.7.5 dev) with Nvidia 9600 GT. AMD is same for lowend cards - only a bit more bugs (OpenCl apps didnt work correct) were fixed (OpenCl on AMD makes more probs that nvidia)

Blender needs an slg (smart lux gpu) plugin. You get it from the luxrender page (with install instruktions).

Link to comment
Share on other sites

You must DL the Luxrender RC1 (OS X, standalone) . Some files which are needed for Blender are within that DL.

Maybe some more informations (older install way) may help to understand how to install that.

http://www.luxrender.net/wiki/LuxBlend_2.5_installation_instructions

If problems you can use the Luxrender Forum:

http://www.luxrender.net/forum/viewtopic.php?f=16&t=1964

Link to comment
Share on other sites

hi gents,

 

Just upgraded to Mountain Lion RTM but no CL. According to netkas it should be working out of the box. I have nvidia Quadro 1000M 1GB.

I got error 2012-07-12 03:10:24 - RUNTIME ERROR: PathOCL kernel compilation error

 

Any ideas ?

 

 

 

Quadro 1000M:

 

Chipset Model: Quadro 1000M

Type: GPU

Bus: PCIe

Slot: Slot-1

PCIe Lane Width: x16

VRAM (Total): 1024 MB

Vendor: NVIDIA (0x10de)

Device ID: 0x0dfa

Revision ID: 0x00a1

ROM Revision: 70.08.50.00.04
Link to comment
Share on other sites

Check other OS X OpeCL benches like OceanWave (less complex code). Maybe OS X 10.7.5 dev / 10.8 GM has problems with luxmark (very complex code) on some cards (bugs in driver, not in luxmark).

Link to comment
Share on other sites

Then you must wait for Drivers updates.

Some kind of gpus, which apple doesn support official may not work with OpenCL even with later updates.

So your problem maybe cannot fixed with the known OpenCL patch.

Does OpenCL info (very lowlevel+easy access to OpenCL driver) works ?

It shows (run in Terminal )some Infos about available OpenCL devices.

At least CPU is shown as OpenCL device (0) - if also gpu shown as OpenCL device (1,....) your problem may get fixed with later driver updates. If OpenCL Info fails (only CPU info) you may not get that fixed ever.

OpenCL Info is within the OpenCL OceanWave (show package content..., ./resources/oclinfo )

Link to comment
Share on other sites

Thx I will verify later. But is OpenCL worked flawlessly on 10.7.5 so did CUDA which isn't working either now.

Upgrading to 10.8 was easy but I have a few issues now. I think to go back to 10.7.5.

Power management (Sleep) doesnt work, Cuda and OpenCL doesnt work.

I have patched Intel Power Management Problems is I think something with P-States because I get error 18 but I will ask in other forum cause

its too offtopic. (Sleep works - but wake up stops screen is corrupted or freezed mouse works)

 

Thx !

Link to comment
Share on other sites

[Device 0]

Name: Quadro 1000M

Vendor: NVIDIA

Type: GPU

Device Version: OpenCL 1.1

Driver Version: CLH 1.0

Compute Units: 2

Work Group Size: 1024

Clock: 1400 MHz

Global Memory: 2048 MB

 

Only OpenCL1.1 ... ?

Link to comment
Share on other sites

I Did put the lib file replacement in GeForceGLDriver.bundle as suggested by netkas. But Still I got the problem. I have 2 issues left to be a perfect LM GM. One is sleep the other is OpenCL. Would be nice to get this beeing fixed :-)

Link to comment
Share on other sites

 Share

×
×
  • Create New...