Jump to content

OpenCL Nbody Simulation bench (works 10.9+)


mitch_de
 Share

81 posts in this topic

Recommended Posts

893 Gigaflops (81920 stars) :superman:

 

You can try to set even more stars (run much slower but more Gigaflops on very fast cards possible).

You used 81920 , you can try 262144 star count also. Not recommended for slow / low end gpus - may stall OpenGL output / run out of VRAM.

Link to comment
Share on other sites

For me OpenCL is not working and my card is not being recognized either as an OpenCL device...

 

I have a Quadro 5000 and it's working pretty well OOB, tried to find the hex string to patch it under ML, but the string has changed, so I don't know where to patch the file.

Link to comment
Share on other sites

For me OpenCL is not working and my card is not being recognized either as an OpenCL device...

 

I have a Quadro 5000 and it's working pretty well OOB, tried to find the hex string to patch it under ML, but the string has changed, so I don't know where to patch the file.

 

I didnt have to patch anything under ML for OpenCL - it just worked (GTX570)

Link to comment
Share on other sites

  • 2 weeks later...

Managed to get 10 Gflops more in ML (863) vs Lion (853), GPU settings/framebuffer same on both, ML on SSD while Lion install fills more than half of 2x1TB HDs in RAID-0, and a lot more stuff running in background/startup on Lion might explain the difference.

 

Mountain Lion Galaxies.pngLion Galaxies.pngScreen Shot 2012-03-15 at 6.49.32 PM.png

Link to comment
Share on other sites

  • 2 weeks later...
  • 3 weeks later...

Nun, Galaxies startet gleich im GPU Mode, also SIM: Grafikkartennamen.

Mit key s kann man in den CPU mode umschalten.

Wenn Galaxies nicht im GPU Mode startet sondern im CPU MODE, dann klappt was mit OpenCL bei dir nicht.

Funzen denn die anderen OpenCL Benches (Luxmark2 oder OpenCLWeave)?

 

galaxies starts up in gpu mode by default. If galaxies starts up in an cpu mode (single/multi thread) then something is wrong with the OpenCL device gpu.

Try other OpenCL Benches.

Link to comment
Share on other sites

  • 3 weeks later...

yeah, i tried that, switching between single & multicpus is working...next switch galaxies quits unexpected!!!!

logfile is in your pm box...thanx mitch!!!

 

...same problem here after updating to 10.7.3 (Zotac GT 520) doesn't load the gpu sim only gt 520 + cpu (single -multi)

Link to comment
Share on other sites

Try an new BUILD compiled under 10.7 / BUILD for 10.7.

For me both BUILDS working 10.7.3 + 10.7.4 : this test build today and the other (10.7+10.8) from first page on Nvidia 9600 GT - and getting same GigaFlops (138).

Dont use command line 80K Stars! Simple start galxies without changing star cont , so it can use 65K (default)!

May even also not work on some GPUs. Dont worry, until other OpenCL Benches like Luxmark 2.0 or OpenCL_Oceanwave work on your GPU.

Galaxies_BUILD_10.7.zip

Link to comment
Share on other sites

Try an new BUILD compiled under 10.7 / BUILD for 10.7.

For me both BUILDS working 10.7.3 + 10.7.4 : this test build today and the other (10.7+10.8) from first page on Nvidia 9600 GT - and getting same GigaFlops (138).

Dont use command line 80K Stars! Simple start galxies without changing star cont , so it can use 65K (default)!

May even also not work on some GPUs. Dont worry, until other OpenCL Benches like Luxmark 2.0 or OpenCL_Oceanwave work on your GPU.

...that build also a no-go here...on 10.7.3...still runs fine on 10.7.2

Link to comment
Share on other sites

damn...found my problem...a bad edit in geforcegldriverbundle... fixed after reading this:

in 10.7.3 there is a change in the GeForceGLDriver... see below

From Netkas Thread:

 

 

Find

EB A8 83 F8 02 7C 15

replace 02 with 03 to get

EB A8 83 F8 03 7C 15

Find

78 E8 83 F8 02 7C 11

replace 02 with 03 to get

78 E8 83 F8 03 7C 11

i didn't do the edit in bold...(in HexEdit) :| ...seems all good now...my bad...( i only get 60-65 gigaflops with my GeForce GT 520)

Link to comment
Share on other sites

Good to hear you get your fermi lowend GT 520 cardto work. 9600 GT gets 131 (10.7.3 and 10.7.4)

What result with Luxmark 2.0 in the small Luxball Scene ? i get around 740 (and 74 in the middle complex scene, 25 very complex scene)

http://www.luxrender...uxMark#Download

 

Oceanwave (attached) gets around 90 FPS. All results for Nvidia 9600 GT.

 

Fermi gpu type doesnt mean its always faster than the old non fermi types. For example the GTX 680 is great in OpenGL but very bad in OpenCL speed compared to GTX 580. GTX 680 plays in OpenCL region of GTX 460-GTX 470. Upcoming GTX 690 is much faster in OpenCL (ok, it has 2 OpenCL gpus on the card ;) )

OpenCL_OceanWave_32Bit.zip

Link to comment
Share on other sites

damn...found my problem...a bad edit in geforcegldriverbundle... fixed after reading this:

in 10.7.3 there is a change in the GeForceGLDriver... see below

From Netkas Thread:

 

 

Find

EB A8 83 F8 02 7C 15

replace 02 with 03 to get

EB A8 83 F8 03 7C 15

Find

78 E8 83 F8 02 7C 11

replace 02 with 03 to get

78 E8 83 F8 03 7C 11

i didn't do the edit in bold...(in HexEdit) :| ...seems all good now...my bad...( i only get 60-65 gigaflops with my GeForce GT 520)

 

 

thanx for da headsup....wrong edit too!!!fixed

 

432 gflops...thanx guys

Link to comment
Share on other sites

  • 4 months later...
  • 1 month later...
 Share

×
×
  • Create New...