Jump to content

OpenCL Oceanwave & Bandwidth Bench - 07. March 2013

OpenCL AMD NVIDIA

  • Please log in to reply
343 replies to this topic

#221
gils83

gils83

    DJ Officiel d'InsanelyMac

  • Members
  • PipPipPipPipPipPipPipPipPipPip
  • 1,894 posts
  • Gender:Male
  • Location:France
  • Interests:le soleil du var et l'informatique
Acer 8930G core2duo / GT 9600m 512


Caractéristiques de l'objet Point d'origine: Guangdong China (Mainland)

Description

Carte vidéo GeForce 9600M GS 512MB, DDR3 graphiques II originaux des tous neufs de Nvidia MXM

Dispositifs :

Fabricant NVIDIA Série GeForce 9M Nom de code NB9P Canalisations 32 - unifié Vitesse de noyau * 430 mégahertz Vitesse de Shader * 1075 mégahertz Vitesse de mémoire * 800 mégahertz Largeur d'autobus de mémoire Bit 128 Type de mémoire GDDR2 Quantité maximale de mémoire Mb 256 Mémoire partagée non
DirectX DirectX 10, Shader 4.0 Transistors 314 millions technologie 65 nanomètre Dispositifs PCI-E 2.0, 400 mégahertz de Speichertakt d'und du bei GDDR2 bei GDDR3 de 800 mégahertz Taille de cahier moyen Date d'annonce 03.06.2008






Posted Image

#222
mitch_de

mitch_de

    InsanelyMacaholic

  • Retired
  • 2,896 posts
  • Gender:Male
  • Location:Stuttgart / Germany


i'm on OS X 10.7.4 , could it be that ?


Normally not, even i build it with 10.8 SDK.
Try OpenCL_Oceanwave ( the OpenCL code part, put unzipped into OpenCL_Oceanwave/Content/Resources) which was build using SDK 10.7

For me, 10.8.3 doenst matter in speed / working same using the 10.7 SDK build.

Attached Files



#223
gils83

gils83

    DJ Officiel d'InsanelyMac

  • Members
  • PipPipPipPipPipPipPipPipPipPip
  • 1,894 posts
  • Gender:Male
  • Location:France
  • Interests:le soleil du var et l'informatique
AMD Phenom x6 / HD 4850 1024


Posted Image

#224
TH3L4UGH1NGM4N

TH3L4UGH1NGM4N

    (~_~)

  • Retired
  • 1,158 posts
  • Gender:Male
  • Location:Wonderland
  • Interests:(~_^)
Surprised to see my 6870 outscore a 7870 on page 11 of this thread o.o

Posted Image

#225
mitch_de

mitch_de

    InsanelyMacaholic

  • Retired
  • 2,896 posts
  • Gender:Male
  • Location:Stuttgart / Germany
Yep, but small FPS diffs can happen between different CPu /Bus / PCIe speeds and different clocked (GPU/VRAM) speeds to.
Even on same system speed diffs between two runs can happen - in area of 400 FPS at least 10-20 fps diff between runs is normal.

Perhaps the OpenCLbandwidth test may help to find out such speed diffs (CPU<>GPU data transferspeed, GPU<>GPU(=devicetodevice) speeds)

Here are my results for 9600 GT:
Last value, device to device shows VRAM acess speed, limited by VRAM type + VRAM clock and VRAM bandwidth (how many Bits, like 64 bit for low end gpus like GT210/GT220, 128 bits, 192 bits or 256 Bits)
The first two values are CPU<>PCI<>GPU transferspeeds and limited by PCIe speed, CPU type and GPU type+clock



Developer/GPU Computing/OpenCL/bin/darwin/release/oclBandwidthTest Starting...
Running on...
GeForce 9600 GT
Quick Mode
Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 2200.0

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1784.8

Device to Device Bandwidth, 1 Device(s)
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 27501.7

[oclBandwidthTest] test results...
PASSED > exiting in 3 seconds: 3...2...1...done! logout

Attached Files



#226
TH3L4UGH1NGM4N

TH3L4UGH1NGM4N

    (~_~)

  • Retired
  • 1,158 posts
  • Gender:Male
  • Location:Wonderland
  • Interests:(~_^)
Very much true that there will always be slight differences when having same gpu with different cpu/bus/pci speeds etc but looking at his system with an i5 paired with the 7870, I'd only assume that my cpu overclock has to do with the higher fps from the gpu (just an assumption I'll test with stock clock settings to verify).

And thank you mitch_de the test proves quite interesting to see those differences in bandwidth. The last one involving the vram speed was the most interesting for me to see as I've never seen the actual speeds in any tests before.

I might pick at the other guy to run the test as well so I could see how the numbers look on his end.

Running on...
ATI Radeon Barts XT Prototype
Quick Mode
Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 4531.3
Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5770.6
Device to Device Bandwidth, 1 Device(s)
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 102000.4
[oclBandwidthTest] test results...
PASSED
> exiting in 3 seconds: 3...2...1...done!


#227
k3nny

k3nny

    InsanelyMac Legend

  • Members
  • PipPipPipPipPipPipPip
  • 553 posts
  • Gender:Male
Mh, here are my results:

/Users/alex/Downloads/oclBandwidthTest Starting...
Running on...
AMD Radeon HD Pitcairn XT Prototype Compute Engine
Quick Mode
Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 9859.1
Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 11576.8
Device to Device Bandwidth, 1 Device(s)
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 84047.9
[oclBandwidthTest] test results...
PASSED
> exiting in 3 seconds: 3...2...1...done!


#228
TH3L4UGH1NGM4N

TH3L4UGH1NGM4N

    (~_~)

  • Retired
  • 1,158 posts
  • Gender:Male
  • Location:Wonderland
  • Interests:(~_^)
Talk about quick to post, thank you k3nny :)

As I somehow assumed, it's definitely only the gpu not so much the OC because your first 2 test scores no doubt beat mine since the cpu is in the party.

Both our cards share the same 256 bit interface so the difference lies within the cards independent of the other system components.

#229
mitch_de

mitch_de

    InsanelyMacaholic

  • Retired
  • 2,896 posts
  • Gender:Male
  • Location:Stuttgart / Germany
Yep, interesting is the big speed diff for transferspeed CPU>PCIe bus> GPU (VRAM) and GPU (VRAM) > PCIe bus> CPU.
Twice as fast for the one system compared to other. 4500 vs 9800 MB/sand 5700 vs 11500 MB/s !
GPU VRAM >GPU VRAM transferspeed diff isnt that much 84000 vs 102000.
Because OpenCL Apps must always (more or less) submit data (for computing) to gpu and receive result data that transferspeed (CPU<>PCie<>GPU) can
make the difference even gpu+VRAM speed has less differences.
Thats why real (science) gpu computing tries to minimise transfered data or like PCie 3.0 (+ MB Chipset support + GPU chipset support) bus systems.
(Old AGP or PCI bus would be a no go for today gpu computing : only 1/4 - 1/10 of your transferspeeds over PCie)

#230
oSxFr33k

oSxFr33k

    InsanelyMac Legend

  • Members
  • PipPipPipPipPipPipPip
  • 845 posts
  • Gender:Male
  • Interests:Sound and Graphic Design. Electronics in general.
DOes anyone know what the Hex edits are for OSX 10.8.2 in:

/System/Library/Extensions/

GeForceGLDriver.bundle/Contents/MacOS/GeForceGLDriver



and



/System/Library/Extensions/

GeForceGLDriver.bundle/Contents/MacOS/libclh.dylib



#231
Gringo Vermelho

Gringo Vermelho

    The Jan Bird fix

  • Supervisors
  • 6,121 posts
  • Gender:Male
  • Location:Brazil
*First Kepler score*

892.2 FPS

Brand new Mountain Lion 10.8.2 install
Nvidia 304.00.05f02 driver, CUDA 5.0.37
OpenCL unpatched
No AGPM edits
GraphicsEnabler=n - no injection!
MacPro3,1 system definition
EVGA Geforce GTX 660 2GB (vanilla model)
Core 2 Duo E8500 @ 3.16GHz
Asus P5Q-E (P45 Express/ICH10R) 4GB RAM

Attached File  Screen Shot 2013-02-04 at 1.07.30 AM.png   388.45KB   6 downloads

Running on...

GeForce GTX 660

Quick Mode

Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 4227.2

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5542.3

Device to Device Bandwidth, 1 Device(s)
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 81757.6

[oclBandwidthTest] test results...
PASSED

  • p.H likes this

#232
TH3L4UGH1NGM4N

TH3L4UGH1NGM4N

    (~_~)

  • Retired
  • 1,158 posts
  • Gender:Male
  • Location:Wonderland
  • Interests:(~_^)
I can't believe dis!

Posted Image

You now own the stripper pole GV

#233
eep357

eep357

    Triple Platinum

  • Retired
  • 2,527 posts
  • Gender:Male
  • Location:Dark Side of The Wall
  • Interests:things and stuff
I went to New Jersey and all I got was this stupid shirt......Oh, and a GTX 660! :)

#234
gothic860

gothic860

    InsanelyMac Protégé

  • Members
  • PipPip
  • 84 posts
  • Gender:Male
  • Location:Germany, Bavaria

*First Kepler score*

892.2 FPS

Brand new Mountain Lion 10.8.2 install
Nvidia 304.00.05f02 driver, CUDA 5.0.37
OpenCL unpatched
No AGPM edits
GraphicsEnabler=n - no injection!
MacPro3,1 system definition
EVGA Geforce GTX 660 2GB
Core 2 Duo E8500 @ 3.16GHz
Asus P5Q-E (P45 Express/ICH10R) 4GB RAM


And why i get only 485fps with my GTX680 :wallbash:

#235
TH3L4UGH1NGM4N

TH3L4UGH1NGM4N

    (~_~)

  • Retired
  • 1,158 posts
  • Gender:Male
  • Location:Wonderland
  • Interests:(~_^)
That's the part that's blowing my brain cells since I saw your 680 fps and it was half that of a 660? (~_^)

#236
mitch_de

mitch_de

    InsanelyMacaholic

  • Retired
  • 2,896 posts
  • Gender:Male
  • Location:Stuttgart / Germany

And why i get only 485fps with my GTX680 :wallbash:


Hmmm, does show OpenCL bandwith test some bottleneck compared to similar cpu/gpu systems?
But i heared that, also Luxmark (OpenCL) speed of some GTX 6xx ( i dont remember which) was much worse than older GTX 5xx card.
Maybe it was the GTX 680 which si fast OpenGL card but way slower than possible in CUDA/OpenCL because some internal design changes
to get more FPS out in OpenGL in cost of much less OpenCL / Shader speed ?

attached luxmark DB results (fastest, medium scene), GTX 680 has less compute units than other highend gpus.

Attached File  Bildschirmfoto 2013-02-04 um 11.54.57.jpg   145.2KB   23 downloads

#237
Regi Yassin

Regi Yassin

    Who am I ?

  • Members
  • PipPipPipPipPip
  • 278 posts
  • Gender:Not Telling
here is my results
hw details, in sig

nvidia official driver for 10.8.2 + cuda 5.0.37 + agpm edit
iMac 12,2

Attached Files



#238
mitch_de

mitch_de

    InsanelyMacaholic

  • Retired
  • 2,896 posts
  • Gender:Male
  • Location:Stuttgart / Germany
Great to see GTX 650 Ti works well & fast.

#239
Gringo Vermelho

Gringo Vermelho

    The Jan Bird fix

  • Supervisors
  • 6,121 posts
  • Gender:Male
  • Location:Brazil
*lol* guys!!

#240
RobertX

RobertX

    Yosemite Sam

  • Members
  • PipPipPipPipPipPipPip
  • 570 posts
  • Gender:Not Telling
...and now for something completely different...

/Users/leslie/Downloads/oclBandwidthTest Starting...

Running on...

GeForce GT 430

Quick Mode

Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 161.5

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 202.9

Device to Device Bandwidth, 1 Device(s)
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 12265.7

[oclBandwidthTest] test results...
PASSED

> exiting in 3 seconds: 3...2...1...done!

logout





0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users

© 2014 InsanelyMac  |   News  |   Forum  |   Downloads  |   OSx86 Wiki  |   Mac Netbook  |   PHP hosting by CatN  |   Designed by Ed Gain  |   Logo by irfan  |   Privacy Policy