16 replies to this topic

[robertx]

...well, trying out this card, but in 10.8.2 and in windows8 the card (pci-e x16) appears to be running at x1
 pci-e lane width.png   97.48K   29 downloads
any ideas why? or if it is in fact running at x16, my WEI scores in win8 went way up, and openglviewer on the mac side scores more than doubled...here's a shot from cuda=z performance tab as well
 cuda-z GT430.png   77.23K   23 downloads

[iFIRE]

this is only cosmetic problem:

the card (pci-e x16) appears to be running at x1

performance see this:
AppleGraphicsPowerManagement (AGPM) for nVidia

[robertx]

...hmm, on windows8 now, still shows only pci-e x16 @ x1 ...can it be just cosmetic if both operating systems show the same x1 lane width? With the GT 520 installed, gpu-z shows x16 @ x16 but with GT 430 only x1 in windows8 but any benchmarks i've tried on the windows side show 3 times the scores i receive in osx 10.8.2...what gives?
edit: back on 10.8.2 ...openglextensionviewer now matches my scores from windows 8  very odd
 opengl.png   116.88K   22 downloads

[Gringo Vermelho]

Possible lane sharing with another expansion card? Do you have any other PCI-E cards plugged in?

There should be a diagram in your motherboard manual that shows which slots share available lanes.

It could also be due to some kind of energy saving feature of your motherboard? Try GPU-Z on Windows, I don't remember exactly but I believe it has some button you click that makes the card go from x1 to x16 or something?

[robertx]

hey Gringo...i've tried gpu-z...  cpu-z is the one with the little renderer window...but it doesn't change the lane width for me...Would the GT 430(128bit) really be 3 times faster than the GT 520(64bits) when running only x1 instead of x16?...i'm using a lenovo thinkcentre m57 SFF... it only has one pci-e x16 slot ...and one pci slot(when the pci-e slot is occupied, the onboard graphics are disabled)

[Gringo Vermelho]

Yeah iFIRE is probably correct in saying that it's just cosmetic.

I'm sure it is GPU-Z, CPU-Z does not have anything for video cards..? never mind.

[robertx]

I stand corrected..it IS GPU-Z with the renderer...CPU-Z shows the pci-e lane width on the motherboard tab

[robertx]

sooo... here's my results from OpenCL_OceanWave
 oceanwave.png   387.83K   2 downloads

...this is almost double my score(from results using GT520)  Would this be possible from a GT430 running at pci-e x1 speed?

[Gringo Vermelho]

I don't know how to do the maths. Your result is on par with the existing GT 430 score on mitch_de's OceanWave post.

[mitch_de]

You could also compare Bandwidth speed values : Host(=CPU/Bus)> device(=GPU) and Device  > Host is usefull for PCIe Speed, device>device is VRAM speed
PS: My 9600 GT is shown in Systemprofiler as x16 Lane card. So i don't think the x1 Lane is only cosmetic!
 Bildschirmfoto 2013-02-07 um 21.44.27.jpg   67.68K   7 downloads

Here my old 9600 GT oclBandwidthTest values:

oclBandwidthTest] starting...
/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 2380.2
Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1763.1
Device to Device Bandwidth, 1 Device(s)
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 26650.5
[oclBandwidthTest] test results...
PASSED
> exiting in 3 seconds: 3...2...1...done!

And bandwidthTest_CUDA ( CUDA code based bandwidth tests)

/Users/andreasm/Desktop/bandwidthTest_CUDA ; exit;
[bandwidthTest_CUDA] starting...
/Users/andreasm/Desktop/bandwidthTest_CUDA Starting...
Running on...
Device 0: GeForce 9600 GT
Quick Mode
Host to Device Bandwidth, 1 Device(s), Paged memory
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432   1986.6
Device to Host Bandwidth, 1 Device(s), Paged memory
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432   1479.6
Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432   35298.9
[bandwidthTest_CUDA] test results...
PASSED   > exiting in 3 seconds: 3...2...1...done!

Attached Files

•  oclBandwidthTest.zip   37.46K   3 downloads
•  bandwidthTest_CUDA.zip   35.65K   6 downloads

[robertx]

yeah...i'm stumped, i mentioned in your opencl_oceanwave topic that when i downgrade drivers(back to stock 10.8.2) i get way better results in the bandwidth test, but my device to device score went down...and the drivers needed tweaking some as i was getting screen artifacts and lag... i'll keep the retail nvidia drivers for now(i guess) both drivers still show me at x1

"rolled back my drivers... new results

  /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 10129.9

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432 31610.5

Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432 6642.8

[oclBandwidthTest] test results...
PASSED

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

and with nvidia drivers...
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 162.4

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432 203.1

Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432 11875.8

[oclBandwidthTest] test results...
PASSED

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

[mitch_de]

Yep, seems you have some driver troubles - which may confuse those test also.
Do you mean you have non Apple (Nvidia OSX retail ) drivers installed?
I had in the past troubles with them - since 10.8.1 i only use Apple drivers.
If you have Nvidia retail drivers installed or some patched (OpenCL patch,...) drivers i would try to install fitting Apple OS X combo update to be sure all
non Apple Nvidia drivers (and patches) will get overwritten.

[robertx]

Leslies-Mac-Pro:~ leslie$ /Users/leslie/Downloads/bandwidthTest_CUDA ; exit;
dyld: Library not loaded: @rpath/libcudart.dylib
  Referenced from: /Users/leslie/Downloads/bandwidthTest_CUDA
  Reason: image not found
Trace/BPT trap: 5
logout
...also this from the other Cuda test....

an aside...my cuda performance scores from cuda-z match exactly the scores from opencl bandwidth test....normal?

[mitch_de]

CUDA-Z and OpenCLbandwidth values normally vary light. But if you meaned that if both are very low it means something is wrong.

dyld: Library not loaded: @rpath/libcudart.dylib

  Referenced from: /Users/leslie/Downloads/bandwidthTest_CUDA

This required lib ( for my CUDA bandwidh test) normally got installed by CUDA install. Located :

/usr/local/cuda/lib/libcudart.dylib
I attacht it for you, put it beside (same folder) of

bandwidthTest_CUDA and run again.

Attached Files

•  libcudart.dylib.zip   216.79K   5 downloads

[robertx]

...hey mitch, i put the dylib file in a folder with your test file...
still getting error

Leslies-Mac-Pro:~ leslie$ /Users/leslie/Downloads/untitled\ folder/bandwidthTest_CUDA ; exit;
dyld: Library not loaded: @rpath/libcudart.dylib
  Referenced from: /Users/leslie/Downloads/untitled folder/bandwidthTest_CUDA
  Reason: image not found
Trace/BPT trap: 5
logout

[Process completed]

...also in my /usr/local/cuda/lib/ the file is named libcuda.dylib not "libcudart.dylib" that your test refers to...
...and thanks for your interest and help


....removed card...bad, bad card...back on the GT 520 your test results

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

Running on...

GeForce GT 520

Quick Mode

Host to Device Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            2413.5

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            1875.6

Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            6322.5

[oclBandwidthTest] test results...
PASSED

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

logout

[Process completed]

[robertx]

...well.i decided one more attempt...after measuring my card and enclosure for clearances, i found the card's heatsink/fan block was touching on a neighbouring resistor cap(preventing full card contact)...i removed the heatsink block and adjusted it up as far as i could(of couse reapplied thermal paste)...re-inserted into pci-e x16 slot and am now reporting x16 link width...

 PCI-Ex16 @ x16.png   97.7K   15 downloads

...all seems well, continueing to test...thanks to you few who responded
...and from  Mitch's other test
Last login: Sun Feb 10 21:25:37 on console
Leslies-Mac-Pro:~ leslie$ /Users/leslie/Downloads/oclBandwidthTest ; exit;
[oclBandwidthTest] starting...

/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 2421.5

Device to Host Bandwidth, 1 Device(s), Paged memory, direct access
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432 1831.8

Device to Device Bandwidth, 1 Device(s)
   Transfer Size (Bytes) Bandwidth(MB/s)
   33554432 18300.2

[oclBandwidthTest] test results...
PASSED

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

logout


[Process completed]
device to device now appears nominal(for this card at least)
...this topic could be marked as solved...and left as a reminder,   or deleted.

[mitch_de]

YES that values(hsot>dev , dev>host) >> 1 Gb/sec  i wanted to see with x16lane!
You see: x1 Lane entry (and low 150 MB/sec!)  in Systemprofiler WASN´T only cosmetic - there was a real problem.