netkas.org forum
November 16, 2018, 03:41:24 PM *
Welcome, Guest. Please login or register.
Did you miss your activation email?

Login with username, password and session length
News: Information for registering users http://forum.netkas.org/index.php/topic,2246.0.html
 
   Home   Help Search Login Register  
Pages: [1]
  Print  
Author Topic: Nvidia Tesla c1060  (Read 11683 times)
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« on: June 06, 2011, 03:20:28 PM »

Hi,
I would like to get my tesla c1060 working for CUDA/OpenCL on my mac. I have installed all of the possible parts of the CUDA package, yet deviceQuery is reporting no available CUDA devices. My main card is ATI6870, which works fine.

I am using latest Chimera as a bootloader with GraphicsEnabler=Yes, is there any way to enhance the injector to support the Tesla? I believe the only reason why it's not included is its lack of monitor output but it should still be usable for computational usage.

To my surprise, following kexts are actually loaded on startup:
com.apple.NVDAResman
com.apple.GeForce
com.apple.nvidia.nv50hal

Which tells me that the drivers are getting loaded - but what does that actually mean?

Is there any way to give it a "kick" to initialise? My suspicions are that the card is somehow disabled when it does not have a monitor connected to it. I had an issue getting the card running under linux (Ubuntu 11.04), deviceQuery also can not see it until I run sudo nvidia-smi -L, which seems to activate the card in some way. Is there anything similar to this on a Mac side?

Thanks a lot, getting this working would save me a great deal of dualbooting headaches.

Please see the attachment - it shows what I am currently getting in the System Profiler


* Screen shot 2011-06-05 at 18.33.36.png (17.06 KB, 272x276 - viewed 502 times.)
Logged
netkas
Administrator
Hero Member
*****
Offline Offline

Posts: 827



« Reply #1 on: June 06, 2011, 03:28:07 PM »

Obviously injecting didnt work, try to install ATY_Init.kext
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #2 on: June 06, 2011, 03:36:25 PM »

.. and that would help with the Tesla? As I said, the ATI is working fine. What is puzzling is similar behaviour under linux - after cold boot, the card is not working for CUDA until I run the nvidia-smi -L, which despite just listing the devices activates the card somehow. Also the fan speed decreases after the nvidia-smi (same as it does during windows boot)

Is there anything similar I can run on Mac to give it a kick?

How come that the injection did not work and yet I still have all the NVidia drivers loaded?

Sorry for so many questions, I would really appreciate some help. Also, thanks for all the great work for the Hackintosh community!
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #3 on: June 06, 2011, 06:40:10 PM »

Basically, what I believe is happening:
- driver loaded properly
- it does not see any monitors connected, puts the device to some kind of sleep mode (also documented on other systems)
- CUDA cannot see the sleeping card

My issue is, as I can not connect any monitors to the Tesla (it does not have any outputs), how do I fool the driver into believing that the device is indeed supposed to be fully operational and working, despite no monitors being connected?

As I have mentioned, on linux I simply need to call nvidia-smi -L and it wakes the card up (you can tell by fan speed change if booted from cold or by CUDA suddenly starting to work). Any thoughts how to do that on the Mac side?
Logged
netkas
Administrator
Hero Member
*****
Offline Offline

Posts: 827



« Reply #4 on: June 06, 2011, 09:52:40 PM »

Basically, what I believe is happening:
- driver loaded properly
wrong, screenshot proofs it
Quote
- it does not see any monitors connected, puts the device to some kind of sleep mode (also documented on other systems)


wrong, due to 1st
Quote
- CUDA cannot see the sleeping card

wrong

Quote
My issue is, as I can not connect any monitors to the Tesla (it does not have any outputs), how do I fool the driver into believing that the device is indeed supposed to be fully operational and working, despite no monitors being connected?

As I have mentioned, on linux I simply need to call nvidia-smi -L and it wakes the card up (you can tell by fan speed change if booted from cold or by CUDA suddenly starting to work). Any thoughts how to do that on the Mac side?
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #5 on: June 06, 2011, 09:54:35 PM »

ok, so back to my initial question - would the ATY_init help with the tesla initialisation?

Also, why do all the NVidia kexts load?

I have also been looking at the chameleon's injector source for my device ID (at http://forge.voodooprojects.org/p/chameleon/source/tree/HEAD/trunk/i386/libsaio/nvidia.c), but it does not contain the string for the Tesla. Is that even relevant? It does not seem to use the information from NVKnownChipsets[] anyway...

Thanks for your help, you are doing great service to bioinformatics Smiley
« Last Edit: June 06, 2011, 09:58:58 PM by zard_cz » Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #6 on: June 06, 2011, 10:01:21 PM »

I have not used the ATY_init before, is my assumption right that I disable GraphicsEnabler and drop the kext into the /Extensiond directory to make it active?
Logged
netkas
Administrator
Hero Member
*****
Offline Offline

Posts: 827



« Reply #7 on: June 07, 2011, 02:40:43 AM »

dont disable graphicsenabler
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #8 on: June 07, 2011, 11:10:19 AM »

will try, thanks
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #9 on: June 07, 2011, 11:22:47 AM »

I've tried ATY_Init from http://netkas.org/?p=798 and nothing had changed.. I have dropped it into /Extra/Extensions and ran Kext utility
Logged
netkas
Administrator
Hero Member
*****
Offline Offline

Posts: 827



« Reply #10 on: June 07, 2011, 06:36:13 PM »

drop it to /S/L/E, not /E/E
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #11 on: June 08, 2011, 02:54:12 PM »

Done.. and again, same result. It seems to see only the ATI card, I got this in the console:
08/06/2011 14:46:22   kernel   ATY_Init: Starting
08/06/2011 14:46:22   kernel   ATY_Init: Checking dev-id 6738
08/06/2011 14:46:22   kernel   ATY_Init: Card 0x6738 was posted
08/06/2011 14:46:22   kernel   ATY_Init: Quiting for card 0x6738

And that's all. Systems seems to work fine otherwise, like nothing had happened.
Logged
netkas
Administrator
Hero Member
*****
Offline Offline

Posts: 827



« Reply #12 on: June 08, 2011, 07:42:42 PM »

so, it looks like card has different pci class, so you need to edit plist of aty_init
Logged
zard_cz
Not Newbie
*
Offline Offline

Posts: 9


« Reply #13 on: June 08, 2011, 09:12:03 PM »

ok, any hints on that? I can see there is an entry for NVIDIA in the Info.plist but honestly don't know what to insert there...
I am not asking for exact directions, but a pointer in the right direction what to read and try would be greatly appreciated..
« Last Edit: June 09, 2011, 01:40:41 AM by zard_cz » Logged
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.21 | SMF © 2015, Simple Machines Valid XHTML 1.0! Valid CSS!