Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

CUDA Errors - All CUDA-capable devices are busy

576 views
Skip to first unread message

Richard Biehl

unread,
Mar 8, 2012, 9:27:34 PM3/8/12
to
I am having a problem getting the Matlab Parallel Computing toolbox to work with my Tesla M2070. When I submit the gpuDevice I get the following error:

Error using gpuDevice (line 22)
An unexpected error occurred during CUDA execution. The CUDA error was: all
CUDA-capable devices are busy or unavailable.

I get the following information from the nVidia SMI utility and I am using the card in TCC mode. I have it installed on a Windows Server 2008 R2 server with MatLab R2011b. The latest nVidia drivers are installed. The display adapter is a Matrox card and the Tesla is for GPU computing only. Any thoughts??

Richard

==============NVSMI LOG==============

Timestamp : Thu Mar 08 20:16:54 2012

Driver Version : 295.73

Attached GPUs : 1

GPU 0000:08:00.0
Product Name : Tesla M2070
Display Mode : Disabled
Persistence Mode : N/A
Driver Model
Current : TCC
Pending : TCC
Serial Number : 0323710045651
GPU UUID : GPU-896b9442dd076af3-ac3a92e7-8157433c-3d0de69
9-af9ee5af7da574b1866fa0d5
VBIOS Version : 70.00.3E.00.03
Inforom Version
OEM Object : 1.0
ECC Object : 1.0
Power Management Object : 1.0
PCI
Bus : 0x08
Device : 0x00
Domain : 0x0000
Device Id : 0x06D210DE
Bus Id : 0000:08:00.0
Sub System Id : 0x083010DE
GPU Link Info
PCIe Generation
Max : 2
Current : 2
Link Width
Max : 16x
Current : 16x
Fan Speed : N/A
Performance State : P8
Memory Usage

Ben Tordoff

unread,
Mar 12, 2012, 5:00:13 AM3/12/12
to
Hi Richard,

This message appears either when no device is found or when the default device is locked out of use. In order to diagnose the problem, can you try the following and post the results:

>> gpuDeviceCount()
>> gpu = parallel.gpu.GPUDevice.getDevice(1)

Note that this latter call asks for information about the device without trying to select it, which should help in working out whether finding the device or selecting the device is the problem. Hopefully you will see the details for your card with "DeviceSupported" set to 1 and "ComputeMode" set to "Default".

Cheers

Ben


"Richard Biehl" wrote in message <jjbpqm$899$1...@newscl01ah.mathworks.com>...
> I am having a problem getting the Matlab Parallel Computing toolbox to work with my Tesla M2070. When I submit the gpuDevice I get the following error:
>
> Error using gpuDevice (line 22)
> An unexpected error occurred during CUDA execution. The CUDA error was: all
> CUDA-capable devices are busy or unavailable.

[snip]
0 new messages