Hi,

the new gpu-backend print the PCI number that is also shown by nvidia-smi.
So this allow to match them.

https://github.com/Theano/Theano/issues/5176

Also someone wrote about using the env environment CUDA_​DEVICE_​ORDER=
PCI_BUS_ID

to change cuda device order to be the same as nvidia-smi.

Example of the new output.

$ nvidia-smi
Tue Nov 29 11:38:29 2016
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 367.44                 Driver Version:
367.44                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr.
ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute
M. |
|===============================+======================+======================|
|   0  GeForce GTX 750     On   | 0000:05:00.0      On |
N/A |
| N/A   66C    P0    11W /  38W |    702MiB /  1999MiB |     64%
Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 1080    On   | 0000:06:00.0     Off |
N/A |
| 30%   46C    P8     7W / 180W |      1MiB /  8113MiB |      0%   E.
Process |
+-------------------------------+----------------------+----------------------+

$ THEANO_FLAGS=device=cuda python -c "import theano"
/Tmp/lisa/os_v5/anaconda/lib/python2.7/site-packages/skcuda/cublas.py:273:
UserWarning: creating CUBLAS context to get version number
  warnings.warn('creating CUBLAS context to get version number')
Mapped name None to device cuda: GeForce GTX 750
PCI Bus ID: 0000:05:00.0
Using cuDNN version 5103 on context None



On Tue, Nov 29, 2016 at 11:30 AM, Ludwig Schmidt-Hackenberg <
[email protected]> wrote:

> Hi,
>
> we are facing the same issue, that people need to remember which GPU
> device ID is which GPU in nvidia-smi to not accidentally kill an already
> running training.
> Did you find any better solutions in the mean time or is Russel's script
> still the preferred way?
>
> Thanks,
> Ludwig
>
>
> On Sunday, January 25, 2015 at 12:52:02 PM UTC+1, Sander Dieleman wrote:
>>
>> I've noticed that on multi-GPU machines with different types of GPUs, the
>> ordering in 'nvidia-smi' and Theano's ordering (gpu0, gpu1, ...) can be
>> different (and both can differ from the actual physical order of the PCI
>> slots).
>>
>> This is only a minor annoyance, but is there a way to make Theano use the
>> same order as nvidia-smi, or vice versa? This would make it much easier to
>> see which GPUs are free to use.
>>
>> Thanks!
>>
>> Sander
>>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups
> "theano-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to