Re: AW: AW: AW: FATAL ERROR: PME offload requires exactly one CUDA device per process.

From: Maxime Boissonneault (maxime.boissonneault_at_calculquebec.ca)
Date: Fri Aug 22 2014 - 08:59:26 CDT

I did. As far as I understand it, the PPN number is the number of thread
per *process*.

In the multicore version, there is *one* process.

The only way I managed to use 8 GPUs is by using the MPI version. In
that case, there are as many processes as the number of MPI ranks, so I
can start it with
mpiexec -np 8 namd2 +ppn2 ...

That will start 8 processes (one for each GPU), and 2 threads per process.

Maxime

Le 2014-08-22 09:56, Norman Geist a écrit :
> So did you try using "EXACLTY" 8 cores?
>
> Norman Geist.
>
>> -----Ursprüngliche Nachricht-----
>> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
>> Auftrag von Maxime Boissonneault
>> Gesendet: Freitag, 22. August 2014 15:23
>> An: Norman Geist
>> Cc: Namd Mailing List
>> Betreff: Re: AW: AW: namd-l: FATAL ERROR: PME offload requires exactly
>> one CUDA device per process.
>>
>> There are 8 K20 GPUs per node.
>>
>> I was hoping to be able to use the multicore binaries instead of the
>> MPI
>> one.
>>
>> Maxime
>>
>> Le 2014-08-22 09:21, Norman Geist a écrit :
>>> How many GPUs do you have?
>>>
>>> Norman Geist.
>>>
>>>
>>>> -----Ursprüngliche Nachricht-----
>>>> Von: Maxime Boissonneault
>> [mailto:maxime.boissonneault_at_calculquebec.ca]
>>>> Gesendet: Freitag, 22. August 2014 14:45
>>>> An: Norman Geist
>>>> Betreff: Re: AW: namd-l: FATAL ERROR: PME offload requires exactly
>> one
>>>> CUDA device per process.
>>>>
>>>> Threads do not work.
>>>>
>>>> So that means that it is impossible to do PME offload on multiple
>> GPUs
>>>> without MPI.
>>>>
>>>> Thanks,
>>>>
>>>> Maxime
>>>>
>>>> Le 2014-08-22 08:44, Norman Geist a écrit :
>>>>> I guess the message tells everything. You need to start more
>>>> processes.
>>>>> Maybe also threads will work so +ppn
>>>>>
>>>>> Norman Geist.
>>>>>
>>>>>> -----Ursprüngliche Nachricht-----
>>>>>> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
>>>>>> Auftrag von Maxime Boissonneault
>>>>>> Gesendet: Freitag, 22. August 2014 13:40
>>>>>> An: Namd Mailing List
>>>>>> Betreff: namd-l: FATAL ERROR: PME offload requires exactly one
>> CUDA
>>>>>> device per process.
>>>>>>
>>>>>> Hi,
>>>>>> I'm getting this error for some computations using the Multicore
>>>>>> binary.
>>>>>> FATAL ERROR: PME offload requires exactly one CUDA device per
>>>> process.
>>>>>> I was wondering, is it a matter of configuration of the problem,
>> or
>>>>>> just
>>>>>> of the problem size ? I understand I can use the MPI binary to
>>>> address
>>>>>> that and use multiple GPUs, but I'm wondering if I can use the
>>>>>> multicore
>>>>>> one (because it performs slightly better than the MPI one).
>>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>>
>>>>>> --
>>>>>> ---------------------------------
>>>>>> Maxime Boissonneault
>>>>>> Analyste de calcul - Calcul Québec, Université Laval
>>>>>> Ph. D. en physique
>>>>> ---
>>>>> Diese E-Mail ist frei von Viren und Malware, denn der avast!
>>>> Antivirus Schutz ist aktiv.
>>>>> http://www.avast.com
>>>>>
>>>> --
>>>> ---------------------------------
>>>> Maxime Boissonneault
>>>> Analyste de calcul - Calcul Québec, Université Laval
>>>> Ph. D. en physique
>>> ---
>>> Diese E-Mail ist frei von Viren und Malware, denn der avast!
>> Antivirus Schutz ist aktiv.
>>> http://www.avast.com
>>>
>>>
>>
>> --
>> ---------------------------------
>> Maxime Boissonneault
>> Analyste de calcul - Calcul Québec, Université Laval
>> Ph. D. en physique
>
> ---
> Diese E-Mail ist frei von Viren und Malware, denn der avast! Antivirus Schutz ist aktiv.
> http://www.avast.com
>
>

-- 
---------------------------------
Maxime Boissonneault
Analyste de calcul - Calcul Québec, Université Laval
Ph. D. en physique

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:21:09 CST