From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Mon Jan 16 2012 - 19:29:26 CST
hi,
its been a long time coming, but i finally found a need
to start updating out GPU nodes to CUDA 4.x (4.1rc2 right now)
and also updated the gpu far control hack for headless
compute nodes running nvidia GPUs (i have a version
for AMD ones, too, btw), since we have started testing
and tuning the OpenCL accelerated multi-GPU support
in LAMMPS on AMD hardware, too.
as i suspected, the persistence mode setting is required
to achieve the same effect as with the logger feature before.
please find the updated script attached.
cheers,
axel.
On Fri, Jul 22, 2011 at 10:19 AM, Axel Kohlmeyer <akohlmey_at_gmail.com> wrote:
> On Fri, Jul 22, 2011 at 10:12 AM, Ajasja Ljubetič
> <ajasja.ljubetic_at_gmail.com> wrote:
>>
>>> as of recent, nvidia seems to have made
>>> it a policy to make it harder to use consumer grade hardware
>>> for GPU computing by dropping access to helpful features that
>>> used to be (semi-)supported.
>>>
>>
>> Yes, sadly I noticed. Quad buffering is not supported on the GTX line
>> either, but requires the Quadro line.
>
> that is something else. this distinction has always been the case.
>
> the change i was commenting on has been an arbitrary and recent
> one and particularly comes after a long time of advertising that
> _all_ nvidia gpus offered are capable for GPU computing.
>
>> Still, setting fan speed in windows does not seem to be a problem, so there
>> is still hope.
>> Could you comment on what "nvidia-smi -d -l -i 900" was supposed to do?
>
> it is supposed to emulate what is now called "persistence mode".
> normally GPU settings (like fan speed control) persist only until
> the driver is re-initialized. with a headless GPU server, this would
> happen _every_ time a new GPU application starts. in order to
> keep the GPU driver from re-initializing the GPUs, one has to keep
> a context open. in a desktop environment that would be the X server.
> for a headless workstation one needs something else. this command
> serves the purpose and does (almost) nothing.
>
> cheers,
> axel.
>
>> Thank you & best regards,
>> Ajasja
>
>
>
> --
> Dr. Axel Kohlmeyer
> akohlmey_at_gmail.com http://goo.gl/1wk0
>
> Institute for Computational Molecular Science
> Temple University, Philadelphia PA, USA.
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 College of Science and Technology Temple University, Philadelphia PA, USA.
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:07 CST