[gmx-developers] bs_nix1310 broken

Mark Abraham mark.j.abraham at gmail.com
Mon Nov 20 16:15:08 CET 2017


Hi,

Looks like I needed to purge the installs of old drivers before adding new
ones on bs_nix1310. I didn't need to do that on bs_nix1204, but I've
learned a lesson now :-).

Mark

On Mon, Nov 20, 2017 at 11:36 AM Szilárd Páll <pall.szilard at gmail.com>
wrote:

> Hi,
>
> CC 2.0 devices are indeed unsupported by CUDA 9.0 RT so I could imagine
> that they "forgot" to test the 384-series (9.0 driver version) drivers them
> properly.
>
> I suggest trying a different driver. We have been using 384.90 on a couple
> of dev-machines, although none with CC 2.0 cards. Otherwise, reverting to
> 375 would be the best option.
>
> Cheers,
>
> --
> Szilárd
>
> On Mon, Nov 20, 2017 at 8:13 AM, Mark Abraham <mark.j.abraham at gmail.com>
> wrote:
>
>> Hi guys,
>>
>> Thanks for the ideas. I can use CUDA_VISIBLE_DEVICES to get the test to
>> run on any of the devices, but only with some of the possible combinations
>> of the three devices visible. Several working combinations includes the old
>> GT 480 GPU, but that shouldn't matter because no builds are using CUDA 9.0.
>> And nvidia-smi reports that all devices are in the default non-exclusive
>> mode. mdrun is perfectly happy with the GPUs.
>>
>> Mark
>>
>> On Mon, 20 Nov 2017 08:07 Millad Ghane <mghane at cs.uh.edu> wrote:
>>
>>> Hello Mark,
>>>
>>> It might be due to existence of another program that makes the GPU
>>> busy for itself. If there is a program that uses GPU, the CUDA driver
>>> might halt your program and return that error. It makes sense, right?
>>> No sharing.
>>>
>>>
>>> Bests,
>>> Millad
>>>
>>>
>>> On Sun, Nov 19, 2017 at 11:26 PM, Mark Abraham <mark.j.abraham at gmail.com>
>>> wrote:
>>> > Hi,
>>> >
>>> > I seem to have broken bs_nix1310 while trying to install CUDA 9.0 + a
>>> driver
>>> > than can use it. It worked fine on bs_nix1204, but something is
>>> different
>>> > about this machine. It doesn't seem to be just that it has one really
>>> old
>>> > GPU.
>>> >
>>> > On gerrit, you will see errors from Jenkins builds on that slave
>>> consistent
>>> > with
>>> >
>>> > cudaFuncGetAttributes failed: all CUDA-capable devices are busy or
>>> > unavailable
>>> >
>>> > for some of the PME-related unit tests, until we sort out how to get a
>>> > functional driver working on it. Sorry for the inconvenience!
>>> >
>>> > (More details at https://redmine.gromacs.org/boards/6/topics/851?r=855
>>> for
>>> > those with access to it.)
>>> >
>>> > Mark
>>> >
>>> > --
>>> > Gromacs Developers mailing list
>>> >
>>> > * Please search the archive at
>>> > http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List
>>> before
>>> > posting!
>>> >
>>> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>> >
>>> > * For (un)subscribe requests visit
>>> >
>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
>>> or
>>> > send a mail to gmx-developers-request at gromacs.org.
>>> --
>>> Gromacs Developers mailing list
>>>
>>> * Please search the archive at
>>> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
>>> posting!
>>>
>>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>>
>>> * For (un)subscribe requests visit
>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
>>> or send a mail to gmx-developers-request at gromacs.org.
>>>
>>
>> --
>> Gromacs Developers mailing list
>>
>> * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
>> posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
>> or send a mail to gmx-developers-request at gromacs.org.
>>
>
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20171120/aece877b/attachment.html>


More information about the gromacs.org_gmx-developers mailing list