On Sun, Apr 8, 2018 at 9:40 PM, Dominik 'Rathann' Mierzejewski
<dominik(a)greysector.net> wrote:
On Sunday, 08 April 2018 at 11:05, Peter Robinson wrote:
> > * Brainstorming (sgallagh, 20:07:36)
> > * Feedback: Fedora Server has too short a lifecycle (sgallagh,
> > 20:08:35)
> > * Feedback: Fedora Server should be more minimalistic (sgallagh,
> > 20:08:35)
> > * Feedback: SELinux needs better usability (sgallagh, 20:08:35)
> > * Feedback: Support OpenCL (sgallagh, 20:08:36)
>
> What does that even mean? OpenCL on it's own is basically useless
> without particular frameworks on top of it that enable it to be
> useful. Like most other hardware features just enabling it doesn't
> often make it useful, eg an ethernet NIC on it's own isn't very useful
> with out say a full IP stack on top.
If a "compute node" can be considered a valid use case for Fedora Server,
then there are quite a few scientific applications supporting OpenCL.
It's not as simple as ticking the "OpenCL box" though, Fedora already
has reasonable OpenCL support, the issue in this space is most of the
industry is focused on things like TensorFlow, CAFFE and other such
frameworks that make compute useful and all of those use CUDA. You
need toolchains, frameworks and a lot of other stuff on top of the "we
support OpenCL" box to take it from a tick box to useful.