Re: [Fsuk-manchester] Open ARM GPU drivers

[Dave Love]

Michael Dorrington <address@hidden> writes:

>> I'd be interested if anyone knows about the possibility of using the
>> drivers for floating point work, presumably via OpenCL.  If so, what
>> sort of performance might Mali 400, for instance, provide?  I couldn't
>> find much relevant information even for proprietary drivers.
>
> I don't think there are many people in the world that can answer that
> question about the free software driver.  The best way to find out would
> be to go FOSDEM this weekend and ask the developers. :)

Afraid I'm failing to do enough free software maintenance as it is,
while trying to ensure new systems can use essentially only free
software :-(.

> If you can't do
> that then maybe get a proxy to ask them or perhaps you can ask questions
> via IRC at the end of the talk (as they do at some events).
> https://fosdem.org/2013/schedule/event/operating_systems_open_arm_gpu/
>
> Another possibility is to ask nicely on the powervr-devel list:
> http://listas.gnu.org.ve/cgi-bin/mailman/listinfo/powervr-devel

There are some specific free software points behind this.  One is that
this is one of the few areas where free firmware is really relevant as
far as I know.  (I think objections to downloadable firmware per se are
somewhat overblown, given that there's typically no great need to
program them any more than rms' toaster.)  Another is that GPU support
is the significant missing free software component for current HPC
systems, even if these ones have limited possibilities.

>> -- Dave [research computing muggins, bemused that Southampton claim
>>          a Pi-based "supercomputer"]
>
> Perhaps they just have lots and lots of Pi computers and perhaps it was
> a cheap way of getting ARMs and headline grabbing.

Apparently headline grabbing, and it seems particularly unfortunate for
an educational system.  [http://apt.cs.man.ac.uk/projects/SpiNNaker/ is
"lots and lots" of ARMs.]  64 Pis must compare poorly with a single node
in our ~100-node university clusters, which might be considered
supercomputers.  They're also of limited use for teaching about such
systems -- not that I'm unsupportive of Pis etc. per se.