Re: [Fsuk-manchester] Open ARM GPU drivers

[Andrew Back]

On 4 February 2013 10:20, Bob Ham <address@hidden> wrote:
> On Mon, 4 Feb 2013 07:43:03 +0000, Andrew Back <address@hidden>
> wrote:
>
>> You may also like to take a look at the Parallella board:
>>
>>   http://www.parallella.org/board/
>>
>> The 16-core Epiphany floating-point hardware provides 25 GFLOPS
>
> Interesting.
>
> Out of curiosity, are there any performance comparisons with GPGPUs?  I
> couldn't find any on the CoreMark website.

There are some comparisons here for the 64-core device:

  
http://www.adapteva.com/white-papers/ten-myths-debunked-by-the-epiphany-iv-64-core-accelerator-chip/

Since the Kickstarter stretch target wasn't met there will only be a
small number of 64-core Parallella prototypes made this time round.

>> There is also an LGPL v3 licensed OpenCL implementation:
>>
>>   http://www.browndeertechnology.com/coprthr.htm
>
> Also interesting, and could be useful.  I find it noteworthy that the
> library targets multi-core CPUs, rather than GPGPUs.  I've viewed OpenMP as
> the de facto method for parallelism on CPUs and OpenCL for GPGPUs but on
> reflection, OpenMP seems only like a quick and easy way to get instant
> parallelism.  It makes me wonder what advantages there are to implementing
> algorithms in OpenCL kernels rather than in parallel OpenMP loops, for a
> CPU.  They're quite different environments and OpenCL has a lot of overhead
> just to get a kernel to run.

Note that the OpenCL implementation includes a simplified API that is
similar to CUDA to use:

  http://www.browndeertechnology.com/docs/stdcl-reference-rev1.3.html

You can see a matrix multiplication example for Epiphany here:

  
http://www.adapteva.com/support/examples/opencl-examples/efficient-matrix-multiplication-using-opencl/

Cheers,

Andrew

-- 
Andrew Back
http://carrierdetect.com