[petsc-dev] http://www.hpcwire.com/hpcwire/2012-11-12/intel_brings_manycore_x86_to_market_with_knights_corner.html

[John Fettig]

The K20 sounds like it will be priced at around $3200, which makes this
card cheaper than the NVIDIA offering.  Plus, the floating point throughput
really isn't the sticking point for PETSc-based applications that are
memory bound.  Also consider that 3 GB of memory to split up the work
hundreds of ways is a little on the small side.  8GB is a little more
reasonable.

Then add on top of this the fact that you could simply recompile PETSc, run
it natively on the card, and still run it on your CPU's as MPMD.  All in
all this seems like a big win for portable, maintainable code that can take
advantage of an accelerator.

Regards,
John


On Mon, Nov 12, 2012 at 7:06 PM, Karl Rupp <rupp at mcs.anl.gov> wrote:

> Thanks for the link.
>
> In terms of raw numbers, $2,649 for 320 GB/sec and 8 GB of memory is quite
> a lot compared to the $500 of a Radeon HD 7970 GHz Edition at 288 GB/sec
> and 3 GB memory. My hope is that Xeon Phi can do better than GPUs in
> kernels requiring frequent global synchronizations, e.g. ILU-substitutions.
>
> Best regards,
> Karli
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20121112/bce14e7f/attachment.html>