CUDA and POY

[Buz Wilson]

Hello POY group

I just installed a new workstation (Ubuntu mate 16.04) and I'm building
software for phylogenetic analysis. The workstation has a decent Gforce
card (1050 TI) with 768 CUDA cores so I'd like to get those running with
POY, too. I need to compile OpenMPI with CUDA support (apparently not
availablein the repositories), but do I need to compile 5.1.2 with any
flags to support CUDA? Incidentally, POY 5.1.2 compiled just fine with
the OCAML from the repositories, so the previous Ocaml bug seems to have
gone away. Any advice appreciated.

Best wishes for 2016!

Buz


--
George D.F. (Buz) Wilson, Ph.D.
Saugatuck Natural History Laboratory, LLC
email: gd...@SNHLab.com website:www.SNHLab.com
http://www.researchgate.net/profile/George_Wilson13

[Nick Lucaroni]

Unfortunetly, CUDA isn't free parallelism for software and would require careful consideration and programming.

For example, we take advantage of SSIMD floating point registers (while avoiding spillover) and bit packing data, but other than that our mode of parallelism is geared toward large scale MPI based parallelism. 

--

nL

--
You received this message because you are subscribed to the Google Groups "POY - Phylogenetic Analysis Software" group.
To unsubscribe from this group and stop receiving emails from it, send an email to poy4+unsubscribe@googlegroups.com.
To post to this group, send email to po...@googlegroups.com.
Visit this group at https://groups.google.com/group/poy4.
For more options, visit https://groups.google.com/d/optout.