[Pw_forum] If some geek wants some fun: pwscf on cell.

Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu
Sun Nov 26 03:44:20 CET 2006

On Sun, 26 Nov 2006, Antonio Suriano wrote:

AS> IBM has developed a new fast parallel processor called Cell with 8 cores.
AS> This processor is specialized in operations on vectors and matrices and 
AS> it is piloted by an evolution of PPC.
AS> Some people can run Linux on this kind of machines (Yellow DOG Linux).
AS> I think it will be very interesting if some computer geek would find an 
AS> hack to run PWSCF on this kind of machines.

note, that in its current form, cell has only single precision floating
point. that does not hurt (much) with classical MD codes like gromacs
(which is what folding at home uses) if they are carefully programmed, and 
you don't need accurate dynamics, but only reasonable statistics.

AS> Actually there is this supercomputer that has a fast evolution of G5 as 
AS> core unit + an 8 core
AS> CELL for fast vectorial operations + 2 very fast floating points unit id 
AS> est 11 overall processor
AS> It has 20 gigabytes hard disk and 512 mega ram (256 for the core and 256 
AS> for the floating point vectorial units) that runs at the same speed of 
AS> the processors+ a dvd player.
AS> This supercomputer has an overall performance in floating point of 1.8 
AS> teraflops.

please be careful with those numbers, this is most likely peak 
performance, i.e. pointless for any practical applications...

AS> This supercomputer  costs only 599 euros, it is developed by Sony  and 
AS> its name is Playstation 3.
AS> It is not a joke, some people are doing protein stuff only using the 
AS> free processor time of machines running videogames meanwhile:
AS> http://en.wikipedia.org/wiki/Folding%40home

you also need a very special compiler for that kind of hardware.
last time a heard. a cell development kit (including hardware) 
was around $20k... this was about half a year ago, though.

for applications like quantum espresso, the most promising
route to make this kind of hardware useful, would be to
a) make sure the code uses BLAS/LAPACK whereever possible and
b) help people like jack dongarra to build a BLAS/LAPACK that 
works with a mix of single and double precision, yet provides
full double precision accuracy.

btw: if you go and look at the folding at home statistics, you'd
see that you can actually make a supercomputer from your desktop
by adding a 3d graphics card. current GPUs have even more 
power than cell (it is just a PITA to program them and get
the data off them).


AS> Chiacchiera con i tuoi amici in tempo reale! 
AS>  http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com 
AS> _______________________________________________
AS> Pw_forum mailing list
AS> Pw_forum at pwscf.org
AS> http://www.democritos.it/mailman/listinfo/pw_forum

Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
If you make something idiot-proof, the universe creates a better idiot.

More information about the users mailing list