Hi Gang,
Parallelisation efficiency will depend on the number of MPI nodes and
threads as well as on the number of calculations to be made per particle
(more for 3D refinement than for 2Dclass averaging). You can monitor
efficiency of threads and MPI nodes using top on the nodes you´re running
on. N threads should lead to N*100% CPU in top for perfect efficiency.We
often see N*70-80%.
Hope that helps,
Sjors
> Hi£¬
>
> I am wondering whether anyone have run Relion 1.2 on AMD cpu clusters. I
> have successfully compiled Relion 1.2 with GNU compiler under
> openmpi1.4.3-gnu. Each node of our cluster is equipped with 2¡Á AMD
> Opteron 6136 £¨2.4G£©CPU and 64G DDR3 1333MHz Memory and connected with
> 40GB QDR Infiniband. Unexpectedly, the calculation with 8 such nodes is
> even at least four times slower than a single computer with 8 core I7
> (2.97G) and 8G DDR3 1333MHz.
>
> Could anyone help accelerate the parallel computation with AMD cpu
> cluster? Could I compile Relion 1.2 with pgi compiler? Could pgi
> compilation or upgrade the openmpi could make a difference? Many thanks
> in advance!
>
> Gang Cai
>
> =======================================
>
> Gang Cai, Ph.D.
>
> Professor, Principal Investigator
>
> Schoole of Life Science,
>
> University of Science and Technology of China
>
> No. 443 Huangshan Rd, Hefei,
>
> Anhui, China, 230027
>
> Tel/Fax: +86-551-3603802
>
> Email: [log in to unmask]
>
> http://cii.ustc.edu.cn/cailab/
>
> =======================================
>
>
--
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres
|