open TELEMAC-MASCARET The mathematically superior suite of solvers

In the file enclosed you will find tables showing the scalability of Telemac-2D and 3D from 1 to 8 processors (cores ?) in the Malpasset test-case (small and large mesh), on a desktop HP Z600 Linux machine with Intel compiler.

Jean-Michel Hervouet

Thank You Jean-Michel,

Interesting to see the "super"-linear behaviour still with 2 cores. Also, this continues to confirm that multi-core processors do not behave like several processors. Using 6 cores off you 8 core processor is our recommended maximum. This of course has nothing to do with TELEMAC's scalability (it would be 8 time faster on the proper architecture) but just show the limitation of the current multi-core processors.

Sebastien.

Hello

I recently installed Telemac 6.2 on a new computer at Port of Antwerp.
This is an IBM x3850 X5 (2 modules) which offer 1 computational node with 8 processors Intel I7-8837 8C (64 cores at all).
The compiler is Intel Cluster Studio XE 2013 (include IMPI).

On the large mesh of malpasset validation case in 2D, we obtain the following results:

Number of Cores	Computation Time in s	Speed-up
1	519	1
2	230	2.26
4	111	4.68
8	59	8.80
12	42	12.36
16	32	16.22
20	28	18.54
24	25	20.76
28	22	23.59
32	20	25.95
40	18	28.83
48	17	30.53
64	16	32.44