I have an extensive use of MKL libraries. I am using PARDISO for my applications. At present, our system configuration is SGI with 8 single core processors and total memory is 32 GB.
Our problem size is very large (1.5 million degrees of freedom). Total number of right hand side (RHS) vectors is in the range of 15000. At present I am solving the problem by taking around 100 RHS at a time, and therefore it takes at least a week to finish the 15000 RHS.
We are planning to buy a new computer system that will reduce this processing time (by solving more RHS at a time). My question is can I reduce this time from 8 days to 2-3 days? Is this possible if I go for 128~256 GB of memory from 30 GB?
I have selected following systems at present,
1. Intel Xeon E5-2670(8 cores, 2.6 GHz, 128 GB Memory) - total 16 cores
2. Xeon Processor (4 cores, 1.86 GHz, 128 GB Memory) - total 16 cores
which of the above will do the job faster? Is there any other suggestion like Xeon E5-4650?
Thank you very much.