Hi u can go for a system with i7 processor, SSD, good clock speed.

if you have 32 gb ram you can go with that otherwise you can go with 16 gb ram.

Dear Louis,

I would say that you need to consider the size of your problem and the kind of algorithm you are going to use. CPLEX is just one of the possible solver.

The point is that you are likely to use a solver based on branch&bound techniques that usually requires a lot of memory and recently also exploit multicore/multiprocessor architectures. So If your problem is hard to solve and/or of large size, take as much memory as you can, and then try to increase the number of core.

Hi,

As already mentioned, you should prefer CPUs having higher number of physical cores (Intel Xeon E5 CPUs have up-to 10 cores).

Memory above 32GB is probably useless, IP problems are NP-complete in the general case, so running times will be an issue before memory space.

Hard drive IOps shouldn't be an issue unless you spend your time reading/writing your IP problems into files, so SSD is not mandatory.

I configured a machine with the latest i7 Extreme processor at the time (two months ago.. still th ebest I think). I execıtes CPLEX like I have not seen before! You can give permission to CPLEX to use all of your cores in parallel, so a good number of physical cores will make a big difference in solving MIP problems where parallelization makes a huge difference in phatoming the branches. RAM and SSD wont matter much, unless, as mentioned, you will do a lot of parameter-reading and result writing (though I do have 32 GB RAM and 512 GB SSD).

@Akiner Tuzuner

Is it possible to explain how to give permission for increasing the dedicated memory? thank you  

I am using AIMMS mathematical software with CPLEX as the solver. However, I found that when my problem is getting bigger, laptop is hang and not responding. What do you think? Im using laptop features Core i5, 8.00 RAM and 1.87 GHz cpu.. How can I explain this in my study?

No single computer configuration works best for all integer programs.   The performance challenges associated with integer programming vary.   Some MIPs are very modest in size, but very difficult to solve due to a huge branch and bound tree.   In that case, 12-16 cores, or maybe even more, may be helpful, as the loads on each thread will probably be will balanced enough to benefit from a relatively large number of cores.    Other MIPs may be much larger, and the node LP solve time rather than tree management may be the performance bottleneck.    In those cases, the loads on each thread will be less balanced, and you will be better off with a configuration with fewer cores but higher chip speed and larger cache size.   And of course there are other models that are in between these two extremes, where both efficient tree management and fast floating point computation for the node LPs can help.   But, overall we find that most MIPs run best with somewhere between 4 and 16 threads for current and recent versions of CPLEX.  

Regarding the notion of "balanced loads" that I mentioned above, if you are unfamiliar with this term, it refers to differences in workloads among different threads in a multi-threaded program.   For branch and bound in MIPs, this means that different parts of the tree are solved on different threads, and then after a while the algorithm synchronizes the threads and updates the branch and bound tree to account for the work of each individual threads.    If the subtrees solved on each thread involve substantially different amounts of work, the loads are unbalanced, and the algorithm may need to wait for the more time consuming threads to finish when it synchronizes, slowing performance.   Thus, the balance (or lack thereof) of the work among the threads can play a significant role in performance.

Regarding memory usage, here is a link to a technote that provides some guidance on CPLEX and memory usage: 

http://www-01.ibm.com/support/docview.wss?uid=swg21399933

Summarizing, consider the characteristics of the kinds of MIPs you need to solve when assessing the best computer configuration.   A good configuration for one class of MIPs may not be as good for others.