Decision Optimization

Expand all | Collapse all

CPU temperature on solving difficult MIPs

  • 1.  CPU temperature on solving difficult MIPs

    Posted Tue July 27, 2021 03:21 AM
    Hello,

    I am solving a series of difficult MIPs on Windows that I terminate after 2 hours of computation and record the unclosed gap at termination and number of unexplored nodes. While I expect this to be computationally intensive, I notice that the CPU temperature is around 96 degree celsius or so. See image here.

    While this is not an immediate problem, atleast according to this site where it appears that the Intel Core i7 7700 can go upto 100 degrees or so, I am concerned that the CPU temperature is so close to the maximum.

    Is there any recommended hardware / operating system benchmarks done by CPLEX folks that shed light on how intensive their computations can get on different platforms? For instance, there seemed to be a blog post here that talks of performance considerations on linux environments. Is there any recommendation / benchmarks available on running difficult MIPs on linux over windows platforms, or one hardware over another?

    Thanks.

    ------------------------------
    CPLEX User
    ------------------------------


  • 2.  RE: CPU temperature on solving difficult MIPs

    Posted Sun August 15, 2021 03:04 PM
    As an end-user, I can't give any specific benchmarks, or even much specific guidance. I will say that I once fried a laptop motherboard running a long (multi-hour) MIP model with CPLEX. Laptops are generally less adept at cooling than desktops, and this particular laptop happened to be sitting in a moderately warm location. Even so, it was very much an unpleasant surprise.

    If you are worried about CPU temp, you might start by making sure that the computer is in a cool, well-ventilated location and that the air filter in the case has been cleaned recently. (Sometimes dust accumulates and impairs ventilation.) Having a fan or a/c unit blowing cool air at it might also help. You can also lower the number of threads used by CPLEX to something less than the number of cores in the machine, which hopefully would reduce CPU usage and therefore heat generation ... at the likely cost of slowing solution of your model.

    ------------------------------
    Paul Rubin
    Professor Emeritus
    Michigan State University
    ------------------------------