Advice Needed: Choosing HPC Nodes for Large-Scale MIP – AMD EPYC vs. Intel Xeon

Answered

Mohammad Derakhshi

• April 16, 2024 17:56

Hello Gurobi Community,

I am seeking advice on selecting HPC nodes for solving large-scale MIP models using Gurobi with MPI4PY. We have two options on the table:

• Option A: R6525 with dual AMD EPYC Rome 7452, 32-core, 128 GB RAM
• Option B: R650 with dual Intel Xeon Ice Lake 6338, 32-core, 128 GB RAM

Could anyone share their experiences or insights on the performance of Gurobi with these CPUs, especially in terms of scalability and efficiency in large-scale optimization? Any benchmarks, comparative insights, or suggestions would be greatly appreciated as we aim to make an informed decision.

Thank you in advance for your help!

Best regards,
Mohammad 

Related to

• Performance

0

• 

  Mario Ruthmair

  Gurobi Staff

  • April 26, 2024 07:21

  Hi Mohammad,

  Here are a few general recommendations:

  • What hardware should I select when running Gurobi?
  • One Gurobi run might not be able to keep all 2x32 cores busy all the time since not all parts of the solver can be parallelized, so the single-thread performance of a CPU is very important, too. Especially the turbo-boost frequency is relevant here.
  • The more L1/2/3 cache you have, the better.
  • The wider and faster the memory bus, the better.

  In the end, it is hard to compare Gurobi's performance based on spec sheets. You might just need to try. Maybe you can temporarily get AWS/Azure machines with similar specs to do a few tests.

  We have 16-core AMD EPYC 7313P CPUs that perform very well.

  Best regards,
  Mario

  0
• 

  Mohammad Derakhshi

  • April 26, 2024 17:13

  

  

  Dear Mario,
  
  Thank you for your detailed and invaluable response.

  Best,
  Mohammad

  0

Please sign in to leave a comment.