0

My question is general, though I have a specific example. My question is this: How can I predict whether a powerful CPU will be fully exploited in a scenario with memory that is (relatively) low?

My example is this: I'm specing out a Dell Precision 7920. I'm working with a programmer who will be using the system. We're going to install Linux and he'll be using his own code for a lot of number crunching. He's selected a powerful dual processor with 20 cores but he's selected only 32 GB RDIMM for the memory. I told a third person that the CPU wouldn't be fully exploited in this scenario but my programmer disagrees strongly and I may be wrong. I've realized that I don't know enough about this kind of question to answer with authority.

In general, any information about predicting bottlenecks is welcome.

MasterOfNone
  • 174
  • 1
  • 8

1 Answers1

0

That really depends on the application used, and what the application need to do. I guess for number crunching 32 GB RAM is enough, you may not even need that much, you may get away even with 2-4 GB. But if the app stores a lot of data in memory (intermediary results for example ), lot being more than that 32 GB, then you would need more memory. Otherwise app either dies from out of memory errors, or starts to swap which means time spent in IO wait instead for number crunching. Since the programmer knows his code, i would go with his recommendation. Especially since more RAM can be added later.

Sandor Marton
  • 1,564
  • 9
  • 12
  • I guess I'm really wondering whether or not there's a more general rule or method I can use to predict this, and if it varies by application then how it could be predicted for different kinds of tasks. In the case of my example I probably do just need to go with the coder's recommendation, but what about next time? – MasterOfNone Aug 29 '18 at 12:07
  • You always go with the coder's recommendation :) If you move data around ( reading files/databases etc) then you could benefit from having more RAM ( app doesn't even have to store a lot of data; file system caches could benefit from the RAM making faster some reads - when you read the same data/file multiple times ) . But number crunching wouldn't use that much RAM. I don't really see how you can predict, unless you know the app, or have experience with similar tasks – Sandor Marton Aug 29 '18 at 16:25