[matthewe]

Hi All!

Long time reader, first time poster

We are about to buy a new machine which will mostly be used for CFD simulations using Fire Dynamics Simulator (FDS) which is run using MPICH.

Due to customer support issues we are likely getting a Lenovo C30 system with 2 x Xeon E5-2650 v2's (not my choice, IT departments choice).

I understand that the CPU is a key component that affects simulation speeds however memory also has a huge impact.

Am I right in thinking that it would be better to provide 8 x 4 GB of memory (The CPUs have 4 memory channels each) rather than 2 x 16 GB? Most of the models that are run will use from 2GB to 8GB of memory, with more than 20GB being a very rare event.

I have spoken to Lenovo and they don't seem to know much - see the online conversation below. However, they clearly said that less sticks will give you more speed.

I've also been bounced around sales and technical on the phone without much success.

Is anyone able to suggest if the extra cost involved in 8x4GB will give a performance boost over using 2x16GB? I would hate for us to buy a $5k machine and lose out on the best performance because we messed up on the memory. I would really appreciate any help

Regards,

Matt


Online conversation with Lenovo:

Please wait while we connect you with a Lenovo Chat Representative.

Connected to Harathi Vemishetty
Harathi Vemishetty : Thank you for contacting Lenovo Sales Chat. My name is Harathi and my Rep ID is 2900711681. Happy to help you today. Hello, may I know your name please?



You : Hi, my name is Matt
You : I have a quite technical question
Harathi Vemishetty : I will try to answer Matt
Harathi Vemishetty : We have dedicated technical team for this
You : We are about to buy some new machines which will mostly be used for Computational Fluid Dynamics simulations using the Fire Dynamics Simulator program. The simulations are very memory intensive and are normally run using MPICH. We are likely getting Lenovo C30 systems with 2 x Xeon E5-2650 v2's. I understand each Xeon processor has 4 memory channels and there are therefore 8 memory channels in total. Am I right in thinking that it would be better (higher performance) to provide 8 x 4 GB of memory rather than 2 x 16 GB? The key issue I see is that if you only use 2 of the DIMM slots then you are only maximising two of the memory channels. In that case the memory transfer will be limited. If you use all 8 DIMM slots you will use all the memory channels and maximise throughput. Therefore 8 x 4GB will have more transfer bandwidth than 2 x 16GB. I also realise that 8 x 4GB will be more expensive and will limit upgrade options in the future.
You : Sorry, this tool doesn't like paragraphs.
Harathi Vemishetty : Let me read
Harathi Vemishetty : Have you customized C30?
You : That is the question, when we buy the new machines will customising it with 8 x 4GB RAM modules give a performance boost compared to 2 x 16GB?
Harathi Vemishetty : Let me check that for you.
Harathi Vemishetty : Please give me few minutes
You : Ok, no problem, thanks.
Harathi Vemishetty : Thank you
Harathi Vemishetty : Thank you for your time. 2 x 16GB will give better performance
Harathi Vemishetty : May I know if you are customizing it now to order?
You : Yes
Harathi Vemishetty : Okay
Harathi Vemishetty : I will be online while you place the order
You : Why will 2 x 16GB give better performance? Can you explain a little?
You : Why would not using all the 8 memory channels give better performance?
Harathi Vemishetty : I will explain you
Harathi Vemishetty : System will take some of the memory capacity from every RAM stick
Harathi Vemishetty : This is just an approximate figure. When a 8GB RAM is added, the actual RAM you get will be 7.5GB or so, that way you'll loose more capacity of the RAM when you add more sticks.
You : Ok, I can understand that. So 2x16GB will give you more total memory than 8x4GB.
You : However, capacity is not the main concern. I am concerned with speed. I want the fastest memory to CPU transfer available.
Matt, less # of memory sticks will give you more speedYou : Ok, that's interesting. Thanks.
Harathi Vemishetty : You are welcome