Help with RAM choice for Fluent Simulations!!!
Hi guys,
I' m between four RAM choices in order to run demanding Fluent Simulations! 6GB 7-8-7-20 1600MHz 6GB 9-9-9-24 2000MHz 12GB 9-9-9-24 1600MHZ 12GB 9-10-9-27 2000MHz What is the best solution to achieve: 1) run a simulation faster 2) run a grid with 2-3 million cells or more 3) run many fluent simulations parallel or which characteristic of RAM (size, speed rating and tested latency) affects 1) 2) and 3)?? Thanks, spk |
I think i' ll choose 12 GB to run more cells!
My question now is between these RAM, which is faster? 12GB 9-9-9-24 1600MHZ 12GB 9-10-9-27 2000MHz I read a previous thread in this forum that a change from 1600MHz to 2000MHz speeds up only 10%. How the tested latency affects the RAM speed? |
Buy better 24 GB of cheaper RAM (with worse timing) than 12 GB. RAM is now very cheap. I can recommend OCZ, G.Skill, Patriot or GEIL modules
|
You might need to make sure your cpu/motherboard can handle ram at 2000. The westmere xeon's I got last year only officially support ram at 1333. Could probably mess around a little with them to get them to run higher speeds.
Faster ram will help, not sure if you need the extra capacity if you only doing 2-3 mil |
Thanks for the answers!!
Do you know how many cells can solve with 24GB or 32GB ram? |
I have also an another question!
I think to buy a motherboard with 8 slots with 64GB max. One 6-core CPU could use all this memory 64GB? |
Just for fun ran some tests on the 2-d code I'm working on. I have a 24gb of ram in 6 4gb chips. I managed to max out at roughly 64 million cells. This code have several extra fields it has to worry about because it is a two phase model on a staggered grid, and I probably could optimize it for memory and get the cell count into the 100's of millions.
That being said I rarely go over a million or two cells because the simulations just take so long. I would say you would run out of cpu power long before you run out of addressable memory. So while you could buy 64gb of RAM you probably couldn't solve any problem that used all 64gb. Better choice would be to buy lower capacity, faster ram. For regular grids you will tend to be bound by memory access. You are correct though in the ram speed question, roughly a ten percent increase in speed between the two timings you lists. The faster ram has slower timings so the increase isn't as much. Yes your processor could access all the ram as long as the mother board supports it. Just because a processor has tri-channel or quad-channel memory doesn't mean it can only access three or hour memory chips. If you really want to do large simulations >10 mill cells it might help to get a gpu, although I'm not sure if fluent supports them and if they do it probably costs extra. |
Quote:
Do you think that i would run out of 4-core cpu power long before i run out of a 32 GB memory? Or do you suggest me 24 GB? |
It really depends on what type of problem you are solving but if your expecting to get results back in a reasonable time then your probably not looking at solving problems with more than a couple mil variables.
For reference the jaguar super computer consists of compute nodes with dual hex core AMD processors with a total ram of 16GB. This is 1.3 Gb RAM per core. |
All times are GMT -4. The time now is 08:04. |