[Rashi]

Hi, the company I'm working for is a startup electric sports car manufacturer. For our simulations we are planning to build a workstation with following specifications. The simulation software's we are mainly going to use are following
Ansys/solidworks for CFD and FEA simulations.
Abacus CAE for non-linear simulations.

Specifications for the workstation.
Processor 6900k
M.B. - Gigabyte x99P-SLI
Ram - Corsair 3000 Mhz 128GB
VGA - Quadro m2000
SSD - Samsung 950 Pro
HDD - 2tb WD SE

I'm planning to use the m.2 ssd as the cache drive for ansys and simulation file storage drive.

The issue I'm facing is with the amount of RAM. On a previous occasion I've used a 8 core workstation with 96GB ram for external body aerodynamic simulation. The mesh was around 10 million cells when I used all 8 cores in Fluent the machine would get stuck while consuming all 96gb of ram. The mesh had to be increased to get more finer result but was limited due to the ram limitation. So I ended up using less cores for the simulation.

That's the reason I went for 128gb of ram which is quite expensive. But think for more complex simulations like non linear analysis it will be help full and also when the simulations get more complex in the future.

What do guys think shall I go for 64gb ram and invest the rest of the money for some other component? But if I go for 64gb and if there's a ram limitation I'll have to wait long time to get money approved to go for 128 gb of ram.

Also if there's any one using a similar setup for their simulations please share your experience.

Thanks in advance.

[Rashi]

will 128 gb for a single 8 core cpu is too much?

[flotus1]

There is not really a "too much" when it comes to RAM
For external aerodynamics with cell counts typically in the range of 10-100 million 128GB of RAM seems reasonable. A very rough rule of thumb for simple aerodynamics is 1GB per 1 million cells.

On a previous occasion I've used a 8 core workstation with 96GB ram for external body aerodynamic simulation. The mesh was around 10 million cells when I used all 8 cores in Fluent the machine would get stuck while consuming all 96gb of ram. The mesh had to be increased to get more finer result but was limited due to the ram limitation. So I ended up using less cores for the simulation.

Not sure if I understand correctly. The memory usage decreased significantly with less cores? This is rather unusual. There is slight overhead due to the parallelization, but this should barely be noticeable.

[Rashi]

Originally Posted by flotus1

There is not really a "too much" when it comes to RAM
For external aerodynamics with cell counts typically in the range of 10-100 million 128GB of RAM seems reasonable. A very rough rule of thumb for simple aerodynamics is 1GB per 1 million cells.


Not sure if I understand correctly. The memory usage decreased significantly with less cores? This is rather unusual. There is slight overhead due to the parallelization, but this should barely be noticeable.

Flotus1 Thank you for your reply. That means going for 128 gb is acceptable right?

Sorry for the confusion. What I meant was when I used less cores for the simulation in fluent (4 cores) the memory didn't maxed out.

Also Flotus1 what do you think about the storage selection(m.2 SSD + HDD), and gpu selection (quadro m2000).

[flotus1]

Yes, 128GB are reasonable keeping in mind that RAM is very inexpensive at the moment and will probably become more expensive in the near future.
I am still baffled by your description concerning memory usage. The amount of memory used by a Fluent simulation should not depend on the amount of cores used. That is one of the key features of the parallelization with MPI and domain decomposition.
An PCIe-SSD might help to some extent, some Ansys software is notorious for disk I/O during simulations. When on a tight budget, a normal SATA-SSD will be enough. The Quadro M2000 seems to be a good choice in your case.