Best high performance computer build for finite element simulations

[KyleGranger]

TL;DR

I want to build a computer that will be optimized for finite element analysis.

I want to build a high performance computer optimized for FEA. If anyone has any suggestions or experience with this, please chime in.
I'm looking for suggestions on processors, RAM, video card, ALL of it!

 

[Vanadium 50]

Budget? If cost is no object, I'd suggest something like this:

 

[jrmichler]

The software manufacturer will tell you this. Search XXX fea computer requirements to find it easily.

You are far better off learning how to make models that run quickly and accurately. Mesh sizes appropriate for the needs, defeature as appropriate, optimal restraints, etc.

I once helped an analyst by taking a model that was running for several hours, then crashing. After defeaturing, cleaning up the restraints, and adjusting the mesh size, it ran in one minute with accurate results.

 

[KyleGranger]

The software manufacturer will tell you this. Search XXX fea computer requirements to find it easily.

You are far better off learning how to make models that run quickly and accurately. Mesh sizes appropriate for the needs, defeature as appropriate, optimal restraints, etc.

I once helped an analyst by taking a model that was running for several hours, then crashing. After defeaturing, cleaning up the restraints, and adjusting the mesh size, it ran in one minute with accurate results.

I will definitely work on learning how to model and mesh for speed and accuracy. I would still like to have a machine capable of running more complex models as I see fit.

I'll primarily using Ansys and Comsol as well as Matlab periodically. I don't expect Matlab to be to demanding for what I'll be doing there though. Ansys offers support for high performance computing (HPC) so I don't think I'm limited in terms of what I can benefit from. I'll surely be exceeding their minimum requirements though.

If this is more of a Reddit question and not appropriate here, just let me know. I figured I would ask it here since there are sure to be people in the STEM field that have experience with this or who have the same question.

The key parts that I need are CPU, RAM, GPU, and motherboard. I've tried reading on AMD vs Intel and have settled on AMD, unless I can be swayed the other way. I'm not sure when it comes to motherboard and graphics card. For the full system, I would like to keep it between $2500-$3000. I know that I need other pieces such as cooling, case, etc so I'll consider that as well. I'm hoping to find feedback on the computational performance.

CPU: AMD Ryzen 9 5950x
RAM: 4x32GB Patriot Viper Steel DDR4 3600

 

[Vanadium 50]

If you are using commercial ANSYS, the computer cost as you describe is less than 10%. Possibly much less. Don't you want something as fast as you can get?

If you are using student ANSYS, you're limited to two cores. Why do you need 16?

 

[KyleGranger]

If you are using commercial ANSYS, the computer cost as you describe is less than 10%. Possibly much less. Don't you want something as fast as you can get?

If you are using student ANSYS, you're limited to two cores. Why do you need 16?

I'm a student, but I have access to the HPC license so I'll be able to use all 16 cores. I would like to focus what I have to spend on the pieces that will be the most impactful. I currently have an older i7, 32GB RAM, and a SSD.

 

[Vanadium 50]

Whether you would do better with a fast CPU, GPU, more memory, etc. depends on the model you are running (and on the license you have). I don't think we can tell. However, performance is much mode dependent on the skill of the modeler than the hardware.

I would also question whether, as a student, spending $2500 on this is wise. If you are taking out student loans, this might be a bad idea. I would say it makes a difference if you finish your degree sooner, not your homework sooner.

 

[bigfooted]

As a student, I'd:
1. find out how to get access to the group/faculty hpc (usually a small local cluster somewhere in the basement), this is sometimes possible for students even outside of MSc/PhD projects if they have good project ideas.
2. spend money on a 'normal' pc (not a 'power beast') or maybe a small, energy efficient workstation. I find that it's usually the memory and (fast!) storage that limits you. Get one or two good quality monitors if you haven't already. Especially for CAD visualization, this is important. Do the real computing on the faculty hpc.

 

[KyleGranger]

I would also question whether, as a student, spending $2500 on this is wise. If you are taking out student loans, this might be a bad idea. I would say it makes a difference if you finish your degree sooner, not your homework sooner.

These are not loans and it will definitely help me finish sooner and/or have better quality work. If nothing else, I'm seeing this as an investment in my education/future. I'm sure plenty of students spend this much or more just going out on weekends...that's how I'm looking at it anyway. But I posted to get others opinions on this that have a better idea than myself so thanks for your input. I will look at a 'normal' pc to weight my options. I definitely need to improve my modeling skills.

 

[KyleGranger]

As a student, I'd:
1. find out how to get access to the group/faculty hpc (usually a small local cluster somewhere in the basement), this is sometimes possible for students even outside of MSc/PhD projects if they have good project ideas.
2. spend money on a 'normal' pc (not a 'power beast') or maybe a small, energy efficient workstation. I find that it's usually the memory and (fast!) storage that limits you. Get one or two good quality monitors if you haven't already. Especially for CAD visualization, this is important. Do the real computing on the faculty hpc.

We have HPC we can get access to through the university. Unfortunately, I have to submit my jobs through the command prompt and that's what has pushed me away from using them. I know it's something that I should familiarize myself with so I'm not as limited. Thanks for the suggestions. I think I'll reconsider this build and look at less expensive options. I thought it would be a good idea to help move me along my way more efficiently but you have brought up some good points.