What computer setup for CFD simulations 1000$-2000$ budget?

Click For Summary

Discussion Overview

The discussion revolves around the optimal computer setup for running computational fluid dynamics (CFD) simulations within a budget of $1,000 to $2,000. Participants explore various hardware options, including CPUs, GPUs, memory, and storage solutions, while considering the specific needs of CFD applications.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested
  • Mathematical reasoning

Main Points Raised

  • One participant suggests purchasing a refurbished machine for under $500 and upgrading it with additional memory and a GPU to fit within the budget.
  • Another participant mentions using two Intel i5-4570 CPUs with 8 GB of RAM each for different operating systems, questioning the necessity of dual CPUs for a single Linux setup.
  • Some participants discuss the importance of a familiar operating system and the use of publicly available libraries versus more advanced programming skills.
  • Concerns are raised about the limitations of high-performance computing (HPC) access and the benefits of transitioning to high-end workstations with ample memory and multiple GPUs.
  • There is a suggestion that approximately 16 GB of RAM per core is advisable, with a recommendation for 64 GB total for an i5 CPU with 4 cores, though this is noted to depend on specific workload requirements.
  • Participants express differing views on the reliability and longevity of SSDs versus traditional spinning disks for data storage, highlighting the finite write cycles of SSDs.

Areas of Agreement / Disagreement

Participants do not reach a consensus on the best approach to building or purchasing a computer for CFD simulations, with multiple competing views on hardware specifications, the necessity of certain components, and storage solutions remaining unresolved.

Contextual Notes

Participants mention various assumptions regarding workload requirements and the specific needs of their CFD applications, which may influence hardware choices. The discussion reflects a range of experiences and opinions on the trade-offs between different setups.

member 428835
I'm running computational fluid dynamic simulations and am looking into building a desktop. But there are so many options. At this budget, what would you buy?

Is it better to buy used CPUs, etc?
 
Computer science news on Phys.org
you can get a refurbished machine for less than $500, add as much memory as you can and add a gpu and get into the $1-2K range. you should be able to run cfd calculations on a machine that size.

I bought two intel i5-4570 @ 3.2GHz with 8 GG of ram for ~$300 each. One I use as a Linux box, the other as a Windows box.
 
Dr Transport said:
you can get a refurbished machine for less than $500, add as much memory as you can and add a gpu and get into the $1-2K range. you should be able to run cfd calculations on a machine that size.

I bought two intel i5-4570 @ 3.2GHz with 8 GG of ram for ~$300 each. One I use as a Linux box, the other as a Windows box.
Thanks for the response. Is there a reason to get 2 cpus if I'm only running Linux? Also, is there a reason to purchase the gpu?
 
joshmccraney said:
Is it better to buy used CPUs, etc?
The answer depends on your level of sophistication. Do you want a familiar OS? Do you want to use publicly available libraries? High level languages?

Or are you handy with a soldering gun and bit level machine programming?
 
anorlunda said:
Do you want a familiar OS? Do you want to use publicly available libraries? High level languages?
A familiar OS? The CFD platform I use (OpenFOAM) is native with Ubuntu, so I think I should stick to that. Public libraries: I use open-source software that run on C++, but I don't do much programming. Just edit some of the directories mildly and then let the program do it's thing. I run simulations in parallel, currently 4 cores.

anorlunda said:
Or are you handy with a soldering gun and bit level machine programming?
Hahahaha not at all, no machine learning or a soldering gun (only done this in a lab, and wouldn't trust myself on an expensive machine).
 
joshmccraney said:
Thanks for the response. Is there a reason to get 2 cpus if I'm only running Linux? Also, is there a reason to purchase the gpu?
No, I mentioned that because I have two separate machines. I'd get one, add memory then possibly add a gpu. I have interacted with cfd groups for years, one stopped using the HPC's they had access to to convert ttheir codes to high end workstations with gobs of memory and multiple gpu's. If your wondering why, my former employer, although they wanted us to use the hpc for large runs, they also severely limited our access, i.e., we had to request time on a machine a year in advance and they told us what block of time we had to use it and if you missed your run-times, oh well, put in a request for next year and wait and see.
 
  • Like
Likes   Reactions: BvU
Dr Transport said:
No, I mentioned that because I have two separate machines. I'd get one, add memory then possibly add a gpu. I have interacted with cfd groups for years, one stopped using the HPC's they had access to to convert ttheir codes to high end workstations with gobs of memory and multiple gpu's. If your wondering why, my former employer, although they wanted us to use the hpc for large runs, they also severely limited our access, i.e., we had to request time on a machine a year in advance and they told us what block of time we had to use it and if you missed your run-times, oh well, put in a request for next year and wait and see.
Thanks for the response! So your recommendation is to not build one from scratch, but purchase a refurb, upgrade RAM, SSD? I was told approximately 16 GB ram/ core. Does that sound right to you?

Yea, I'm not a CFD guy, but some people at my school are, and they go through the same issue of requesting time on the clusters. Seems like a real pain.
 
joshmccraney said:
Thanks for the response! So your recommendation is to not build one from scratch, but purchase a refurb, upgrade RAM, SSD? I was told approximately 16 GB ram/ core. Does that sound right to you?

Yea, I'm not a CFD guy, but some people at my school are, and they go through the same issue of requesting time on the clusters. Seems like a real pain.

An i5-cpu has 4 cores, so 64 GB is reasonable. It all depends on your budget and what your codes need. Memory is the big ticket item, I've only got 8 GB on my systems and I run fine, but I'm not doing huge sims at the house either. I could go with divorce class memory for my machine, but I don't want a divorce and I don't need it right now.

Be careful with ssd's also, I found out the other day that we've been seeing them fail due to lifetime issues, they only can be be written to so many times before they fail. I use spinners for all of my data and ssd's for OS only.
 
  • Like
Likes   Reactions: BvU
The best specs for OpenFOAM depend on your workload. Large models need memory more than CPU. Small models it's the reverse.

It is true that SSDs have a finite lifetime - about 10,000 writes. It's also true that spinning disks have a lifetime. Spec sheets have numbers like 180 TB/year, which for a 6 TB drive is 30 reads per year. Depending on your workload one is better than the other.
 

Similar threads

  • · Replies 2 ·
Replies
2
Views
2K
Replies
4
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 7 ·
Replies
7
Views
6K
  • · Replies 24 ·
Replies
24
Views
3K
  • · Replies 20 ·
Replies
20
Views
5K
Replies
2
Views
1K
  • · Replies 5 ·
Replies
5
Views
8K
  • · Replies 1 ·
Replies
1
Views
5K
Replies
10
Views
4K