Cluster or workstations for turbomachinery CFD?

k_k · March 11, 2012, 04:23

Hello all,

Need some advice.

I currently use CFD for turbomachinery, with models little more than half a million cells (one blade passage steady state) and the solution converges in about 3 hours (2000 sweeps) on a 6 core PC. I am about to use several full geometry model with about 10 million cells for both steady and unsteady computation. I am sure that I need additional computational resources.
I am planning to expand the computational resources so that I would be able to see the solution within 12 hrs.
I am confused what to get, single or several high end PC/work station or a small cluster??

1. which would be suitable for my case?
2. which would be better on a longer run for lesser CFD turn around times?
3. which is more advantageous in terms of maintenance, flexibility and cost effectiveness?

Kindly share your expertise

Kind Regards,
K

abdul099 · March 15, 2012, 03:42

For your case, you might go for a small cluster or several workstations. A single PC will not give you enough power to reach your desired turnaround times (just a guess).

The full case has about 20 times more cells and you've got 4 times more time until you want to get the results. That means, you need at least 5 times the computing power. Maybe even more due to the worse scaling on high CPU counts.

A small cluster with infiniband interconnect will give you a better performance compared to several workstations. That could be interesting for the unsteady cases since the usually run for a long time. But cluster hardware is pretty expensive, you need a good cooling and an upgrade is usually too complicated.

Several workstations are more flexible in terms of flexibility, mainenance and I guess also cost effectiveness.

The cheapest way and the most flexible in terms of hardware upgrade etc. are several PC's, but for the cost of the lowest performance. But maybe you can purchase much more PC's for the price of the small cluster or the workstations, so you can run 2 cases at a time...

March 11, 2012, 04:23	Cluster or workstations for turbomachinery CFD?	#1
k_k Member Kailash Join Date: May 2011 Location: London, UK Posts: 45 Rep Power: 14	Hello all, Need some advice. I currently use CFD for turbomachinery, with models little more than half a million cells (one blade passage steady state) and the solution converges in about 3 hours (2000 sweeps) on a 6 core PC. I am about to use several full geometry model with about 10 million cells for both steady and unsteady computation. I am sure that I need additional computational resources. I am planning to expand the computational resources so that I would be able to see the solution within 12 hrs. I am confused what to get, single or several high end PC/work station or a small cluster?? 1. which would be suitable for my case? 2. which would be better on a longer run for lesser CFD turn around times? 3. which is more advantageous in terms of maintenance, flexibility and cost effectiveness? Kindly share your expertise Kind Regards, K

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CFD Design...The CFD Future	John C. Chien	Main CFD Forum	20	November 19, 2015 23:40
is there any money in CFD?	T	Main CFD Forum	35	May 9, 2001 19:35
CFD for fans & blower housings	David Carroll	Main CFD Forum	8	August 24, 2000 17:25
PC vs. Workstation	Tim Franke	Main CFD Forum	5	September 29, 1999 15:01
public CFD Code development	Heinz Wilkening	Main CFD Forum	38	March 5, 1999 11:44

March 15, 2012, 03:42		#2
abdul099 Senior Member Join Date: Oct 2009 Location: Germany Posts: 636 Rep Power: 21	For your case, you might go for a small cluster or several workstations. A single PC will not give you enough power to reach your desired turnaround times (just a guess). The full case has about 20 times more cells and you've got 4 times more time until you want to get the results. That means, you need at least 5 times the computing power. Maybe even more due to the worse scaling on high CPU counts. A small cluster with infiniband interconnect will give you a better performance compared to several workstations. That could be interesting for the unsteady cases since the usually run for a long time. But cluster hardware is pretty expensive, you need a good cooling and an upgrade is usually too complicated. Several workstations are more flexible in terms of flexibility, mainenance and I guess also cost effectiveness. The cheapest way and the most flexible in terms of hardware upgrade etc. are several PC's, but for the cost of the lowest performance. But maybe you can purchase much more PC's for the price of the small cluster or the workstations, so you can run 2 cases at a time...