CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > General Forums > Hardware

Yet another workstation configuration question

Register Blogs Community New Posts Updated Threads Search

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   October 16, 2012, 15:13
Default Yet another workstation configuration question
  #1
RSE
New Member
 
Reuven Shenkar
Join Date: Oct 2012
Posts: 3
Rep Power: 14
RSE is on a distinguished road
Hi there,

After not working with CFD for about a decade I am returning to that from another direction. This time the issue is environmental & wind energy (wind resources, flow over hills etc.). Therefore I would like to buy my first workstation and start fiddling around with OpenFOAM. This will be the main task of the workstation, though in the future I may want to do some medium-level solid CAD (SolidWorks etc.).

I have already read some of the threads here and generally I understand that faster cores are more important than the actual number of cores.

I don't have that much knowledge of HPC and I have never built a machine of my own, therefore I would opt for an off-the-shelf machine, though I understand that it comes at a (big) price.

My budget is ~$2K. I was looking at the Dell T3600 with Xeon E-1650 and 8GB, DDR3 RDIMM Memory, 1600MHz, ECC memory, it is a bit above $2K. As I would like to have Red Hat Linux installed instead of Win7, I may be able to get it for $150 or so less (if they let me).

I was also looking at the T5600 with dual four-core E5-2609 and 16GB, DDR3 RDIMM Memory, 1600MHz, ECC memory. At about $3K it is way beyond my budget; however, if you think it is really worth the extra money than I may well consider it. I must say that I would really like this machine to stay relevant for 3-4 years to come (if possible...).

I would appreciate to have your recommendations. Any comments? Suggestions?

Thanks, R.
Attached Images
File Type: png T3600.png (28.1 KB, 26 views)
File Type: png T5600.png (33.8 KB, 19 views)
RSE is offline   Reply With Quote

Old   October 17, 2012, 08:58
Default
  #2
Senior Member
 
Charles
Join Date: Apr 2009
Posts: 185
Rep Power: 18
CapSizer is on a distinguished road
If you are chasing the best performance for your money, I doubt that Xeon is the best option. There are companies that will build you a tricked out Core-i7 "gaming" rig with good cooling and super-fast memory. Typically very good value for money if you don't overdo the bling and the graphics card.

If you are going to be running OpenFOAM, and therefore not paying for each parallel process, having more cores can be more useful than having fast cores. It is even more useful to have as many channels of fast RAM as you can get. Look at www.elnexus.com , where you can configure and price a white box system on-line. You will be able to put together a dual socket Opteron for just about the upper / over limit of your budget. Now, as is well known by now, Opteron cores are not winning any prizes for their speed, but you can get a lot of them for your money, and more importantly, you can get the memory bandwidth of 8 memory channels, provided of course that you configure it with 8 memory modules.


I think you will find that many of us are still using machines that are 3 to 4 years old, but it is false economy (also known as cash-flow reality). Whatever you buy, you will be able to get one with the double the performance at the sime price no more than 2 years later.
CapSizer is offline   Reply With Quote

Old   October 17, 2012, 15:58
Default
  #3
Senior Member
 
Join Date: Oct 2009
Location: Germany
Posts: 636
Rep Power: 22
abdul099 is on a distinguished road
I recommend to spend some thoughts on the memory. Try to get an educated guess how much memory you need, or you won't be happy with your purchase. Slow CPU's or having just a few cores mean, you wait longer for results. Too little memory means, you're not able to get any result. 8GB is not that much, when you run stationary cases you can easily have a bigger mesh which needs more memory. But that's all up to your requirements.

I can't give you any advice on memory, for two simple reasons: I'm not familiar with OpenFOAM and I'm not familiar with your case. But maybe somebody else can tell you better how much memory you need.
__________________
We do three types of jobs here:
GOOD, FAST AND CHEAP
You may choose any two!
abdul099 is offline   Reply With Quote

Old   October 17, 2012, 17:56
Default
  #4
Senior Member
 
scipy's Avatar
 
Alex Pasic
Join Date: Aug 2011
Location: Croatia
Posts: 199
Rep Power: 16
scipy is on a distinguished road
Send a message via Skype™ to scipy
This is what you need: http://i.imgur.com/vJXt8.png

Case/power supply/even the WD magnetic HDD are up for discussion, well even the motherboard could be had for ~100 $ less (MSI X79 8D version).. but this is MUCH more PC for the same money as a "built workstation".

You can have pretty much any computer store (where you are buying the components) build this for you free of charge. However, I wouldn't let them install anything. :P

Abdul is right, more memory is crucial and it doesn't get much better than CL9 2133 MHz Quad channel 64 GB kit. It is way out of the ideal 2-4 GB per core ratio, but if you will be doing steady state cases you won't mind waiting for a few hours or even a day longer to solve, as long as the case (mesh) can fit inside the RAM.
scipy is offline   Reply With Quote

Old   October 17, 2012, 18:06
Default
  #5
RSE
New Member
 
Reuven Shenkar
Join Date: Oct 2012
Posts: 3
Rep Power: 14
RSE is on a distinguished road
CapSizer and abdul, thanks for your insight. I definitely have to do some homework before I spend the money.

I spoke to a friend of mine who is doing exactly the same type of analysis as I am planning to do (OpenFOAM as well). He mentioned that 10M+ nodes is considered "nice". What would you suggest considering that?
RSE is offline   Reply With Quote

Old   October 17, 2012, 18:09
Default
  #6
Senior Member
 
scipy's Avatar
 
Alex Pasic
Join Date: Aug 2011
Location: Croatia
Posts: 199
Rep Power: 16
scipy is on a distinguished road
Send a message via Skype™ to scipy
Don't know about 10+ mil nodes, but 10.2 mil cells is using 22 GB of RAM (out of 32 available) in Fluent single precision Coupled algorithm (non-segregated).

You need at least 32 GB of RAM to be "comfortable", but if at any point you have a need to switch your big case to double precision.. you'll find yourself crying for more, hence - 64 GB right away.
scipy is offline   Reply With Quote

Old   October 17, 2012, 18:15
Default
  #7
RSE
New Member
 
Reuven Shenkar
Join Date: Oct 2012
Posts: 3
Rep Power: 14
RSE is on a distinguished road
scipy, thanks for your input. Definitely seems the way to go. Local computer store - here I come!
RSE is offline   Reply With Quote

Reply


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Question Re Engineering Data Source imnull ANSYS 0 March 5, 2012 14:51
Workstation setup - advice sought before making the purchase Micael FLUENT 0 January 6, 2012 15:45
internal field question - PitzDaily Case atareen64 OpenFOAM Running, Solving & CFD 2 January 26, 2011 16:26
Question about Table applicaiton. universez OpenFOAM Running, Solving & CFD 0 January 12, 2010 21:31
Poisson Solver question Suresh Main CFD Forum 3 August 12, 2005 05:37


All times are GMT -4. The time now is 04:14.