We have a few Simulation Premium users and they are asking for a faster PC to run their sims. What does the real world recommend... not those lame requirements that SWX puts out.
We have a few Simulation Premium users and they are asking for a faster PC to run their sims. What does the real world recommend... not those lame requirements that SWX puts out.
what problem do they have specifically?
the BEST fix for slow solve time is proper setup of an analysis. for example we had a customer that had an analysis that took days. with proper setup we got it down to less than an hour.
for me, i have a quad core 2.6ghz
ssd
24gb of ram
quadro 2000M
it can solve most problems in reasonable times. about twice a year i need something more powerful.
Just slower than desired solve times. The majority of their runs can be done in a few hours, but some do take more (over night). At one point they too were having extremely long solve times, and failures. We got with our VAR and a lot of that was resolved with proper setup.
We have a pretty good sized HPCluster, however Simulation will not work with it.
We use Xeon across the board in all engineering workstations, but going double processor seems to be a good move. We likely will move up a notch or two on the processor speed too- currently at 3.40. The current machine has 24GB RAM and it has run out... so increasing that is a must.
Moving to SSD also seems like a good move... maybe even the PCIe version.
FEA uses practically no graphics, so I'm not sure we would opt for the top end video. We currently use the Quadro K2000.
I have a new workstation and I did this test:
I am on page 13 if you sort Workstation 2015 and the CPU speed ascending order.
| Solidworks | : | 2015:2.1: x64 |
| OS | : | Microsoft Windows 7 Professional:Service Pack 1:x64 |
| Computer | : | Hewlett-Packard:HP Z640 Workstation |
| CPU | : | Intel(R) Xeon(R) CPU E5-2637 v3 @ 3.50GHz [Intel64 Family 6 Model 63 Stepping 2] (8 processors) |
| RealView | : | 9.1 |
| Graphics | : | NVIDIA Quadro K4200 |
| Ram | : | 32675 |
| Hard Disk | : | 1000:SCSI |
SolidWorks Preformance Test (sec) (overall= cpu + Graphics + I/O)
Overall=68.6 Cpu= 36.2 Graphics=8.7 I/O=23.7
Render=8.5
Right now I am using Simulation, it will take probably 3 hours to solve so what it really matters in terms of solving time is the number of DOFs:
If you have a smart mesh by using mesh controls where it really matters you can reduce the number of DOFs and the solving time by a lot.
Edit: I just realized my time step is 0.01 which means at least 100 steps until the final time. The time steps also are very important. If you use "Auto" setting in the time steps which usually is 1/10th of the final solution time it will take just 10 steps to solve but sometime it may take longer for each step if your system is not converging like in the problems with contacts and buckling situations. So it may happen that a smaller time step to converge faster to the final solution than the "Auto" setting.
I have one disagreement with Andrei Popov, in my (limited) experience, when using contact sets or fasteners/connections the solve time goes way up. I frequently solve 2-10M DOF linear static problems and when using multiple load cases all running at the same time my workstation frequently hits >75% CPU utilization and >50% of 128GB RAM. One I recently ran took 100 hours per load case.
And I agree with you, in my current project I have 8 contact sets with 20 contact surfaces each which means at each time step I have 25 points per surface x same number of points or double per target which is approximately 4M points contact conditions. The issue is that contact conditions work better if the mesh is refined in the contact points and the time interval is refined too. The first run I had to change the contact conditions because at the first step it was hanging too much, in my experience if it takes more than 10 min to solve the first time step it means the problem is badly conditioned, so I changed the contact conditions, I added some mesh controls and changed the time step to 0.05. I run only one case so that is how much it took: 400 k DOF, 4M points contact conditions, minimum 20 time steps expected, in reality it took 40 time steps and 4 hours to solve 80% of max time step. So it depends on the problem you solve but in my case it took in average 6 min per time step.
Dual Xeon processors, SDD HD - 500GB min (2x - one for data), 132MB DDR4 RAM, top-of-the-line graphics card(s) w/ 16MB DDR5; in short, go to HP and max out a Z840 ($18k +) and add two 1900x1600 30" monitors.
(I dream in binary.)