38 Replies Latest reply on Dec 18, 2014 4:21 AM by Hayden Krause

    flow simulation cpu utilization

    Dan Hofstetter

      Hi all,

       

      I have been playing around with Flow Simulation on my new work PC, and have noticed that the solver does not utilize much of the resources I seem to have available.  I'm running an Intel 3960X six-core processor with 32GB of memory with Windows 7 x64 and SW2011/SP5.0, and during the solving process I only see a maximum of 20% cpu utilization.  Memory usage hovers around 3.5 to 4.0 GB (the model is not very large so this is probably not an issue).  From watching the resource monitor, it doesn't seem like there is any large amount of hard disk activity going on either.  I've tried changing the priority for SW to high with no luck.

       

      I have turned off hyperthreading, and was told it would actually slow down the calculations.  Are there any settings I should change that would let Flow Sim use more of the PC?  I would have expected the flow solver to eat up as much cpu power as was available, which is why I went with a fast six-core processor in the first place.

       

      Thanks,

       

      Dan

        • Re: flow simulation cpu utilization
          Bill McEachern

          I just sent off a job on my machine which is a dual socket 6 core zeon 3.3 GHz (12 cores all in). I told Flow to use 6 cores and the cpu utilization in task manager shows peaks in the low 40's% with drops to ~18% - average looks like about 35%. The first iteration is pretty low utilization - down around 8% for me.

          • Re: flow simulation cpu utilization
            Samuel Barnum

            I would guess that flow simulation is a mostly single threaded application.  The % CPU shown in Task manager is not that easy to interpret since a single thread is not just assigned to one core, but is switched back and forth between cores.  Note that 20% is quite close to 16.6667%, which is equal to 100% CPU usage / 6 Cores.  Some of the minor associated tasks probably run in parallel, and account for the remaining ~4%.  The regular SW Simulation behaves the same way.   If you turn Hyperthreading back on and retest, you will probably see Task Manager show about 8%-11%, since Windows now thinks you have 12 cores.  That doesn't mean the simulation runs slower, windows just thinks the computer can calculate more at once.

              • Re: flow simulation cpu utilization
                Bill McEachern

                The applicatiojn has a multi threaded solver though scaling effeciency is not perfect. Also the scaling eff. is also dependent on the physics of the problem.

                  • Re: flow simulation cpu utilization
                    Samuel Barnum

                    Bill, do you know of a any resources that provide information on under what circumstances the multithreaded capabilities of Simultion are maximized, and when they are not?   One could make some educated guesses about this, but it would be more informative to have something directly from those in the know.

                      • Re: flow simulation cpu utilization
                        Bill McEachern

                        I don't know of anything formal. It is just my expereince in usinghte code. I would suggest you run your own experiments and see what you get for time to complete for avarious scenarios. You were not given bad advice on the hardware selection though. I went bigger and badder and no complaints, though I am thinking of spending some dough on the disk subsystem  - my single SAS drive is not really doing it for me.

                    • Re: flow simulation cpu utilization
                      Dan Hofstetter

                      The 20% I see for cpu usage in task manager is 20% total cpu utilization, taking into account that there are six active physical cores.  If I look at each core individually, they are working at less than 4% most of the time.  There must be another bottleneck somewhere that I can't see since the cpu isn't working very hard.  I tried watching the performance monitor too, which is where I noticed not a whole lot of activity was taking place on the disk or in memory.  With 32GB of ram, I could  enable a large RAMdisk if the hard disk were the bottleneck, even though I am using a 120GB SSD as the boot/program drive.  I have virtual memory turned off (no paging to disk since so much memory is available).  Applications such as Photoworks360 seem to do a much better job of utilizing available system resources.

                       

                      Also, I turned off hyperthreading (which would show up as 12 total cores in task manager) because I was told it would actually increase the solution time slightly over just using the 6 logical cores.  I haven't tested this yet.

                        • Re: flow simulation cpu utilization
                          Bill McEachern

                          What physics do you have enabled in your simulation that is showing poor multi-core perforamnce?

                            • Re: flow simulation cpu utilization
                              Dan Hofstetter

                              I just threw some Parker hydraulic fittings together quickly as a test to see how well the simulation performs on this PC, and to learn more about using Flow Simulation.  I set up the boundary conditions so that I have static pressure at the outlet, and inlet volume flow of 10 gpm.  I was looking for the pressure drop through the fittings at that flow rate, so I set goals for average static, total, and dynamic pressure.  I used water as the fluid.  (need to add hydraulic oil - is there a library I can import?)  The first time I ran the simulation, I chose the default value of 3 for initial mesh level, but bumped it up to 8 to force the simulation to run longer.  This was mainly so I could hibernate/wake the PC a few times to see if it affected the simulation.

                        • Re: flow simulation cpu utilization
                          Anna Wood

                          SW2012 utilizes the extra cores in flow simulation much better then previous versions.  After four cores in previous versions the speed improvement flattens out.  SW2012 improves that curve.

                           

                          I posted an image of the graph shown at the SolidWorks press event for the flow improvement in another thread.  I will see if I can find it.

                           

                          Cheers,

                           

                          Anna

                          • Re: flow simulation cpu utilization
                            Peter Griffin

                            I have a very similar set up to yours, Dan (6 core xeon, which i usually run hyperthreaded) and have recently had exactly the same question-- CPU usage isn't very high (with the smaller mesh size models anyway). I am also still running Flow Sim 2011, because some of the other CFD folks I've talked to have encountered some bugs in 2012 and ended up rolling back. I've been talking with our VAR and they gave me the same graph that Anna showed, which basically shows Flow Sim 2011 doesn't get much faster with more than 6 cores but Flow Sim 2012 seems to leverage up to around 12 cores with performance gains. The only time I really see the CPU stressing is when I'm running larger simulations (say larger than a couple million cells).

                             

                            Today I ran a series of simulations to see how time to solve varied with number of cores. I'll post my chart tomorrow. The big takeaway (atleast for this model) was that solver time was fastest when using only two cores or 8 cores. Worst time to solve was using a single core, and second worst (by a significant amount) was 10 cores.

                             

                            In talking with a coworker about the low CPU usage he speculated that it may have something to do with RAM speed. The CPU could crunch faster than the RAM can feed it? I'll honestly admit that I don't know much about the interaction between RAM and the CPU, but it sounded like an interesting thought. I'm already running 3 channel RAM though, and the next step would be quad channel, which would require (i think) a big hardware overhaul.

                             

                            I'm a little surprised SW hasn't already answered this question. At the very least, it seems like the programmers should be able to quickly say where the biggest Flow Sim performance boosts come from.

                              • Re: flow simulation cpu utilization
                                Ron Reiners

                                We I first purchased Flow I was using a Core 2 processor and would see that the CPU was running at or near 100% all of the time. When I finally was approved to get a new computer I asked my VAR and hear as to what the best system would be. My VAR said get as many cores and processors as possible. I ended up getting a system with dual 8 core Xeon processors. Expecting a significant improvement, I was really surprised to find out that although there was a calculation time improvement, it wasn't near what i thought it would be considering I change from a 2 core system to a 16 core system. I did notice, as discussed here, the processor utilization went down to about 20%.

                                 

                                I don't know if this is correct or not but after asking VAR, forum, and tech support I finally talked to someone that has more in depth knowledge of how the program work. He indicated that Flow calculates in more of a serial method in which it needs to calculate and get the results of one cell before it can calculate the next cell. because of this one cell at a time method your system will have 1-2 cells working and the others doing minimal processing which averages out to 20%. In his opinion the best system would one that has the faster clock speed.

                                 

                                I don't know if anyone else has notice this but in my particular cases I typically have 4 different configurations in my model. Using my old Core 2 system I could only simulate one at a time. Using my new dual Xeon processor system I am able to run 4 simulations at the same time in effect cutting my total calculation time down to 25%. Also when I am running 4 simulations at one my CPU utilization is now somewhere around 70%-80%.

                                  • Re: flow simulation cpu utilization
                                    Dan Hofstetter

                                    Great replies - glad to know I'm not necessarily doing something wrong that is causing the "slow" performance.  I have 32GB of quad-channel RAM, currently clocked at 1600 MHz.  I'll try to clock it faster and see if the speed improves.  Watching the performance monitor in Windows 7 shows hard disk and memory writes, and I really don't see a whole lot of anything going on that is causing any part of my system to work very hard during calculations.  It is almost like there is a delay programmed into the routines.  It seems to me that even if Flow solves mostly serially that at least one processor would be cranked up during solving.

                                     

                                    I was also told to get as many cores as possible for simulation.  I could have saved a lot of money by getting an i7-2600 quad core with a less-expensive motherboard - that seems like the best bang for the buck at the moment.  I have also wondered about SW2012.  Does anyone know if any of the other SW-compatible programs solve faster?  I know price differs quite a bit, but I want to make sure I am working as efficiently as possible.

                                     

                                    Edit:  I just watched it again after speeding up the memory to 1866 MHz and noticed a lot of disk writes to the simulation folder.  I'm going to try moving the files around and maybe even try a RamDisk and see what happens.

                                      • Re: flow simulation cpu utilization
                                        Peter Griffin

                                        Here are a couple charts I compiled based on solver times for the simulation I described earlier. I hope the text is big enough to read. What I did was I ran the same simulation (same boundary conditions, mesh, etc...) on two different computers with different processors. I defined scalability as (time to run with 1 core/time to run with X cores). I can post the model and conditions if anyone else is curious about how their set up compares.

                                         

                                        Looking at the charts, it seems like for a given core count processor speed really dominates. The i7 and Xeon I'm running are the same generation but one has twice the clock speed of the other. This translated to roughly twice the solve speed when using 2-8 cores. (By the way, I'm hyperthreading in all these simulation runs, which I now hear is NOT recommended).

                                         

                                        Flow Simulation Run Time Comparison Charts.JPG

                                          • Re: flow simulation cpu utilization
                                            Dan Hofstetter

                                            I'd be curious to try that model using my rig.  I just moved all temp files and my work files to a 4GB RAMDisk (created using the DataRAM RAMDisk utility, which is free for 4GB and smaller disks), and my simulation seems to run faster now, especially when doing initial meshing.  I benchmarked the RAMDisk at over 4,000 MB/sec, versus 550 MB/sec for my SSD.  I am seeing between 35% and 50% overall CPU usage with that model now, compared to the ~20% I saw before.  I am running this processor at 3.6 GHz right now, but can crank it back up to 4.5 GHz and see how it compares at a higher core speed, too.

                                              • Re: flow simulation cpu utilization
                                                Peter Griffin

                                                Dan, do you have a chart with some of the speed comparisons you were describing here?

                                                  • Re: flow simulation cpu utilization
                                                    Dan Hofstetter

                                                    I didn't chart anything - I'll try to re-run the simulation that you posted below with various setups and record some solving times.

                                                      • Re: flow simulation cpu utilization
                                                        Mark Larson

                                                        If you have not read the following thread, you should:

                                                        see Ian Hogg post

                                                        https://forum.solidworks.com/message/181109#181109

                                                        note what he has to say about time to solve from one release to the next, and time to solve as you increase the number of cores.

                                                         

                                                        IMO, scalability is sales & marketing mumbo jumbo. It obscures what is really important by confusion over what it really means because it allows changing the scale on graphs, and what does scalability really mean? Try that with a graph of time, everybody knows what a second and a minute is.

                                                         

                                                        I have only compared a single CPU vs. a dual cpu, running 2 tests, and found the reduction to be about of 30% (26% & 35% actually in the two tests I ran) in the amount of time to solve, which is in line with what Ian reports in the link above. CPU utilization? My observation was 60-90% (although I actually saw it vary from 16-100%) was my seat of pants for a single cpu and 40-50% for the dual.That may tell you something but I really don't care, anything other than time to solve is just fodder for musing IMO. Ultimately, if your hardware tweaks and changes don't reduce time to solve, it was just an exercise in futility.

                                                         

                                                        BTW, if you want to reduce time by half, forget your hardware, just use symmetry in the boundary condition of the computational domain. (OK, truth in advertising, it won't reduce by half any more than doubling the number of cores will, but it will reduce significantly)

                                                        IOW, your biggest bang for the buck may be between your ears (knowledge of the software).

                                                          • Re: flow simulation cpu utilization
                                                            Peter Griffin

                                                            I agree, at the end of the day all I want is for the sims to run faster. I've tried doing a lot of things within the software to optimize mesh size, symmetry, CFD Zooming, etc... But most of the stuff I do can't be simplified with things like symmetry or CFD Zooming.

                                                             

                                                            CPU utilization itself doesn't mean much. It's just an indication that Flow isn't making good use of the resources it has available, so if you're going to build a new machine you don't waste money putting in resources it can't use.

                                                             

                                                            Scalability is the ratio of computational time normalized to the number of cells. SW does this is because the real geometry number of cells in 2011 and 2012 is different for the same task. SW computes scalability (time to solve)/(number of cells). So if you keep the number of cells constant, you can set up a ratio where cell count becomes irrelevant and you get purely a time reduction percentage. I agree it would be nice if they would just publish that instead of scalability.

                                                              • Re: flow simulation cpu utilization
                                                                Mark Larson

                                                                What I meant  can be seen in looking at the graphs you posted, from the graph of time it can be seen that 6 cores from one core almost cuts the time in half, and that going to 8 cores give almost no gain. It also shows there really isn’t that much reduction in time going from 4 cores to 6. It is easy for the brain to process the time vs. core graph because it is easy to translate the scale of the graph because we all know what a minute & second is.

                                                                 

                                                                Now look at the scalability graph, the performance gains look a little more impressive, and could have been made to look even more impressive by changing the scale of the graph on the Y  axis, and it would not be immediately obvious since the concept of scalability is nowhere near as familiar to us as time. From this graph it actually looks like going from 4 cores to 6 is significant, but it really isn’t because it is only about 30 seconds. Going from 6 cores to 8 looks like it gives a little performance boost but looking at the time saved, it is so small to be imperceptible.

                                                                 

                                                                As for cpu utilization, If it were me, I’d be more concerned if it is running at 90-100% utilized because that would mean my computer is the bottleneck. If it is running low, it means the computer is more than adequate for the job. The computer is the only thing I have control of, I can improve the hardware or fiddle with Windows to make it faster. But if it is more than adequate for the job (cpu utilization low), there is nothing further I can do to lower the solve times.

                                                      • Re: flow simulation cpu utilization
                                                        Peter Griffin

                                                        I just ran another set of sims to compare Flow 2011 to 2012 (un-hyperthreaded for both). Also, I compared the hyperthreaded vs. un-hyperthreaded for Flow 2011. Here are the results.

                                                         

                                                        Flow 2012 actually took about 10 more iterations to solve than 2011, at about 2 seconds per iteration.  tweaks. Flow Simulation - 2011 vs. 2012 Solve Time.JPGFlow Simulation 2011 - Hyperthreaded vs. Un-Hyperthreaded.JPG

                                                      • Re: flow simulation cpu utilization
                                                        Anna Wood

                                                        SSD drives will help your performance.  Simulation reads/writes to disk a lot.  So fast SSD drives in a RAID 0 configuration will get you a good performance boost.

                                                         

                                                        Cheers,

                                                         

                                                        Anna

                                                      • Re: flow simulation cpu utilization
                                                        Peter Griffin

                                                        I had an email in my inbox from SW (via my VAR) this morning saying basically the same the you mentioned, Ron, about some Flow calculations being serial in nature. What it said was the Navier-Stokes equations (which are the fundamental eq's Flow is trying to solve for fluid flow mass, momentum, and energy conservation) have no closed form solutions, meaning they have to be solved one after the other and can't be run in parallel. There is still some benefit to running multicores (perhaps because some of the other eq's can be solved simulataneously?) and the results seem to be more pronounced with larger mesh sizes. However, meshing and pre-processing themseleves are single core activities.

                                                         

                                                        I'll post the sim conditions in another reply.

                                                          • Re: flow simulation cpu utilization
                                                            Peter Griffin

                                                            Here's how to set up the simulation I ran.

                                                             

                                                            The model I used was the Heat Exchanger example in the Flow Simulation folder. File path will be something like:

                                                             

                                                            C:\Program Files\SolidWorks Corp\SolidWorks Flow Simulation\Examples\Tutorial 3 - Heat Exchanger

                                                             

                                                            Open that assembly file and set up the following:

                                                             

                                                            General Settings:

                                                                 Analysis Type: Internal

                                                                 Check box for "Heat Conduction in solids"

                                                                 Default Fluid: Air

                                                                 Default Solid: 321 Stainless Steel

                                                                 Roughness: 15 micrometer

                                                                 Other than that, just leave everything else default

                                                             

                                                            Fluid Subdomains: set two subdomains, both air, and leave them at their default conditions.

                                                             

                                                            Boundary Conditions:

                                                                 Tube Side Inlet-- 2000 l/min @ 20.05 C (all other options default)

                                                                 Tube Side Outlet -- Environmental Pressure (all other options default)

                                                                 Shell Side Inlet -- 3000 l/min @ 400 C (all other options default)

                                                                 Shell Side Outlet -- Static Pressure @ 20 psig (all other options default)

                                                             

                                                            Initial Mesh: See the screen shot. This should give you 48720 fluid cells, 13150 solid cells, and 32508 partial cells.

                                                             

                                                            Surface Goals: (Be sure to select "Create Separate Goals for Each Surface" if you try to add all the goals at once)

                                                                 Tube Side Outlet: Bulk Avg Fluid Temp and Bulk Avg Total Pressure (Use for Convergance check for both)

                                                                 Shell Side Outlet: Bulk Avg Fluid Temp and Bulk Avg Total Pressure (Use for Convergance check for both)

                                                             

                                                            I can just send the zip file of the model already set up in Flow for anyone who wants it via email. Just provide your email.Model Screen Shot.JPGInitial Mesh Screen Shot.JPG

                                                              • Re: flow simulation cpu utilization
                                                                Dan Hofstetter

                                                                Can you email the zipped file you mentioned?  I get a warning message about a vortex, and failure to complete mid-way through the calculations.  I may have something set wrong.  My email should be available in my profile (don't want to post here to avoid spam).

                                                                  • Re: flow simulation cpu utilization
                                                                    Peter Griffin

                                                                    I couldn't find your email in your profile. Perhaps you can find mine and just send me something?

                                                                      • Re: flow simulation cpu utilization
                                                                        Dan Hofstetter

                                                                        heh, privacy settings must be correct then - I assumed since I saw it that it was public.  Please try sending it to detroit54@comcast.net - that one already gets a lot of spam so I don't much care if it gets more.

                                                                          • Re: flow simulation cpu utilization
                                                                            Peter Griffin

                                                                            Were you able to get the file that I sent to you to run ok?

                                                                              • Re: flow simulation cpu utilization
                                                                                Dan Hofstetter

                                                                                I got it to run successfully, thanks!  I need to find some time to sit and try some different hardware/software configurations so I can report back with the changes to solving speed.  Will post here when I have a chart created.

                                                                                • Re: flow simulation cpu utilization
                                                                                  Dan Hofstetter

                                                                                  Ok, I spent some time running the simulation, here is a chart for running at full speed (4.4 GHz, 1866 MHz RAM) using a solid-state drive (SSD) for the work and cache files, with Hyperthreading turned on:

                                                                                   

                                                                                  SWSimTime.jpg

                                                                                  I also tried stock cpu and memory frequency settings separately to see the impact of faster cpu and ram.  As you noted, 8 cores seems to yield the best time to solve in all scenarios that I ran.  Here is a summary for 8-core performance using a solid-state drive as the work and cache drive:

                                                                                   

                                                                                  4.4 GHz cpu, 1866 MHz ram, SSD, HT on          3:03 to solve          30-40% cpu utilization

                                                                                  4.4 GHz cpu, 1600 MHz ram, SSD, HT on          3:09 to solve          35-40% cpu utilization

                                                                                  3.6 GHz cpu, 1866 MHz ram, SSD, HT on          3:39 to solve          30-40% cpu utilization

                                                                                   

                                                                                  One run using a RamDisk as the work and cache drive:

                                                                                   

                                                                                  4.4 GHz cpu, 1866 MHz ram, RamDisk, HT on     3:03 to solve          30-40% cpu utilization

                                                                                   

                                                                                  And one instance where I turned Hyperthreading off (only six cores running):

                                                                                   

                                                                                  4.4 GHz cpu, 1866 MHz ram, SSD, HT off          2:57 to solve          50-60% cpu utilization

                                                                                   

                                                                                  I was a bit surprised that the time to solve using the RamDisk was no better than my SSD, but then again solid-state drives are fast.  Watching the performance monitor in Win7, there is a lot of disk I/O going on early in the simulation, mostly during meshing, so I expected to see bigger gains with the RamDisk.  Still, with a larger model maybe the results would be different?  Based on this, I'm likely to keep Hyperthreading turned on all the time since the time to solve isn't that different.  Maybe if I get into simulations that take hours or days to solve I will change my mind.  Another thing worth noting is that as cpu utilization goes up, time to solve goes down in all scenarios.  This is why I started this thread in the first place - it seems like if you have a 500-horsepower engine, but you are only giving it half-throttle, you are going to get to the finish slower than if you mash the pedal to the floor.  It is worth noting that my cpu went to max frequency right away when running all scenarios, i.e. it went from 1.2 GHz idle speed to 4.4 GHz as soon as the simulation started, and it stayed at that speed until the calculations had converged.

                                                                                   

                                                                                  I noticed you had different configurations in the model - 12 core, 8 core, etc.  I was running the default "3 core" configuration in all instances - should I have changed this, or was this simply a way of not having to change the number of cores used for each run?

                                                                                   

                                                                                  - Dan

                                                                                    • Re: flow simulation cpu utilization
                                                                                      Samuel Barnum

                                                                                      Dan, you might like to try TMonitor to look at the Freq. of each core as you are running.  Unless you have some other utility that does the same already.

                                                                                       

                                                                                      It can show you per-core clock speed or multiplier, and can log data per second if you like.. Yellow inicates steps into Turbo Boost.

                                                                                       

                                                                                      http://www.cpuid.com/softwares/tmonitor.html

                                                                                      • Re: flow simulation cpu utilization
                                                                                        Peter Griffin

                                                                                        Interesting results. Did you build your computer yourself or did you buy from someone?

                                                                                         

                                                                                        I had multiple configurations simply because I cloned the project each time I did a new run. This way it forced the calculation to start completely from scratch. I've had problems in the past with Flow not recognizing it needs to remesh and/or pulling old results into a new study.

                                                                                          • Re: flow simulation cpu utilization
                                                                                            Dan Hofstetter

                                                                                            I built the PC myself with Flow Simulation in mind, it is largely based on the November Ultra-high Performance suggested computer posted here a while back.  I'm very happy with its performance so far.  (FWIW Scooby Doo rebuild time is 9.76 seconds).  Clocking this at 4.4 GHz was as easy as raising the turbo multiplier in bios from 36 to 44.  I priced Dell and Boxx workstations before building this one, but they were just too expensive and did not have exactly what I wanted.  I would have bought the Intel 3930K processor for $600 less than the 3960X, since it can be clocked to the same speeds and would perform just as well, but it was sold out everywhere in the US a few days before I needed to place the order for components. 

                                                                                             

                                                                                            I included the parts list below for anyone that might be interested:

                                                                                             

                                                                                            Amazon.com:

                                                                                            NVIDIA Quadro FX 4600 Graphics Card (Condition: New)                                  $199.99

                                                                                            Newegg.com:

                                                                                            CORSAIR H80 (CWCH80) High Performance Liquid CPU Cooler

                                                                                            Item #: N82E16835181016                                                                                        $93.24

                                                                                            Antec Sonata Elite Black 0.8mm cold rolled steel ATX Mid Tower Computer Case

                                                                                            Item #: N82E16811129057                                                                                        $89.99

                                                                                            CORSAIR CMPSU-750TX 750W ATX12V v2.2 SLI Certified CrossFire Ready 80 PLUS Certified Active PFC Compatible with Core i7 ...

                                                                                            Item #: N82E16817139006                                                                                        $129.99
                                                                                            TWO G.SKILL Ripjaws Z Series 16GB (4 x 4GB) 240-Pin DDR3 SDRAM DDR3 2133 (PC3 17000) Desktop Memory Model F3-17000CL11Q-16GBZL

                                                                                            Item #: N82E16820231501                                                                                        $279.98
                                                                                            ($139.99 ea - $35.00 ea on sale)

                                                                                            MSI X79A-GD65 (8D) LGA 2011 Intel X79 SATA 6Gb/s USB 3.0 ATX Intel Motherboard with UEFI BIOS

                                                                                            Item #: N82E16813130624                                                                                        $299.99

                                                                                            Intel Core i7-3960X Extreme Edition Sandy Bridge-E 3.3GHz (3.9GHz Turbo) LGA 2011 130W Six-Core Desktop Processor BX80619i73960X

                                                                                            Item #: N82E16819116491                                                                                        $1,049.99

                                                                                            Seagate Barracuda ST1000DM003 1TB 7200 RPM SATA 6.0Gb/s 3.5" Internal Hard Drive -Bare Drive

                                                                                            Item #: N82E16822148840                                                                                        $129.99

                                                                                            RAZER ABYSSUS Black 3 Buttons 1 x Wheel USB Wired Optical High Precision Gaming Mouse

                                                                                            Item #: N82E16826153056                                                                                        $29.99

                                                                                            LITE-ON DVD Burner - Bulk Black SATA Model iHAS124-04 - OEM

                                                                                            Item #: N82E16827106289                                                                                        $19.99

                                                                                            OCZ Vertex 3 VTX3-25SAT3-120G 2.5" 120GB SATA III MLC Internal Solid State Drive (SSD)

                                                                                            Item #: N82E16820227706                                                                                        $189.99

                                                                                            Total Cost (not including shipping):                                                                           $2,513.13

                                                                            • Re: flow simulation cpu utilization
                                                                              Hayden Krause

                                                                              Hi Guys,

                                                                               

                                                                              I wondered if the CPU utilization from 2012 to 2014/15 has increased the scalability of your duel processor Xeon processors.

                                                                               

                                                                              I run simulations but also need to do CAD work while simulations are running.  So far, in order to do this i have two computers, one is a dedicated simulation machine (i7-5960X OC to 4.4GHz).  The processor is lightning fast but there isn't a chance that the PC is usable while simulation is running.  I am thinking of moving to Duel Xeon or Quad server system so that I can run simulations and do CAD happily.  It would only be worth doing if i could get similar performance as I am currently having with my current setup.

                                                                               

                                                                              Any thoughts or experience with this would be appreciated.

                                                                                • Re: flow simulation cpu utilization
                                                                                  Jared Conway

                                                                                  i run 2-3 sessions of solidworks at the same time if i need to run multiple solutions at the same time and can't say that i have stability problems

                                                                                  performance is what i expect, it varies because my resources are being tapped to run simulations

                                                                                  i haven't tried it yet but in 2015 you can pre and post process at the same time

                                                                                  how did you convince management to get you a second swx license to run the simulation on one and do cad on the other

                                                                                   

                                                                                  in the end, the best way to reduce solve time is to optimize the analysis setup and process. throwing hardware at it has limited returns

                                                                                  1 person found this helpful
                                                                                    • Re: flow simulation cpu utilization
                                                                                      Hayden Krause

                                                                                      Cheers Jared,

                                                                                       

                                                                                      The reason for having a dedicated machine for simulation is because we run large simulations 24/7.  We optimize the machine to have maximum throughput and run the machine to its full capacity (currently using 8 cores).  We are designing instruments where accuracy is key, therefore simulation size are very large.  We are in a very niche market - One simulation for our next product in line will take 5 days to accurately solve.  We always optimize the analysis, simplify geometry were we can and spend months on mesh refinement - i would be shot if i didn't (by myself). 

                                                                                       

                                                                                      The sim machines running at full tilt becomes impossible to use productively for other areas of the job.  I still need to design parts and systems, oversee production, I also need to run FEA... So to do all this effectively (with SW) I need the separate system.  Bottom line is that in doing what we are doing we have been managing to create more accurate and robust instruments then our competitors, and we are doing it faster then before.  Therefore a dedicated system has proved its worth - and it isn't unusual for companies to have dedicated systems.

                                                                                       

                                                                                      Having said that, you see me on here allot as I am forever researching and trying to understand the options.  We have not gone to Xeon purely because we think we have the fastest setup possible at the moment, which is our aim.  We have recently taken on new staff.  Rather then purchasing new SW, I would prefer to purchase a Duel CPU Xeon so i can run sims on 8 (perhaps more depending) of the cores and still have the CPU power available to run CAD, FEA etc.  This relinquishes PC and SW to the new staff and all we need is one Xeon PC.  The catch is, if the simulation runs slower, it will slow our dev time.  So the Xeon needs to be able to run as fast as my OC 8 core i7 (which is the fastest processor i have used), at the same time as allowing me to CAD, FEA and all the rest.  This is why I ask about running multiple simulations in one go, doing this increases your throughput so can accept slower individual simulations.

                                                                                       

                                                                                      On top of staff acquired, we are soon to be acquiring more.  Moving forward I want to be absolutely sure that we are at our most efficient.  I take it that with your system you are able to run simulations and do CAD at the same time without any noticeable effects?

                                                                                       

                                                                                      Hayden