I have a model with 2300 elements. Yes, an unusually large model, but not unheard of I suppose. And yes, I should expect the computer to be having a hard time dealing with a model of that size.

So I open the exact same model on three different computers: my old desktop that I had been using for years, my old basic company laptop and my brand new desktop. Surprisingly, the laptop that I otherwise hate greatly outperformed the other two workstations.

Check out the attached comparison table. I can't explain the reason for the sluggishness of the workstations. I'm wondering if CAESAR II is not designed to work well with Quadro. I sure hope I'm wrong. I wanna love this beast of a workstation that I just got yesterday. And I need Quadro for Autodesk Inventor and ANSYS. Any idea what I can do to unleash the power of my new machine on CAESAR II?

Out of curiosity, did you turn on Task Manager to see how system resources are being allocated?

Unfortunately, buying computers these days is no longer like comparing cars and more like comparing business enterprises. As such, there's seemingly always going to be some weird weak link in the inner operations that makes some things super slow from one system to the next.

You might consider remote access software to operate your laptop from The Beast for an optimized experience. You may experience latency, however.

If your laptop and desktop have redundant network cable ports, you might cut that down through connecting them via ethernet cable. (But extra ethernet cards aren't expensive, assuming your hardware will support it.)

Try changing the graphics driver CAESAR II is using. You can find this setting on the "Graphics Settings" pane of the Configuration. The driver setting is the last option in the "Miscellaneous Options" group.

You didn't say what version of CAESAR II you're running, but v.10 supports two flavors of DirectX - try one of these.

I use version 10. I tried DirectX 9 and got basically the same performance as DirectX 11.

The videos that I'll post below will show that the computer is pretty much free.
Yeah I'm hoping that I don't have to remote to the laptop just to edit the model, and bring it back to the desktop for solving.

Thank you for your response all.

I just recorded myself opening and analyzing the exact same file using the three machines. All using version 2018 service pack 2.

New Workstation

Graphic Refresh Time: 21sec
Error Checking Time: 4min 11sec
Solving Time: 15min 49sec
CAESAR II Selected Graphics Driver: DirectX 11

OS: Windows 10
Processor: Xeon W-2145 3.70 GHz (8 cores, 16 logical)
Memory: 64GB, DDR4 2666 MHz
Graphics Card: NVidia Quadro P4000, 1202 MHz
Graphics Memory: 8GB GDDR5, 1901 MHz, 243.3 GB/s
Storage: SSD Samsung MZVKW512HMJP


Old Desktop

Graphic Refresh Time: 13sec
Error Checking Time: 2min 9sec
Solving Time: 18min 7sec
CAESAR II Selected Graphics Driver: DirectX 11

OS: Windows 7
Processor: i7-3770 3.40 GHz (4 cores, 8 logical)
Memory: 16GB, DDR3 1600 MHz
Graphics Card: NVidia Quadro K2000, 954 MHz
Graphics Memory: 2GB GDDR5, 1000 MHz, 64 GB/s
Storage: HDD Seagate ST250DM000


Old Laptop

Graphic Refresh Time: 5sec
Error Checking Time: 5sec
Solving Time: 28min 4sec
CAESAR II Selected Graphics Driver: DirectX 11

OS: Windows 7
Processor: i5-2520M 2.50 GHz (2 cores, 4 logical)
Memory: 8GB, DDR3 666.67 MHz
Graphics Card: Intel HD 3000, 648 MHz
Graphics Memory: shared
Storage: HDD Toshiba MK3261GSYN

Is the original solving time of 2m 11s a typo, or do you think the recording software is interfering?

I can only say that graphics hardware has always been a challenge with CAESAR.

With regards to solving time, it's clear that in Windows 10, CAESAR relies on a single core for the heavy lifting, as one sits at nearly 100%. In Windows 7, it's not clear, because all cores are being used all the time, say around 70%. Perhaps CAESAR is still using the 1, and other processes are tying up the others, too.

Yeah actually the 2-minute solving time that I posted at the beginning was based on the same file with only 3 load cases. I did that test a while back.

I agree that it seems like only one logical processor is utilized by the solver. Perhaps that hasn't changed since the old days? Would be nice if the user can decide how many processors to use.

FYI, Inventor does *not* need a Quadro or FirePro - it gets along just fine with a GTX 1___ or RTX 2____ series. Not too certain about ANSYS, but if its using DirectX instead of OpenGL it's the same thing.

Agreed. Yeah I shouldn't have said that I need Quadro for Inventor. It's just that for our previous machines, my choices were Quadro or some basic ATI card. The performance difference with Inventor and ANSYS are vastly different as expected.

And the new machines we're getting, I don't have any choice. I wish I'm involved in the decision making process on the hardware though. I'm hoping that I don't have to give up Quadro for CAESAR, if it's indeed the reason for this poor performance I'm noticing.

The thing with multicore computing is that you have to be able to split up the same computation as though it were running on different actual computers.

Analogously, the computer would have to intelligently split up your piping models and add displaced anchors at the split point(s), and solve each one independently, moving displacements around until it gets all of them converge on equal forces and displacements.

My opinion is you'd need to eliminate CAESAR's convergence issues to make this viable.

Yes, all three of my machines struggle to finish solving the analysis.

But modeling time is what I'm most concerned about at this point. 20-second refresh each time I edit any part of the model (unless of course I freeze the graphic, which is my current workaround), and 250-second error check are excessive for a powerful machine, when a much more inferior machine can do each for 5 seconds.

Just tested the same model on my colleague's desktop which is the same as my old desktop, except it uses AMD Radeon HD 7470 graphics card. The graphics refresh and error check is as fast as my laptop. Though not conclusive, I am inclined to believe that Quadro just doesn't work well with CAESAR II.

I use Quadro M1200 laptop with Xenon processor (year old) laptop. It works very well. Your machine is very powerful. Here are few things that can help,

1. Make sure your computer is actually using the dedicated video card (more performance) and not the video card in-built to the mother board (saves power). By default the computer (maybe it is a laptop only thing) is setup to decide which card to use. You can see that in the Nvdia Control panel by right clicking on your desktop and specify your preference there.

2. If you have not already tried it, update the video card drivers and then try again.

3. Are you using the same config settings for these comparisons.

4. Also try opengl2 drivers (you have already tried driectx 11, directx 9).

From what I can see, it looks like your model is very structured. Could you not model sections as individual CAESAR files and then include them into a master CAESAR file come run-time?

Yeah, that's one workaround. If I didn't know that an inferior machine can do much better, I'd be happy with that.

Mandeep,
1) Yes it is the dedicated video card being used.

2) I will try updating the driver and see what happens.

3) Yes I use the same CAESAR II config.

4) I tried them all. Very similar result.

Interesting. IT guy logged in to my new machine so I can install some applications and drivers. I updated the graphics driver as well.

Then before I logged out, I tried running CAESAR. It's real fast. I took notes on the settings of the graphic driver.

When I logged into my account, slow again. Even after copying the settings. I suppose I have more investigating to do.

It may be permissions issue. Ask IT admin to make sure you have read and write access to CAESAR II System folder from your account, e.g.

C:\ProgramData\Intergraph CAS\CAESAR II\10.0

Also check if you have anti-virus or anti-spam or some other monitoring system running. If so, try to whitelist CAESAR II.

Did you disable the in-build video card completely or did you choose CAESAR II to run with Nvdia card. You may have to specify it at exe level so specify prepip.exe to use Nvdia driver.

Just a quick update. It seems like the answer is in the registry. Simply importing the HKCU for CAESAR II from the IT personnel's account to mine solved the problem. I'll report more when I'm certain which values are the culprit.

This had been bugging me for over a year now. After lots of tinkering, I believe I've found the culprits. Not at all what I expected though. They are:
Expanded Temperature and Pressure Highlight Toolbars

In the video below I demonstrate how just popping up the temperature highlight toolbar suddenly causes the problem. I start the video with me refreshing the plot and running error check before popping up the control bar. Both completed within seconds. Right after I popped up the temperature highlight toolbar, plot refresh took 30 seconds and error check took 6min 45sec to complete.



The same would happen with Pressure Highlight toolbar. And I'm not sure about earlier CAESAR II, but I had the same problem with version 2017.

Again, for smaller models this may not be noticeable. But as I have shown you before, the difference is more obvious the bigger the model becomes. I would love to be able to use these toolbars since I typically deal with multiple temperatures and pressures. Hopefully this problem is taken care of in the next release of CAESAR II.

Wow Good investigation!

I could reproduce it to some extend, but I am not seeing the delays that you are seeing. I have sent it to developers to review this.

Thanks,
Mandeep

Thank you. Yeah I think the difference is not very noticeable for smaller models. I just opened a couple of models to see the effect of this. On a 600-element model, I saw little performance difference. On a 1300-element model, I started seeing it. And the difference is more noticeable when running error checks. Additionally, in one test the error check got even slower when both Temperature and Pressure Highlight toolbars were present.

I personally will continue using the toolbars. And when my model gets big enough that I start noticing the performance penalty, I know what to do now.