[flotus1]

Ooof...Abaqus 6.14 is ancient. Have you ever considered switching to a more recent version? Windows 10 is not among the tested operating systems. And based on my limited experience, that really means something with this software.

Anyway, the error message
dbghelp.dll not found in PATH2/19/2020 11:20:50 AM
suggests that there is either something wrong with your installation, or that there really is an incompatibility between your modern operating system and the ancient software version.

[Slash]

I cannot buy a newer version of Abaqus. I'm trying to install Ubuntu SO and check if it is compatible. A windows alternative is windows server but it is expensive and I don't know if Abaqus works well in it. An interesting point is that when I installed windows, the AMD chipset driver hasn't been installed well because there's an incompatibility with Windows Pro and supermicro drivers. Can be this the root cause of the problem?

[flotus1]

If you want to give it a try on Linux, use CentOS. Installing Abaqus on other distributions (like Ubuntu or OpenSUSE) requires you to jump through some additional hoops.

You could also dig a little deeper regarding the missing dll. It seems to be a rather common problem based on a quick google search, and not specific to Abaqus.

[Slash]

flotus 1, I have tried Abaqus in a virtual machine with Ubuntu and it doesn't work. The error is: Abaqus Error: The executable standard
aborted with system error "Illegal instruction" (signal 4).

Did you have tried Abaqus standard on linux?

[flotus1]

Yes, we use it on OpenSUSE with Intel Xeon Workstations. But it has it's pitfalls. Installation requires some tweaking beforehand, and newer OS versions have some incompatibilities. According to support, it is a bug in Abaqus linked to newer versions of glibc (2.25 and above). They intend to fix it with the 2021 release, and implement these fixes down to version 2018.
On our cluster with AMD Epyc Naples CPUs, it runs just fine on CentOS. That's why I said: if you want to give Linux a try, better stick to the distributions (and versions) that your Abaqus release was tested with https://www.3ds.com/support/hardware...-614-products/
You can substitute red hat with CentOS though.

In hindsight, I might not have been so keen on recommending bleeding edge hardware, knowing that you intend to run rather old software versions.

[Slash]

In the cluster with red hat have you tried abaqus standard?

[flotus1]

Yes, I did. On CentOS that is, not RHEL. Both standard and explicit work as intended. I only used Abaqus 2019 though.

[Habib-CFD]

I can confirm both standard and explicit solvers of Abaqus 2016 work properly in CentOS V6.10 (Kernal 2.6.32-754).

[Slash]

Habib-CFD you can confirm that works using AMD Epyc cpus?

[Habib-CFD]

Originally Posted by Slash

Habib-CFD you can confirm that works using AMD Epyc cpus?

The direct answer is No. The mentioned system in our laboratory stands on Intel 6580K. I have never installed the Abaqus in our EPYC 7302P system due to license limitations. However, I can confirm the CentOS version 7.7 compatible with EPYC Rome (2nd gen) servers from Hardware viewpoint.

[Slash]

Hello, I finally installed Abaqus 2019 and it worked fully with windows 10 pro. Now I have another question.

I've done several tests to maximize the performance of my workstation. Earlier tests show that my workstation runs an Abaqus analysis in 1 hour and 57 minutes. After that I've optimized bios parameters according to AMD guideline and now I run the test in 1 hour and 43 minutes. Finally I've decided to test the gpu co-computation but it seems not working well, since the analysis time (with the same bios settings) is increased to 1 hour and 48 minutes with -gpus=1 abaqus parameter.

I see many test in internet that shows a significant improvement of performance with gpu computation. What am I doing wrong? Do you have any experience with gpu computation?

[Slash]

My gpu card is the Nvidia Quadro RTX 4000 with 8 gb of gddr6 memory and 2304 cuda cores.

When I run Abaqus with -gpus=1 parameter I get some warning in the log file: the supernode XXX is too large to be loaded into the GPU. The warning is repeated 4 or 5 times with different supernode number.

[flotus1]

Today is your lucky day. I recently had to do some testing with GPU acceleration on Abaqus/Standard. Ask me anything

The RTX4000 won't help for GPU acceleration. The solver needs DP performance, which most modern Quadro cards don't have. Quadro RTX 4000 has an abysmal 222 GFLOPS for FP64. And the warning about supernode not fitting into VRAM points to 8GB being not enough for your model.
As a matter of fact, I am in the process of sourcing a few old Quadro K6000 (1.73 TFLOPS FP64, 12GB VRAM) after my tests.

[Slash]

Interesting. Where have you found these benchmarks?

[flotus1]

I used one of our real-world simulations as a test case.
Or what do you mean by "benchmark"?

[Slash]

Do you have both the quadro rtx 4000 and k6000 available for real-world testing?

[flotus1]

No, only two K6000, currently not in the same workstation.
I tested different GPUs with low FP64 performance, they actually slowed down the simulation. My attempts at running the simulation in single precision had no effect.

[Slash]

I have only set gpus=1 in the command line. Is it necessary to set other parameters in Abaqus?

[flotus1]

With only one GPU accelerator in the system, there is not much more you can do.
Some minor tweaks concerning core binding, mostly undocumented, are still possible. But my testing on those was not conclusive. Because they don't all behave as intended, and the test case I was given took hours to finish. And support had to admit that they will look into these issues for coming releases.

[Slash]

Good evening. I've a problem with the same motherboard (H11DSI with an Amd Epyc 7352). Since I assembled the workstation, the button to suspend or hibernate never appeared on windows. It is normal with this kind of motherboard or there's a BIOS option to activate the hibernation/suspension? Thanks