Learning Forum

[kevin1986]

Hi guys,

I installed ansys 2022 r1, 2019 r1 and 2022 r1. I use these different ansys fluent to run my test case. I found fluent 2022 r1 is much slower than others. I am quite confused since the hardware (cpu and memory) is the same. After further debugging, I found fluent 2022 r1's bandwidth (slowest is around 100 mb/s, fasters is 2000 mb/s, but the value changes every time) is much slower than others (slowest is around 1500 mb/s).  All of them I used intelmpi, with 64 cores. 

What can I do?

Thanks. 

[kevin1986]

 

I also tried 2022r2. Still slow. I tried only 4 cores. 2020’s bandwidth is around 9000-10000, 2022r2 is around 1000-2000. When more cores are used, 2022r2 is slower. Why?

 

[MangeshANSYS]

Hello,

we could compare two different Fluent versions on same machine or same Fluent version on twi different machines, but not 2 different versions on 2 different computers at the same time..

Are the screenshots from two differrent machines or two different user profiles on same machine? Is there variation seen on the same machine with 2 different versions of Fluent and MPI or are we comparing this data from 2 different machines ?

If only one machine is used for solcing then please also try microsoft MPI from lauuncher parallel settings

If still seeing issues then can you please provide processor make and model, memory quantity and configuration (4 DIMMs of 8 GB etc)

[kevin1986]

Hello Mangesh,

Thanks for your reply. We did the comparison of different fluent version on SAME machines. We do have several workstations, but we did the comparison of different fluent version on SAME machine. We found the bandwidth is different on SAME machine. We suspect it comes from intelMPI. Although microsoft MPI of Fluent 2022 is also slow. 

The processors are amd epyc 7642/7543/7302. 16 DIMM of 16G. We have several workstations with different CPUs. But it should be noted we did the test on same machine. 

It should be noted that linux ansys fluent 2022 is much faster. The calcultion speed of linux fluent 2022 is the same with windows fluent 2020. 

Thanks for your reply.

[kevin1986]

I think its because the new version uses intelmpi2019, but the old version uses intelmpi 2018.  

Is it possible to let ansys fluent 2022 to use intelmpi 2018?

[kevin1986]

In the parallel switch, after typing "intel2018" I can use intelmpi2018. I did several tests, intelmpi2018 is better than intelmpi2019 (intel). But its still slow. I cannot figure out why.

[kevin1986]

The slow parallel bandwidth makes the calculation in Fluent 2022 much slower than that of Fluent 2020...

[Ghobad Bagheri]

Hi Kevin, 

 

Have you found any solution for the bandwidth problem? I have the same issue with Ansys 2023 R1. The bandwidth is so low that a system with AMD 64 cores performs the same as my other system with 16 cores (Ansys 2021 R2). 

[kevin1986]

Nope. This problem looks like a bug, either comes from intelmpi windows version, or fluent 21/22/23 windows version. I dont know.

The latest please check here: /forum/forums/topic/new-version-fluent-is-slower-than-2020r2/

[kevin1986]

Meanwhile, I suggest you switch to Fluent 2020r2, you will see its much faster. Please let me know if you have any updates. 

[Ghobad Bagheri]

I tried Fluent 2020r2, and the Bandwidth was ok. So I am running my simulation on F2020. I am going to report the bug. 

[kevin1986]

Thank you Ghobad. I dont know where to report the bug except this forum. Maybe you have another ways to do it. But please let me know as well when ANSYS fixes it.  Much appreciated! 

[Ghobad Bagheri]

I have a contact point where I can report the bug. I will keep you posted. 

 

[kevin1986]

Ghobad, did ansys team reply your answer? Thanks. 

[Ghobad Bagheri]

Hi Kevin, 

They are still investigating the problem. I shared the info needed to convince them that there is a bug. 

By the way, I found a workaround. You can run Fluent from the terminal (any version) by enforcing INTEL MPI 2018, which is the version with better performance. You can use the following syntax. 

'fluent 3ddp -t24 -mpi=intel2018'

Description:

3ddp= 3D with double precision 

t24= number of cores 

[kevin1986]

Hello Ghobad, I am wondering if they give you any solutions regarding this problem. Thanks. 

D

[Ghobad Bagheri]

Hi Kevin,

I shared the benchmark with them, and it is evident that the performance was not the same. I haven't heard back from them yet.

[kevin1986]

Thank you for your update, Ghobad. Lets see.

D

[kevin1986]

Thanks Ghobad. I also tried to use intel2018 in GUI (type intel2018 manully). Please see the latest results. 

1. default intel's bandwidth is very slow, even 4 cores are used. Only 1000-1400.
2. when 4 core is used. Intel2018's bandwidth is around 9000. Very good. much better than default intel.  
3. when 64 cores are used. Intel2018's minimum bandwidth drops to 900, which slow down the calculation seriously.

1. 

4. msmpi has similar issue. Also slow

 

 

At last, I would say Fluent 2020r2 does not have this issue at all. I hope this information can help them debug further about this issue :-)

[Otari Kemularia]

 

Having the same problem with 2×7532. Any news regarding the issue?

I'm using Win 10 pro, can it be the problem? Moving to Win server may be solve the problem?

 

2020 fluent works well.

2022 R2:

 

2020 R2:

 

[kevin1986]

Win 10 Pro is not the problem. I tested win10Pro and win2019 server. All has this issue. 

This problem is not shown in Cent OS system.

[Otari Kemularia]

Is it pain in the ass to setup a cluster with Cent OS and send the simulations from the main workstation?

[kevin1986]

If you want to use multiple nodes, its a bit tedious to install slurm etc.

If you just want to install ansys on CENTOS. Its very easy.

 

[Otari Kemularia]

Can I open 2022 R2 files, created in Win 10 with ansys on Centos? I have almost zero experience with linux

[kevin1986]

Yes you can. 

[Ghobad Bagheri]

The default setting for MPI doesn't work the best with AMD Threadripper. So we need some fine-tuning as follows:

In the launcher, set MPI to MSMPI 

and use the following syntaxes in the environmental tab. 

FLUENT_AFFINITY=0

MPIEXEC_AFFINITY=3:P

 

Using this setting, Fluent 2023R1 and Fluent 2020R2 perform similarly with comparable computational time. Please update me if you achieve the same improvement.

[kevin1986]

Thank you very much Ghobad. I will try and keep you updated asap.

[Otari Kemularia]

It definately worked for 5800x, now I'm gonna test it for 2X7532.

Thank you Kevin for opening this thread.

Thank you Gobad for helping us out.

[kevin1986]

I justed tried, it works like a charm!

Do does it mean we cannot intelmpi for newer versions?

[Otari Kemularia]

 

Tried on 2×7532, apparently this solution works but needs a little beet tweaking. But again, you guys saved my skin, I was the one who’s idea was buying 2x7532s and them not working was causing me problems at my workplace.

2020 R2 has little bit a better performance, 1.3mil fluent case:

 

 

 

 

[kevin1986]

yeah, I also notice 2020 is faster. But now is much much better than before.