-
-
July 22, 2023 at 1:35 pm
paul_pinto
SubscriberHello,Â
I am a PhD student at KU Leuven and doing my simulation on a 24-core processor using only 40 threads. In the past week or so I was able to do one time step in about 30 seconds or so. This week it is showing me that the time required for one timestep is 40 mins. There is nothing wrong with the computer. I tried reading /forum/forums/topic/ansys-running-extremely-slower-than-normal/, unfortunately, our firewall is controlled by IT and I can't switch it off.Â
Â
How can I slove this problem?Â
Best
SamsonÂ
-
July 25, 2023 at 10:54 am
Rob
Forum ModeratorYou can try shutting down the solver & restarting. That will eliminate memory leaks as a factor. Is anything else running on that computer?Â
-
July 25, 2023 at 10:57 am
paul_pinto
SubscriberI tried restarting the solver multiple times. I also tried restarting the whole system and it didn't help much. There is nothing else running, this is a computer dedicated to ANSYS. Â
-
-
July 25, 2023 at 10:59 am
Rob
Forum ModeratorOK. What models are you using? If you check cpu load are all nodes equally balanced? How much free RAM is there?Â
-
July 25, 2023 at 11:07 am
paul_pinto
SubscriberI am doing a porous media heat transfer simulation using K-omega SST and SIMPLE algorithm with a Second order scheme for all parameters and bounded second order for transient. All CPUs are loaded equally about 2.2%. The mesh count is 375000 elements among which 60000 are solid domain and not porous media. Since I am using the LTNE model I have 690000 elements finally that only use about 12% of my 128GB RAM.Â
-
-
July 25, 2023 at 11:43 am
Rob
Forum ModeratorThat shouldn't be a problem, try running on a different number of cores to force the mesh to repartition and see if that helps. Are there any animations etc? With 700k cells you may find you start to lose parallel efficiency at 15-20 nodes.Â
-
July 25, 2023 at 11:48 am
paul_pinto
SubscriberHow does one know how many cores to use for how many cells? No animations, just some simple mass-weighted temperature and area-weighted pressure calculations with some report definations. I shall take your suggestion and load it on 44 cores.Â
-
-
July 25, 2023 at 11:58 am
Rob
Forum ModeratorIt's a bit model dependent, but typically 50k-ish fluid cells per core is about as low as I'd go in most cases. You're trading cpu speed with memory bandwidth (interconnect stuff) so eventually the data transfer outweighs the boost in cpu power.Â
-
July 25, 2023 at 12:00 pm
paul_pinto
SubscriberThanks for the advice it is working now.Â
Best
SamsonÂ
-
-
- The topic ‘ANSYS Fluent Extremely Slow from a few days’ is closed to new replies.
-
3472
-
1057
-
1051
-
934
-
902
© 2025 Copyright ANSYS, Inc. All rights reserved.