Learning Forum

[a.y.al-wardi]

I am running a transient case on Ansys Fluent 2023 R2, with around a mesh of around 40,000 elements. The run is for 300s (flow time) with a time step of 0.3. I have tried running it using a desktop and it takes around 33mins to complete, using 12 cores and requesting parallelisation. However, running the same case on a high-performance supercomputing cluster takes more than 24 hours, requesting parallelisation and 12 cores. 

Desktop specs: processor: Intel ® Core ™ i7-14700 2.10 GHZ; RAM: 32GB; system type: 64bit operating system, x-64 based processor, cores: 20 cores and 28 logical processors. 
HPC specs: cores: over 7,000 cores, memory: Up to 2 TB per node,  operating system: Linux, scheduler: SLURM job scheduler, storage: multi-terabyte parallel filesystem, parallel processing: supports both MPI and OpenMP parallelism

Checks performed: verified that the HPC is correctly parallelizing the job and utilizing the requested number of cores, tested different numbers of cores, loaded the Intel MPI library via module load mpi and used a dedicated node. Despite all of these checks, the job still takes way longer than expected to run on the HPC compared to the desktop. 

I have attached a copy of the job script used, and the input script for Fluent.

Any suggestions/advice to resolve this are highly appreciated. Please let me know if further information is required. 

[Rob]

I'd probably not try over 4 cores parallel (if that) as you finish up paying more time in message passing than you gain in compute power. Re the slow down, what else is running on the system? Are you running exclusive to that box or is someone else also running parallel and therefore using the RAM channels?

[a.y.al-wardi]

I tried running the job requesting 4 cores, that didn’t help speed the run on the HPC. 

We have also tried submitting to a dedicated node (no one else using it), that also didn’t help. 

 

[Rob]

Were you running the GPU or CPU solver? 

[a.y.al-wardi]

How do you differentiate between the two? How can I tell?

[Rob]

Did you tell Fluent to run GPU version? I have no idea how the CPU solver will run on GPU: assuming I'm reading your cluster spec correctly. 

[a.y.al-wardi]

 

I am using CPU, also using CPU on the HPC. 

The HPC also has 7 nodes that use GPU in addition to nodes with CPUs. 

Aplogies for the confusion. 

 

[Rob]

No worries. In which case uncertain - you may need to check with IT to see what's going on.