This article briefly describes the steps for running Ansys Fluent on a virtual machine (VM) that's deployed on Azure. It also presents the performance results of running Ansys Fluent on Azure.
Fluent is a computational fluid dynamics (CFD) application that's used to model fluid flow, heat and mass transfer, chemical reactions, and more. Fluent provides:
- Task-based workflows, including multiphase modeling, batteries, shape optimization, and aerodynamics, for pre-processing the generation of a CFD-ready mesh for both clean and dirty CAD.
- Accurate simulation of multiphase flows, including gas-liquid, liquid-liquid, gas-solid, particle flows, and DEM.
- A range of turbulence models, including the GEKO model.
Fluent is used in the aerospace, automotive, medical, healthcare, manufacturing, industrial equipment, communication, embedded systems, energy, retail, and consumer goods industries.
Why deploy Ansys Fluent on Azure?
- Modern and diverse compute options to align to your workload's needs
- The flexibility of virtualization without the need to buy and maintain physical hardware
- Rapid provisioning
- Performance that scales well up to 64 or 96 CPUs on a single node and linearly on multiple nodes
Architecture
Download a Visio file of this architecture.
Components
- Azure Virtual Machines is
used to create a Linux VM.
- For information about deploying the VM and installing the drivers, see Linux VMs on Azure.
- Azure Virtual Network is
used to create a private network infrastructure in the cloud.
- Network security groups are used to restrict access to the VM.
- A public IP address connects the internet to the VM.
- A physical solid-state drive (SSD) is used for storage.
Compute sizing and drivers
Performance tests of Fluent on Azure used HBv3-series VMs running the Linux CentOS operating system. The following table provides details about HBv3-series VMs.
| VM size | vCPU | Memory (GiB) | Memory bandwidth GBps | Base CPU frequency (Ghz) | All-cores frequency (Ghz, peak) | Single-core frequency (Ghz, peak) | RDMA performance (Gbps) | Maximum data disks |
|---|---|---|---|---|---|---|---|---|
| Standard_HB120rs_v3 | 120 | 448 | 350 | 1.9 | 3.0 | 3.5 | 200 | 32 |
| Standard_HB120-96rs_v3 | 96 | 448 | 350 | 1.9 | 3.0 | 3.5 | 200 | 32 |
| Standard_HB120-64rs_v3 | 64 | 448 | 350 | 1.9 | 3.0 | 3.5 | 200 | 32 |
| Standard_HB120-32rs_v3 | 32 | 448 | 350 | 1.9 | 3.0 | 3.5 | 200 | 32 |
| Standard_HB120-16rs_v3 | 16 | 448 | 350 | 1.9 | 3.0 | 3.5 | 200 | 32 |
Required drivers
To use AMD CPUs on HBv3 VMs, you need to install AMD drivers.
To use InfiniBand, you need to enable InfiniBand drivers.
Fluent installation
Before you install Fluent, you need to deploy and connect a VM, install Linux, and install the required AMD and InfiniBand drivers.
For information about deploying the VM and installing the drivers, see Run a Linux VM on Azure.
For information about installing Fluent, see the Ansys website.
Fluent performance results
HBv3 VMs with different numbers of vCPUs were deployed to determine the optimal configuration for Fluent on a single node. That optimal configuration was then tested in a multi-node cluster deployment. Ansys Fluent 2021 R2 was tested.
Results, single-node configuration
The single-node configuration was evaluated for the following test cases.
Aircraft wing test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| aircraft_wing_14m | 14,000,000 | Hexahedral | Pressure-based coupled solver, Least Squares cell based, steady | Realizable K-e Turbulence |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 860.67 | 1.00 |
| 32 | 569.03 | 1.51 |
| 64 | 442.69 | 1.94 |
| 96 | 433.45 | 1.99 |
| 120 | 429.54 | 2.00 |
Pump test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| pump_2m | 2,000,000 | Hexahedral | Pressure-based coupled solver, Least Squares cell based, steady | Realizable K-e Turbulence, Mixture Multiphase |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 213.83 | 1.00 |
| 32 | 146.38 | 1.46 |
| 64 | 118.26 | 1.81 |
| 96 | 112.53 | 1.90 |
| 120 | 115.47 | 1.85 |
Landing gear test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| landing_gear_15m | 15,000,000 | Mixed | Pressure-based coupled solver, Least Squares cell based, Unsteady | LES, Acoustics |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 871.37 | 1.00 |
| 32 | 580.31 | 1.50 |
| 64 | 501.02 | 1.74 |
| 96 | 484.46 | 1.80 |
| 120 | 489.96 | 1.78 |
Oil rig test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| oil_rig_7m | 7,000,000 | Mixed | Pressure-based segregated solver, Green-Gauss cell based, unsteady | VOF, SST K-omega Turbulence |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 377.11 | 1.00 |
| 32 | 224.16 | 1.68 |
| 64 | 152.42 | 2.47 |
| 96 | 140.81 | 2.68 |
| 120 | 132.34 | 2.85 |
Sedan test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| sedan_4m | 4,000,000 | Mixed | Pressure-based coupled solver, Green-Gauss cell based, steady | Standard K-e Turbulence |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 154.02 | 1.00 |
| 32 | 99.88 | 1.54 |
| 64 | 79.40 | 1.94 |
| 96 | 74.88 | 2.06 |
| 120 | 75.62 | 2.04 |
Combustor test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| combustor_12m | 12,000,000 | Polyhedra | Pressure-based coupled solver, Least Squares cell based, pseudo transient | Realizable K-e Turbulence, Species Transport |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 3,238 | 1.00 |
| 32 | 2,085 | 1.55 |
| 64 | 1,513 | 2.14 |
| 96 | 1,360 | 2.38 |
| 120 | 1,236 | 2.62 |
Exhaust system test case
| Test case name | Number of cells | Cell type | Solver | Models |
|---|---|---|---|---|
| exhaust_system_33m | 33,000,000 | Mixed | Pressure-based coupled solver, Least Squares cell based, steady | SST K-omega Turbulence |
The following table and graph present the test results.
| Cores | Wall-time per 100 iterations (seconds) |
Relative speed increase |
|---|---|---|
| 16 | 2,685 | 1.00 |
| 32 | 1,628 | 1.65 |
| 64 | 1,334 | 2.01 |
| 96 | 1,205 | 2.23 |
| 120 | 1,112 | 2.42 |
Results, multi-node configuration
As the preceding performance results show, HBv3-series VMs with 64 cores and 96 cores are optimal configurations. The performance improvement when you increase from 64 CPUs to 96 CPUs is between 5 and 10 percent. Taking license costs into consideration, the 64-CPU configuration is the best choice. Standard_HB120-64rs_v3 VMs, which have 64 cores, were used for the multi-node tests.
The multi-node configuration was evaluated for the same test cases.
Aircraft wing test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 442.69 | 1.00 |
| 2 | 128 | 226.06 | 1.96 |
| 3 | 192 | 149.31 | 2.96 |
| 4 | 256 | 109.23 | 4.05 |
This graph shows the relative speed increases:
Pump test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 118.26 | 1.00 |
| 2 | 128 | 55.42 | 2.13 |
| 3 | 192 | 35.53 | 3.33 |
| 4 | 256 | 24.26 | 4.88 |
This graph shows the relative speed increases:
Landing gear test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 501.02 | 1.00 |
| 2 | 128 | 247.17 | 2.03 |
| 3 | 192 | 160.02 | 3.13 |
| 4 | 256 | 117.78 | 4.25 |
This graph shows the relative speed increases:
Oil rig test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 152.42 | 1.00 |
| 2 | 128 | 75.48 | 2.02 |
| 3 | 192 | 52.76 | 2.89 |
| 4 | 256 | 41.38 | 3.68 |
This graph shows the relative speed increases:
Sedan test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 79.40 | 1.00 |
| 2 | 128 | 39.66 | 2.00 |
| 3 | 192 | 23.90 | 3.32 |
| 4 | 256 | 20.15 | 3.94 |
This graph shows the relative speed increases:
Combustor test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 1,512.56 | 1.00 |
| 2 | 128 | 828.63 | 1.83 |
| 3 | 192 | 531.82 | 2.84 |
| 4 | 256 | 359.86 | 4.20 |
This graph shows the relative speed increases:
Exhaust system test case
| Number of nodes | Number of cores | Wall-clock time per 100 iterations (seconds) | Relative speed increase |
|---|---|---|---|
| 1 | 64 | 1,333.72 | 1.00 |
| 2 | 128 | 629.02 | 2.12 |
| 3 | 192 | 399.66 | 3.34 |
| 4 | 256 | 304.05 | 4.39 |
This graph shows the relative speed increases:
Azure cost
Only wall-clock time per 100 iterations of each model is considered for these cost calculations. Application installation time and license costs aren't considered.
You can use the Azure pricing calculator to estimate the costs for your configuration.
For a single-node configuration, you can multiply the wall-clock times by the Azure hourly costs for HBv3-series VMs to compute total costs. For the current hourly costs, see Linux Virtual Machines Pricing. Here are the times for a single-node configuration:
| VM size | Number of CPUs | Wall-clock time (hours) |
|---|---|---|
| Standard_HB120-16rs_v3 | 16 | 2.33 |
| Standard_HB120-32rs_v3 | 32 | 1.48 |
| Standard_HB120-64rs_v3 | 64 | 1.15 |
| Standard_HB120-96rs_v3 | 96 | 1.06 |
| Standard_HB120rs_v3 | 120 | 1.00 |
For a multi-node configuration, you can multiply the wall-clock times by the number of nodes and the Azure hourly costs for HBv3-series VMs to compute total costs. For the current hourly costs, see Linux Virtual Machines Pricing. Here are the times for a multi-node configuration:
| VM size | Number of nodes | Number of cores | Wall-clock time (hours) |
|---|---|---|---|
| Standard_HB120-64rs_v3 | 1 | 64 | 1.15 |
| Standard_HB120-64rs_v3 | 2 | 128 | 0.58 |
| Standard_HB120-64rs_v3 | 3 | 192 | 0.38 |
| Standard_HB120-64rs_v3 | 4 | 256 | 0.27 |
Summary
- Ansys Fluent 2021 R2 was successfully tested on HBv3-series Azure VMs.
- In single-node configurations, performance scaled well up to 64 or 96 CPUs. After that point, the speed increase dropped off.
- In multi-node configurations, performance scaled linearly as nodes were added.
Contributors
This article is maintained by Microsoft. It was originally written by the following contributors.
Principal authors:
- Hari Bagudu | Senior Manager
- Gauhar Junnarkar | Principal Program Manager
- Vinod Pamulapati | HPC Performance Engineer
Other contributors:
- Mick Alberts | Technical Writer
- Guy Bursell | Director Business Strategy
- Sachin Rastogi | Manager
To see non-public LinkedIn profiles, sign in to LinkedIn.
Next steps
- GPU-optimized virtual machine sizes
- Virtual machines on Azure
- Virtual networks and virtual machines on Azure
- Learning path: Run high-performance computing (HPC) applications on Azure