This article briefly describes the steps for running Altair EDEM on a virtual machine (VM) that's deployed on Azure. It also presents the performance results of running EDEM on Azure.
EDEM is an application that's used for bulk and granular material simulation. EDEM uses discrete element method (DEM) to simulate and analyze the behavior of coal, mined ores, soils, fibers, grains, tablets, and powders.
EDEM simulation provides engineers with insight into how those materials interact with equipment during a range of operation and process conditions. It can be used by itself or combined with other CAE tools.
Companies in the heavy equipment, off-road, mining, steelmaking, and process manufacturing industries use EDEM to understand and predict granular material behaviors, evaluate equipment performance, and optimize processes.
Why deploy EDEM on Azure?
- You can use EDEM to model particle shape by using the multi-sphere method.
- EDEM is highly parallelized for use on multi-core shared memory workstations, GPU hardware, and multi-GPU systems.
- The solver engine is fully double precision across all platforms.
- EDEM can simulate large and complex particle systems.
- EDEM provides advanced post-processing capabilities.
Architecture
Download a Visio file of this architecture.
Components
- Azure Virtual Machines is used to create Windows VMs. For information about deploying VMs and installing the drivers, see Windows VMs on Azure.
- Azure Virtual
Network is
used to create a private network infrastructure in the cloud.
- Network security groups restrict access to the VMs.
- A public IP address connects the internet to the VMs.
- A physical SSD provides storage.
Compute sizing and drivers
Performance tests of EDEM on Azure used NCv3, NC A100 v4, and ND A100 v4 series VMs running on Windows. The following table provides the configuration details.
| Size | vCPUs | Memory, in GiB | Temporary storage (SSD), in GiB | GPUs | GPU memory, in GiB | Maximum data disks | Maximum uncached disk throughput: IOPS / MBps | Maximum NICs |
|---|---|---|---|---|---|---|---|---|
| Standard_ND96asr_v4 | 96 | 900 | 6,000 | 8 | 40 | 32 | 80,000 / 800 | 8 |
| Standard_NC24ads_A100_v4 | 24 | 220 | 1,123 | 1 | 80 | 12 | 30,000 / 1,000 | 2 |
| Standard_NC48ads_A100_v4 | 48 | 440 | 2,246 | 2 | 160 | 24 | 60,000 / 2,000 | 4 |
| Standard_NC96ads_A100_v4 | 96 | 880 | 4,492 | 4 | 320 | 32 | 120,000 / 4,000 | 8 |
| Standard_NC6s_v3 | 6 | 112 | 736 | 1 | 16 | 12 | 20,000 / 200 | 4 |
Required drivers
To use EDEM on the previously listed VMs as described in this article, you need to install NVIDIA and AMD drivers.
EDEM installation
Before you install EDEM, you need to deploy and connect a VM and install the required NVIDIA and AMD drivers.
For information about deploying the VM and installing the drivers, see Run a Windows VM on Azure.
To download EDEM:
- Sign in to Altair One Marketplace.
- Select EDEM in the product list.
- Select the appropriate operating system and download.
- Download the license manager.
For EDEM installation instructions, see the documents on Altair One Marketplace.
EDEM performance results
Seven models are used to test the performance of EDEM on Azure VMs. The following table provides details.
| Model | Angle of repose | Bed of material | Hopper discharge | Powder mixer | Screw augur | Mill | Transfer chute |
|---|---|---|---|---|---|---|---|
| Description | Cylinder angle of repose | Bed of material with tillage tool | Hopper emptying into container | Powder mixer operation | Screw augur operation | Mill operation | Transfer chute with dynamic factory |
 |
 |
 |
 |
 |
 |
 |
|
| Particle radius (m) | 0.0005 - 0.001 | 0.002 - 0.004 | 0.003 | 0.0005 | 0.001 | 0.005 | 0.0045 - 0.009 |
| Number of spheres | 3 | 3 | 3 | 1 | 1 | 1 | 3 |
| Size distribution | Random | Random | Fixed | Fixed | Fixed | Fixed | Random |
| Number of particles | 1,000,000 | 1,000,000 | 1,000,000 | 1,000,000 | 1,000,000 | 1,000,000 | 1,000,000 |
| Physics | Hertz-Mindlin | Hertz-Mindlin with JKR | Hertz-Mindlin | Hertz-Mindlin | Hertz-Mindlin | Hertz-Mindlin | Hertz-Mindlin with JKR |
| Time steps | 5.73E-06 | 5.00E-05 | 4.00E-05 | 9.20E-06 | 1.40E-05 | 0.00016 | 5.97E-05 |
| Total time | 0.5 | 1 | 1 | 1 | 1 | 1 | 1 |
| Save interval | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 | 0.1 |
| Grid cell size (x Rmin) | 3 | 3 | 3 | 3 | 3 | 3 | 5 |
| Factory | No | No | No | No | No | No | Yes |
| Periodic boundaries | No | No | No | No | No | No | No |
Results for EDEM 2021.1 on NDv4 and NCv3 VMs
The following table shows the elapsed wall-clock times, in seconds, required to complete each simulation.
| Model | ND96asr_v4, 96 CPUs | ND96asr_v4, 1 A100 GPU | NC6s_v3, 1 V100 GPU |
|---|---|---|---|
| Angle of repose | 12,819.80 | 1,543.66 | 2,319.39 |
| Bed of material | 2,650.56 | 320.24 | 475.04 |
| Hopper discharge | 9,318.89 | 566.59 | 1,030.38 |
| Powder mixer | 14,028.50 | 1,013.98 | 1,312.27 |
| Screw auger | 8,871.59 | 1,295.16 | 1,158.98 |
| Mill | 1,339.11 | 83.18 | 116.49 |
| Transfer chute | 3,859.01 | 310.22 | 437.92 |
The following graph uses a Standard_ND96asr_v4, 96-vCPU VM as a baseline and shows how much the speed increases on A100-GPU and V100-GPU VMs.
Results for EDEM 2022.1 on NC A100 v4 VMs
The following table shows the elapsed wall-clock times, in seconds, required to complete each simulation.
| Model | NC24ads_A100_v4, 24 vCPUs | NC24ads_A100_v4, 1 GPU | NC48ads_A100_v4, 2 GPUs | NC96ads_A100_v4, 4 GPUs |
|---|---|---|---|---|
| Angle of repose | 22,950.80 | 649.59 | 404.05 | 339.38 |
| Bed of material | 4,835.23 | 140.10 | 87.67 | 72.11 |
| Hopper cischarge | 11,457.00 | 301.33 | 187.68 | 144.45 |
| Powder mixer | 13,906.20 | 606.43 | 394.99 | 361.85 |
| Screw auger | 11,089.00 | 536.27 | 343.75 | 278.92 |
| Mill | 1,141.65 | 46.33 | 34.26 | 28.49 |
| Transfer chute | 4,146.64 | 117.67 | 77.98 | 63.80 |
The following graph uses a Standard_NC24ads_A100_v4, 24-vCPU VM as a baseline and shows how much the speed increases on VMs with varying numbers of A100 GPUs.
Results for EDEM 2022.1 on ND A100 v4 VMs
The following table shows the elapsed wall-clock times, in seconds, required to complete each simulation.
| Model | NC24ads_A100_v4, 24 vCPUs | ND96asr_v4, 1 GPU | ND96asr_v4, 2 GPUs | ND96asr_v4, 3 GPUs | ND96asr_v4, 4 GPUs |
|---|---|---|---|---|---|
| Angle of repose | 22,950.80 | 682.66 | 517.99 | 491.00 | 494.08 |
| Bed of material | 4,835.23 | 148.17 | 106.42 | 93.42 | 98.30 |
| Hopper discharge | 11,457.00 | 316.62 | 236.32 | 204.62 | 189.02 |
| Powder mixer | 13,906.20 | 646.77 | 477.97 | 463.59 | 477.86 |
| Screw auger | 11,089.00 | 566.37 | 408.32 | 378.17 | 341.56 |
| Mill | 1,141.65 | 51.79 | 41.29 | 39.93 | 35.96 |
| Transfer chute | 4,146.64 | 126.46 | 90.54 | 85.01 | 81.35 |
The following graph uses a Standard_NC24ads_A100_v4, 24-vCPU VM as a baseline and shows how much the speed increases on ND96asr_v4 VMs with varying numbers of A100 GPUs.
Azure cost
Only model running time (wall-clock time) is considered for these cost calculations. Application installation time isn't considered. The calculations are indicative. The actual numbers depend on the size of the model.
You can use the Azure pricing calculator to estimate costs for your configuration.
The following tables provide elapsed times in hours. To compute the total cost, multiply the elapsed time by the Azure VM hourly cost. For current hourly costs, see Windows Virtual Machines Pricing.
EDEM 2021.1 costs on ND96asr_v4 VMs
| Model | ND96asr_v4, 96 vCPUs | ND96asr_v4, 1 GPU |
|---|---|---|
| Angle of repose | 3.56 | 0.43 |
| Bed of material | 0.74 | 0.09 |
| Hopper discharge | 2.59 | 0.16 |
| Powder mixer | 3.90 | 0.28 |
| Screw auger | 2.46 | 0.36 |
| Mill | 0.37 | 0.02 |
| Transfer chute | 1.07 | 0.09 |
EDEM 2021.1 costs on NCv3 VMs
| Model | NC6s_v3, 1 GPU |
|---|---|
| Angle of repose | 0.64 |
| Bed of material | 0.13 |
| Hopper discharge | 0.29 |
| Powder mixer | 0.36 |
| Screw auger | 0.32 |
| Mill | 0.03 |
| Transfer chute | 0.12 |
EDEM 2022.1 costs on NC A100 v4 VMs
| Model | NC24ads_A100_v4, 24 vCPUs | NC24ads_A100_v4, 1 GPU | NC24ads_A100_v4, 2 GPUs | NC24ads_A100_v4, 4 GPUs |
|---|---|---|---|---|
| Angle of repose | 6.38 | 0.18 | 0.11 | 0.09 |
| Bed of material | 1.34 | 0.04 | 0.02 | 0.02 |
| Hopper discharge | 3.18 | 0.08 | 0.05 | 0.04 |
| Powder mixer | 3.86 | 0.17 | 0.11 | 0.10 |
| Screw auger | 3.08 | 0.15 | 0.10 | 0.08 |
| Mill | 0.32 | 0.01 | 0.01 | 0.01 |
| Transfer chute | 1.15 | 0.03 | 0.02 | 0.02 |
EDEM 2022.1 costs on ND96asr_v4 VMs
| Model | ND96asr_v4, 1 GPU | ND96asr_v4, 2 GPUs | ND96asr_v4, 3 GPUs | ND96asr_v4, 4 GPUs |
|---|---|---|---|---|
| Angle of repose | 0.19 | 0.14 | 0.14 | 0.14 |
| Bed of material | 0.04 | 0.03 | 0.03 | 0.03 |
| Hopper discharge | 0.09 | 0.07 | 0.06 | 0.05 |
| Powder mixer | 0.18 | 0.13 | 0.13 | 0.13 |
| Screw auger | 0.16 | 0.11 | 0.11 | 0.09 |
| Mill | 0.01 | 0.01 | 0.01 | 0.01 |
| Transfer chute | 0.04 | 0.03 | 0.02 | 0.02 |
Summary
- EDEM 2021.1 was deployed and tested on ND A100 v4 and NCv3 VMs with one GPU and two GPUs. EDEM 2022.1 was deployed and tested on ND A100 v4 and NC A100 v4 VMs with one GPU and multiple GPUs.
- On Azure, the GPU technology in EDEM provides faster processing than CPU configurations. Tests demonstrate speed increases of about 80x with NC A100 v4 A100 GPUs and about 60x with ND A100 v4 A100 GPUs.
- The complexity of the model affects GPU scale-up.
- The NC A100 v4 VM demonstrates better GPU acceleration than other VM configurations on Azure.
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
- Saurabh Parave | HPC Performance Engineer
- Kalai Selvan | HPC Performance Engineer
Other contributors:
- Mick Alberts | Technical Writer
- Guy Bursell | Director of Business Strategy
- Sachin Rastogi | Manager
To see non-public LinkedIn profiles, sign in to LinkedIn.
Next steps
- GPU-optimized virtual machine sizes
- Windows virtual machines on Azure
- Virtual networks and virtual machines on Azure
- Learning path: Run HPC applications on Azure