Accelerating water wading simulation using Altair® nanoFluidX® on Azure Nvidia A100 and Nvidia H100

Over the past few weeks, we have been working with Altair engineers to validate and verify the nanoFluidX v2024 product on Azure. This software offers significant benefits to engineers who solve problems where traditional CFD techniques require significant manual time and computational resources. Vehicle wading is a critical durability property for engineers to monitor water coverage and accumulation, and to assess the potential for damage due to water impact.

nanoFluidX’s Lagrangian meshless approach was designed from the ground up for GPU computing using NVIDIA CUDA, making it one of the fastest SPH solvers on the market. It allows for extremely fast model setup, allowing engineers to iterate faster.

With this validation, we wanted to explore the possibilities of GPU computing in two ways: how nanoFluidX would perform on an Nvidia H100 series GPU, and how it would perform while scaling to an 8-way GPU virtual machine (VM). Let’s first look at the A100 and H100.

that NC_A100_v4 There are three flavors, with 1, 2, or 4 A100 80GB GPUs. By default, these are PCIe-based GPUs, but internally they are NVlink paired. The rest of the system consists of 24 (non-multithreaded) AMD Milan CPU cores, 220GB of main memory, and 960TB NVME local scratch disk per GPU. When choosing a 2 or 4 GPU VM, these numbers are multiplied for a total of 880GB of main memory.

that NC_H100_v5 It has grown in GPU capabilities. It comes in 1 or 2 GPU configurations, built around the Nvidia 94GB H100 NVL. This GPU has a PCIe interface to the main system, but many of its features are in line with the SXM H100 series. The CPU cores have been increased to 40 (non-multithreaded) AMD Genoa CPU cores and 320GB of main memory, and it also comes with an upgraded 3.5TB NVME local scratch disk.

The benchmark run for this validation is the Altair CX-1 car model. This benchmark represents a production-scale model of a full-size vehicle driving a 24 m wading channel at 10 km/h for 15 seconds.

“Working with Microsoft and NVIDIA, we have successfully validated nanoFluidX v2024 on NVIDIA’s A100 and H100 GPUs. The latest release boasts a solver that is 1.5X faster than before and provides improved scaling across multiple GPUs. These benchmarks demonstrate significant performance improvements of up to 1.8X with NVIDIA H100, reducing simulation times and accelerating design cycles. These advancements make nanoFluidX one of the fastest Smoothed-particle Hydrodynamic (SPH) GPU codes on the market.” – David Curry, Senior Vice President of CFD and EDEM at Altair

As you can see in the table below, the H100 delivers higher performance than the A100, which is consistent with the performance increase Nvidia announced between the two generations. So both software and Azure VMs can ensure that these GPUs reach their compute potential.

Since nanoFluidX supports multi-GPU systems, we wanted to verify its scalability and test it on an 8-way GPU ND series. Again, we tested it on an Nvidia A100 system, NDads_A100_v4, And the successor based on Nvidia H100: NDisr_H100_v5Both systems have eight GPUs, all interconnected via NVlink.

Chart showing the performance increase of H100(NCv5) compared to A100(NCv4)

As you can see from the table above, nanoFluidX effectively utilizes all GPU power. It achieved a 1-hour simulation period on NDisr_H100_v5, which significantly impacted the turnaround and design cycle time.

You can go to the portal and request quota and spin up these VMs, but we often see customers looking for an HPC environment that is better integrated into their production workflow. Altair offers a solution to run your projects on Azure through the Altair One platform. Work with your Altair representative to enable this Azure-based solution. Or, you can use Altair Unlimited, an Altair SaaS solution in the Virtual Appliance Marketplace, to deploy and manage your own HPC cluster on Azure. Work with your Azure account manager to enable GPU quota for HPC.

#AzureHPCAI