Machine with NVIDIA A800

A customer involved in metal material research contacted us to inquire about introducing a machine for MD calculations.
Specifically, the expected specifications are as follows, and we would like proposals for a configuration that can be realized within a budget of 700 to 800 million yen.

GPU: Multiple A100s or V100s if possible
CPU: Xeon 2CPU configuration
Storage: 2TB SSD + 4TB HDD
Memory: 256GB

The software to be used is Quantum Espresso, LAMMPS, Advance Nanolabo, NeuralMD, CUDA Toolkit, and MATLAB, and the OS is to be pre-installed with AlmaLinux 9. We have also received questions about the benefits of using multiple GPUs for the software to be used.

Additionally, customers have been looking at the proposed clustering examples on the Tegsys website and have been asking questions such as whether a scalable machine or a cluster configuration would be better for this purpose, and whether there are any points to be aware of regarding the behavior of a two-CPU configuration.

CPU Intel Xeon Platinum 8568Y+ 2.30GHz(TB 4.10GHz) 48C/96T × 2
memory Total 256GB DDR5 5600 REG ECC 32GB x 8
Storage 1 2TB SSD M.2 NVMe Gen4
Storage 2 4TB HDD S-ATA
Video NVIDIA T400 4GB (MiniDisplayPort x3)
GPU NVIDIA A800 40GB Active
network on board (10GBase-T x2)
Housing + power supply Tower type housing + 1500W
OS Alma Linux 9

About GPU

If you specify double precision, the NVIDIA A800 40GB is currently suitable. However, since the NVIDIA A800 40GB does not have a screen output, we recommend adding a card such as the T400 for screen output when using Matlab.

About 2CPU configuration

Regarding the 2-CPU configuration, it is recognized as the total number of cores on the OS, so it does not behave specially. However, if you look in detail, the CPUs are connected by a high-speed bus, but the PCI-E bus on which the memory and GPU are mounted is connected to each CPU. Therefore, when accessing the memory address or GPU managed by one CPU, it is necessary to go through the connection bus (UPI) between the CPUs, which may cause some delay.

About NanoLabo

The more CPU cores the NanoLabo has, the greater its processing power will be.

NeuralMD calculations using multiple GPUs have high parallelization efficiency, so using N GPUs will increase the calculation speed by N times. Calculations using two GPUs are twice as fast as calculations using one GPU, but the hardware cost is also twice as high. Therefore, the number of GPUs is determined based on the purpose of use and cost-effectiveness, such as whether calculation time or price is important. In addition, NVLink Bridge and NVSwitch are not required when multiple GPUs are installed.

Regarding whether a scalable machine or a cluster configuration is better, from a software perspective, neither configuration will affect performance such as calculation speed. Please choose from the perspective of hardware such as cooling and noise.

■ Keywords

・What is Quantum Espresso?

Quantum Espresso is open source software written in Fortran95 and distributed free of charge under the GNU General Public License. First-principles electronic state calculation program. It is a first-principles calculation code based on density functional theory. It is capable of optimizing atomic positions and cell structures, calculating electronic states and band structures, analyzing phonon vibrations, and calculating neutron scattering. It operates on multiple platforms and supports parallelization using MPI and OpenMP. It is widely used for theoretical analysis of material properties and the search for new substances, and is also a useful tool in computational biology.

Reference: Home Page – Quantum Espresso *Jumps to an external site

・What is LAMMPS?

LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) is an open-source molecular dynamics calculation application developed by Sandia National Laboratories in the United States.It has a reputation for its wide range and flexibility of force fields that can be handled by parameter settings.

Reference: LAMMPS Molecular Dynamics Simulator *Jumps to an external site

・What is Advance/NanoLabo?

Advance/NanoLabo is a materials simulator designed for beginners. It has features such as an intuitive and easy-to-use GUI, a variety of modeling functions, and support for Neural Network force fields. Its appeal is that even beginners such as experimental researchers can use it easily.

reference:Advance/NanoLabo (AdvanceSoft Corporation) *Jumps to an external site

・What is NeuralMD?

NeuralMD is a molecular dynamics calculation software based on the Neural Network Potential, which enables calculations that are faster than first-principles calculations and more accurate than existing molecular dynamics calculations.

reference:Advance/NeuralMD – Advance/NanoLabo *Jumps to an external site

・What is CUDA Toolkit?
CUDA Toolkit is a parallel computing platform for GPUs provided by NVIDIA. High-speed parallel programming using NVIDIA's GPU architecture is possible from C/C++. The computing power of GPUs can be utilized in various fields such as DeepLearning, scientific computing, and computer graphics.It includes tools such as compilers, libraries, and debuggers, and is provided as an SDK.It also supports multi-GPU environments and can be used in a wide range of environments from workstations to clouds.

Reference: CUDA Toolkit – Free Tools and Training | NVIDIA Developer *Jumps to an external site

・What is MATLAB?

MATLAB is software for numerical analysis used in data analysis, algorithm development, and model creation in a wide range of fields such as engineering, science, and economics, and is also the name of the programming language used internally.

Reference: MATLAB – Technical Computing Language Product Information *Jumps to an external site

Feel free to request a quote based on your usage and budget - Tegsys' simple inquiry form