Machine learning and seismic wave analysis workstation

A customer engaged in research focused on machine learning and numerical calculations contacted us to inquire about the introduction of a workstation that can be used in a 100V environment.
The software we plan to use is PyTorch, TensorFlow, OpenSWPC, SPECFEM, hbi, and our own MPI/OpenMP parallel code.
The desired hardware specifications are a CPU with 64 or more cores, memory of 256GB or more (bandwidth is important), a GPU with 96GB or more VRAM, and an SSD with high-speed I/O support and 4TB or more.
The budget is estimated to be around 500 million yen.

CPU	AMD Ryzen Threadripper PRO 9985WX 3.20GHz (boost 5.4GHz) 64C / 128T
memory	Total 512GB DDR5 5600 REG ECC 64GB x 8
Storage 1	4TB SSD M.2 NVMe Gen5
Storage 2	4TB SSD M.2 NVMe Gen4
Video	NVIDIA RTX PRO 6000 Max-Q 96GB
network	on board (10GBase-T x2)
Housing + power supply	Mid-tower chassis + 1600W 80PLUS PLATINUM
OS	soul linux
security	No HDD return required service for 1 year

Optimal CPU configuration for parallel computing

As the optimal CPU for numerical calculations,64 cores 128 threads OfRyzen Threadripper PRO 9985WXIs adopted.
In addition to the large number of cores, the high operating frequency of 3.20 GHz base clock enables efficient parallel processing using MPI and OpenMP.
In addition, the 8-channel DDR5 memory reduces memory bandwidth bottlenecks, allowing for stable processing even when running multiple processes simultaneously.

Memory configuration

Wave analysis software that handles large-scale matrix data, such as OpenSWPC and SPECFEM, requires sufficient memory capacity and high-speed access performance.
In this configurationDDR5-5600 REG ECC 512GB (64GB x 8 pieces)It is equipped with a network environment with ample capacity and bandwidth.
Even complex numerical calculations and parallel processing can be performed with stable performance.

GPU selection and scalability

For machine learning applications, GPUs haveRTX PRO 6000 Max-Q with 96GB of VRAMis selected.
The large-capacity graphics memory is suitable not only for deep learning but also for technical calculations that use GPUs, and can be used for a wide range of analytical processing.
In the SPECFEM3D Cartesian analysis, NVIDIA officially reported that multi-GPU performance improved.

Reference: SPECFEM3D Cartesian GPU & Software Configurations

This configuration is intended for standalone operation in a 100V environment, and a second GPU can be added if a 200V environment is available.
It comes equipped with a 1600W power supply, allowing for smooth future GPU expansion.

Storage and I/O Performance

To achieve the high-speed I/O required for analysis,Gen5 NVMe SSD 4TB (Read speed over 10,000MB/s)is equipped with
This allows for high-speed reading and writing of large amounts of data, increasing the efficiency of analysis processing.
In addition, it is equipped with a 4TB Gen4 NVMe SSD, and by separating the working area and storage area, you can achieve both efficient and stable data management.

For those who are active in these fields Earthquake analysis Structural mechanics computational physics Artificial intelligence GPU computing

Tegara's custom-made PC production service not only caters to initial use, but also supports system expansion in anticipation of future expansion of research scale.
We not only propose configurations that meet various software requirements, but also accept consultations regarding the construction of an entire research environment.
Please feel free to contact us and we will provide the best solution to suit your needs.

Keyword ・What is PyTorch? PyTorch is a Python-based open-source oscillatory learning framework developed by Meta (formerly Facebook). It features intuitive description using dynamic computational graphs and is widely used by researchers at universities and companies. It supports a wide range of fields, including natural language processing and medical image analysis, and its GPU support and extensive libraries enable fast and flexible model development. reference:PyTorch *Jumps to an external site ・What is Tensorflow? TensorFlow is an open-source machine learning and deep learning library developed by Google that enables efficient numerical computation through tensor operations. It is used by a wide range of engineers, including researchers at universities and companies, and AI engineers, and is capable of handling a variety of tasks such as image recognition and natural language processing. With its high flexibility and scalability, it is used in a wide range of applications, from research to production environments. reference:TensorFlow *Jumps to an external site ・What is OpenSWPC? OpenSWPC is an open-source numerical simulation software that can analyze seismic wave propagation with high accuracy. It is widely used in disaster prevention and structural analysis fields at universities, research institutes, and companies both in Japan and overseas, and is capable of three-dimensional analysis of earthquake motion, tsunamis, ground response, and more. It supports high-speed calculations using GPUs and input/output in NetCDF and SAC formats, and is highly regarded as a research platform with high reproducibility and scalability. reference:OpenSWPC official website *Jumps to an external site ・What is SPECFEM? SPECFEM is open source software for seismic wave analysis using the Spectral Element Method (SEM). It is used by geophysical and seismological research institutes and energy-related companies for earthquake source analysis, ground property evaluation, etc. With high spatial resolution and large-scale analysis performance through parallel computing, it is possible to perform precise simulations from the entire globe to the city scale. reference:SPECFEM *Jumps to an external site ・What is HBI? HBI is a research simulation software that can analyze earthquake cycles in multiple dimensions with high accuracy. Using the boundary element method and H-matrix, it is possible to reproduce complex mechanical phenomena such as 2D and 3D fault movement, friction, and fluid pressure diffusion. It can handle complex fault geometries and friction models, and can easily perform large-scale analysis using parallel processing, making it widely used by university and corporate researchers in the fields of earthquakes and geophysics. reference:sozawa94/hbi (GitHub) *Jumps to an external site

■ Click here for details and inquiries about this PC case Machine learning and seismic wave analysis workstation * Please enter the name of the case or your desired conditions.