Project Background

A world-renowned civilian robotics company in Hangzhou faced a key challenge in optimizing and improving the motion control algorithms for their robots. Motion control algorithms are crucial for the stability, flexibility, and adaptability of robots in different terrains. To achieve more precise and efficient algorithm performance, utilizing advanced computing platforms for inference training has become an indispensable part of the solution.

Scheme

The GST G4DA42-8AD server, as a flagship product designed for high-performance computing, is ideally suited for the inference training needs of such algorithms.

Advantage

High Performance Computing: This server delivers powerful computing capabilities. The AMD EPYC processor, known for its high core count and multi-threaded design, can process large amounts of data in parallel, which is crucial for accelerating the training and inference of deep learning models.

Strong Scalability: The server offers excellent scalability, supporting additional memory, storage, and accelerator cards (such as GPUs), which is highly beneficial for handling large datasets and training complex models. Scalability ensures that as algorithm complexity and data volume grow, the system can continue to meet performance demands.

High Reliability: Reliability is critical for high-performance computing environments that run 24/7. With high-quality components, advanced cooling systems, and redundant designs (such as power supply redundancy), the system ensures high stability and long-term operational reliability.

Project Background

A cloud computing infrastructure service provider has long specialized in the field of mobile Internet and has a deep understanding of mobile Internet business scenarios and user needs. In response to the demand for AI computing power, it has actively laid out two major cloud computing centers under the "Eastern Data, Western Computing" initiative to build an intelligent computing power center for AI training, inference, and R&D. 

Solutions

The Ulanqab AI Computing Center is a key strategic project in 2023. As the core provider of AI computing servers for this project, GST customized and optimized the machine models according to the project background to ensure the smooth progress of the project.

Project Advantage

Adaptability to cold environment: In consideration of the sub-zero natural temperature in Ulanqab during winter, GST optimized the machine pre-heating mechanism, carefully selected machine configurations, and conducted on - site burn - in tests to ensure stable operation in the natural low temperature environment.

Large - scale delivery capability: Regarding the project's large-scale delivery requirement, GST’s six operation centers and the national warehouse logistics system guarantee the ability to deliver goods on time, with high quality and in the required quantity.

Monitoring solution: In response to the cloud platform's server monitoring mechanism, GST's technical team collaborated to customize a BMC monitoring solution to obtain and monitor the GPU operation data.

Project Background

With the continuous evolution of information technologies such as big data, cloud computing, the Internet of Things, and artificial intelligence in the Internet industry, various emerging business models are rapidly developing. Leading companies in the Internet sector are investing heavily in building data centers, resulting in an explosive increase in server demand. According to IDC data, the Internet industry is the largest sector for server procurement, and it is also the fastest-growing industry.

Solution

Diverse customized products: Offer a variety of customized server types, including standard rack servers, multi - node servers, and artificial intelligence servers.

High - quality and rapid system delivery: Establish a connection between the user - side and the manufacturer's factory quality system to ensure that the quality meets the standards. Through the implementation of the factory automation system and the batch deployment of firmware adaptation, achieve rapid system delivery.

Professional operation and maintenance capabilities: Assist users in realizing fully automated operation and maintenance of the basic computing platform, and complete the evolution from standardization and systematization to intelligence.

Advantage

High Performance Computing: This server delivers powerful computing capabilities. The AMD EPYC processor, known for its high core count and multi-threaded design, can process large amounts of data in parallel, which is crucial for accelerating the training and inference of deep learning models.

Ensuring Forward - Looking Needs: Ensure that users can safely and stably apply forward-looking technologies, and help users build a large-scale computing basic support platform with ultra - high computing power, intelligence, and high added value.

Rapid batch delivery: Standardize the operations in the equipment delivery process, reduce manual intervention, achieve batch delivery and deployment, and help users quickly implement the large - scale, high - performance, and scalable computing power practice.

Project Background

Quantitative trading is a process where advanced mathematical models replace human subjective judgment. It uses computer technology to sift through vast amounts of historical data to identify multiple "high - probability" events that can yield excess returns, formulate trading strategies based on them, and then execute trades. As a special type of financial trading, it has extremely strict requirements for hardware computing performance.

Pain Point

Ultra-high clock speed: A clock speed of 6GHz or even higher, along with a corresponding cooling solution.Ultra-low latency: Nanosecond-level network latency. Ultra-high throughput: Real-time reading and processing of live trading data. Stability: Stability under the extreme performance utilization of the machine.

Solution

GST's servers specially customized for quantitative trading.

Earnings

High stability: Through long - term and high - pressure pre - delivery burn - in stability tests in specific environments, the stability of trading is guaranteed.

Low latency: A network penetration speed of 100 nanoseconds or even less than 50 nanoseconds.

High performance: The ability to operate under load at an ultra - high clock speed.

Large throughput: Customized NVMe solutions and highly available NVMe solutions.