New H3C releases H3C UniServer G7 server

Under the rapid development of AGI, modern AI infrastructure has become the key to driving the advancement of the intelligent industry. Recently, New H3C Group, a subsidiary of Tsinghua Unigroup, launched the H3C UniServer G7 series of new computing power platforms, which rely on an AI oriented evolutionary architecture, an endogenous AI operating system and management center, and a diversified computing power system embracing AGI to provide efficient and powerful computing power supply, laying the foundation for the vigorous development of the AIGC industry and helping industry customers move towards a new intelligent future.

Evolutionary architecture for AI
Creating a flexible and intelligent computing power supply

Based on innovative modular design and efficient heat dissipation solutions, the H3C UniServer G7 series provides users with flexible, efficient, and green computing power support, fully meeting different scenarios and business needs.

Modular design decoupling system, on-demand upgrade: GPU subsystem, CPU subsystem, heat dissipation subsystem, etc. all adopt modular design, users can flexibly upgrade according to business needs, ensuring that computing resources are in the optimal state.
Efficient system perfectly solves high power consumption challenges:

Intelligent flow air duct, independent heat dissipation channels for CPU, GPU, and DPU, providing the ultimate fan/machine power ratio;
Intelligent cooling and heat dissipation, increasing the number of heat pipes to 10 and improving heat transfer performance by 8%;
Intelligent static mesh cotton effectively reduces the impact of fan vibration on hard disk performance, with a maximum performance improvement of 22%;
Intelligent rotary fan, fluid dynamics design, eliminates heat dissipation blind spots, increases fan efficiency by 10%;
Smart sensing, with 63 built-in 3D ocean sensors on the server, comprehensively monitors temperature, redundantly monitors key components, forms a thermal map, and provides decision-making basis for heat dissipation strategies.

ALL in GREEN full stack liquid cooling, multi-dimensional evolution:Realize ultimate liquid cooling and high-density computing; Adopting 100% liquid cooling and exclusive Ruiliu technology to reduce the overall energy consumption of the machine; The design of two-phase liquid cooling and condensation reflux greatly improves the heat dissipation efficiency and achieves green energy saving; Achieving zero carbon engines through box immersion and oil immersion technologies to support sustainable development.
Green design reduces carbon emissions and demonstrates corporate social responsibility：Focusing on green design, monitoring the entire lifecycle of products, identifying the main sources of carbon emissions and improvement potential; By replacing with green materials, 1593 tons of equivalent CO2 can be reduced, and a full lifecycle carbon footprint analysis can be achieved to continuously optimize and ensure minimal environmental impact, contributing to sustainable development goals.
Standardize the interface of intelligent computing power, reshape the architecture, and create an open ecosystem: H3C actively hosted and participated in the development of OAI 2.0 standards. The new generation of AI servers is compatible with OAI 2.0 standards and has completed joint design with more than ten AI chip manufacturers to jointly create an open ecosystem.
The industry’s first E3. S CXL2.0 memory pool reshapes data center architecture：H3C UniServer M7200 is the industry’s first CXL2.0 memory pooling solution based on the E3. S form, which achieves 5.6TB of E3. S memory expansion and resource pooling on a single machine. While expanding memory capacity, the single machine can provide up to 512GB of memory bandwidth expansion per second. M7200 supports flexible expansion of both E3. S and AIC forms, and can monitor faults in real-time to ensure high availability. Distributed data scenarios, through memory sharing, can reduce data copying and increase interactivity by 10 times; In the Internet and public cloud scenarios, memory pooling reduces memory costs by 30%.

The operating system and management center of endogenous AI
Realize efficient and stable system operation

As the core of a data center, the stability and efficiency of servers directly affect the operation of the entire data center. The H3C UniServer G7 series is equipped with an endogenous AI operating system and a newly upgraded intelligent management center, ensuring more stable and efficient system operation.

Operating System Panning NingOS: Accumulating over 10 years of experience in the ICT field, New H3C has formed software and hardware collaborative innovation in reliability, performance, and operation, providing customers with the best business experience. In terms of reliability, the Panning operating system has been enhanced for different CPU platforms in terms of kernel security, RAS features, and support for domestic hardware and drivers; In terms of performance, breaking through the key technology of multi-core parallelism, optimizing processor process scheduling and memory scheduling, and maximizing system performance; In terms of maintenance, in addition to conventional vulnerability management, online problem localization, and hot patch upgrades, the Panning operating system can automatically perceive business and form topology, while also being able to perform fault diagnosis, focusing on providing an easy-to-use experience in business management and maintainability.

The five major intelligences of the intelligent management center have been upgraded again: through the comprehensive upgrade of intelligent deployment, intelligent optimization, intelligent energy saving, intelligent diagnosis, and intelligent retirement capabilities, while effectively improving operational efficiency, system performance, and diagnostic accuracy, energy consumption has been significantly reduced.

Embrace the Multi dimensional Computing System of AGI
Fully adapted to AI applications

In the era of AGI, New H3C has comprehensively upgraded multiple intelligent computing infrastructures. With strong compatibility and a diverse ecosystem, it comprehensively adapts to diverse AI scenarios and builds the foundation of computing power for future intelligent applications.

Establishing a new standard for universal servers: As the only 2U server in the industry that supports 6 dual width GPUs, H3C UniServer R4900 G7 supports up to 288 cores, up to 40 E3. S SSDs, and has a massive storage capacity of 1.2PB; H3C UniServer R3900 G7 supports up to 128 cores and significantly increases bandwidth, meeting the high-performance requirements of AI applications.
AI training and promotion integrated engine: H3C UniServer R5300 G7 supports 8 dual width GPUs and 16 single width GPUs, supports 5 GPU topologies, and GPGPU architecture is compatible with 200+mainstream AI models, which can significantly improve the training and inference performance of large models and meet the needs of AI in all scenarios; Equipped with 400TB of ultra large storage space, enhancing reasoning and application capabilities.
AI computing flagship: H3C UniServer R5500 G7 supports training of trillion level parameter large models, seamless migration of 15+mainstream open-source large models, and zero code changes; Realize comprehensive improvements in video memory capacity, scale up interconnect bandwidth, FP16 computing power, and large model training performance.
Targeting trillion dollar model supernode clusters: H3C UniPoD supernodes support southbound interconnection of 32 OAM GPUs, increasing the scale up interconnection scale by 300% and significantly reducing the total training time; Adopting the industry’s only PCIe protocol solution, significantly reducing latency and providing strong support for training trillion parameter models.

Leave a ReplyCancel Reply