Entering 2024, the AIGC boom continues to surge, computing power is no longer only related to a few people, but penetrates into all aspects of production and life, becoming an extension of human capabilities and a key force to promote social change, and computing infrastructure has gradually become a "must fight" for a new round of scientific and technological revolution and industrial transformation. In order to meet the growing demand for computing power in the AI era, New H3C Group, a subsidiary of Tsinghua Unigroup, has released a series of new intelligent computing products such as Alpha Computing Platform 3.0, multi-heterogeneous AI servers, computing-network integrated intelligent edge servers, and full-stack liquid cooling solutions, with the concept of "computing power × AI", to realize the intelligent re-evolution of computing products, fully support the computing power needs in different scenarios, and build a diverse, open, and efficient computing power base for the digital intelligence transformation of hundreds of industries.
Computing power scheduling × AI to achieve the optimal utilization of computing resources
With the continuous expansion of AI application fields, efficient management of computing clusters and reasonable scheduling of computing resources are essential to improve the efficiency of AI task execution and ensure system stability. However, large-scale AI cluster management often faces many challenges, such as difficult management of diverse computing, difficulty in multi-department collaboration, high cost of environment construction, and difficulty in detecting faults.
The newly upgraded Alpha computing platform 3.0 has three core functions: intelligent scheduling of multiple computing power, AI full-process development services, and flexible operation and management, so as to ensure the sustainable development of computing power construction with the optimal utilization of computing power resources.
In terms of intelligent scheduling of multiple computing power, Alpha supports the unified scheduling of heterogeneous computing power of multiple CPUs and GPUs, and the computing power and video memory can be segmented and scheduled on demand at a fine-grained rate of 1% and MB, which significantly improves resource utilization, and supports the computing power scheduling and computing power splitting of Wanka clusters to achieve efficient operation on and off the cloud. In terms ×of operation and management, Aofei supports multi-tenant capabilities in cloud and off-cloud scenarios, and provides prepaid and postpaid models to meet the needs of different tenants, excellent resource management and monitoring capabilities, which can achieve refined resource allocation and ensure the stable and efficient operation of the system.
Heterogeneous computing power × AI are diversified and open to meet the needs of different AI workloads
The diversification and complexity of AI models and application scenarios require diversified computing power supply to promote the rapid development of heterogeneous computing resources. H3C UniServer R5330 G7 and H3C UniServer R5500 G7, the newly released multi-heterogeneous computing platforms of H3C Group, have multiple advantages such as multiple heterogeneous computing platforms, such as multiple heterogeneous computing power, openness and inclusiveness, flexible architecture, and super scalability, providing strong computing power support for different AI workloads.
- Intelligent computing power flagship: H3C UniServer R5500 G7 is designed for large-scale AI training, supports large-scale model training with hundreds of billions of parameters, adopts an advanced modular design of system decoupling, allows users to flexibly choose GPU computing modules, and is compatible with the latest OCP OAI open hardware standards to achieve open and evolvable computing power. The R5500 G7 is equipped with 2 latest processors and a new generation of OAM 8-GPU modules, which has the ultimate computing power of security and power.
- Independent computing engine: H3C UniServer R5330 G7 is equipped with two latest processors + GPU heterogeneous computing power, supports hybrid parallel training of large models, and supports more than 95% of the CUDA operator library, making the underlying adaptation more efficient and convenient, and realizing seamless ecological migration.
Edge computing × AI to address complex and ever-changing edge scenarios
As the convergence of IT, CT, and OT in various industries, edge computing effectively fills the computing power vacuum between the cloud and the device, and is a new strategic fulcrum for leveraging the intelligent upgrading of the industry. In order to meet the needs of complex and changeable edge scenarios, H3C Group has independently developed a new generation of modular edge server H3C UniServer E3300 G6 based on the ODCC OTII-E technical specifications, which has the triple advantages of computing-network integration, edge intelligent computing, and green and efficient, and can easily meet the needs of different edge scenarios such as 5G, edge cloud, and edge AI.
- Computing-network convergence: Compared with traditional edge servers, the E3300 G6 adopts a modular design, and through the "all-in-one" architecture innovation of switches and servers, users can flexibly select different computing and switching nodes to realize the converged deployment of edge point infrastructure, greatly simplifying the complexity of deployment and O&M, and reducing TCO by 20%-30%.
- Edge intelligent computing: The E3300 G6 supports GPU acceleration, providing powerful computing power support for edge inference scenarios, and supports cloud-edge collaboration and unified management and scheduling to achieve more efficient data processing and analysis.
- Green and efficient: The E3300 G6 uses a M_CRPS titanium power supply, which can achieve up to 50% energy savings, smaller body and higher efficiency.
Green and low-carbon × AI breaks through the bottleneck of computing power development and energy efficiency
With the surge in demand for computing power and the in-depth promotion of the "dual carbon" strategy, the pressure of energy saving and consumption reduction of high-density and high-power intelligent computing centers is imminent, and liquid cooling technology has become an important choice for data center construction and optimization due to its outstanding advantages in energy efficiency and heat dissipation. As one of the earliest enterprises in China to invest in liquid cooling R&D, New H3C Group actively promotes the innovation and application of liquid cooling technology, creates a multi-dimensional and evolved full-stack liquid cooling solution, and helps data centers break through the bottleneck of energy efficiency.
New H3C Group's "ALL in GREEN" full-stack liquid cooling solution has covered the two major technical routes of cold plate and immersion, and is accelerating iteration. H3C is committed to building a full cold plate system, and has launched a series of solutions such as hard disk cold plate, CPU+DIMM cold plate, GPU card cold plate, etc., with a cold plate coverage rate of 100% and full liquid cooling of heat-sensitive components above 0.3W, and a liquid-cooled RDHX back-door cooling solution that uses liquid-cooled backdoor instead of air-conditioning refrigeration without air-cooling compensation to achieve 100% In order to further solve the problems of equipment room utilization, liquid cooling heat dissipation ratio and coolant usage, H3C has launched the industry's first rack box immersion server by combining the advantages of cold plate and immersion liquid cooling technology, which is suitable for traditional air-cooled equipment rooms, increasing space utilization by 70%, and the unique "box immersion" design helps increase the proportion of liquid cooling heat dissipation to 100%< In addition, the phase-change immersion solution, combined with the excellent condensation and reflow design, can achieve a 70% improvement in heat dissipation efficiency and recycle coolant in the chassis and cabinets, promoting the evolution of data centers to zero-carbon and sustainable goals.
Computing power is productivity, and the continuous evolution of computing power will promote the acceleration of the AGI era. Facing the slowly unfolding picture of the intelligent world, New H3C Group will adhere to the concept of "intensive cultivation and pragmatism, endowing wisdom to the times", continue to make breakthroughs in key technology fields such as computing power scheduling, heterogeneous computing, edge computing, green and low-carbon, and build a strong computing engine that continues to evolve, so that the "digital artery" of high-quality development will continue to flow.
