DUBAI,UAE,Oct. 16,2024 -- In the era ofAIGC,intelligent computing power has become a core engine driving the development of the digital economy. On October 16th,at the GITEX Global 2024 in Dubai,H3C launched significant new intelligent computing products,which include the general computing flagship product,H3C UniServer R3900/R4900/R4700/R4300 G7,the diverse heterogeneous computing platforms,H3C UniServer R5300/5500 G7,and the next-generation AI storage Polaris series. With the concept of "Evolution × AI," these innovations aim to achieve intelligent further evolution of computing products,accelerating the prosperity and development of the AI industry.
Ray Xu,Senior Vice President of H3C and President of Cloud,Compute,and Storage Product Line,delivered a keynote speech of “H3C Modernized AI Infrastructure,Lighting Up Your Business Innovation”,while launching H3C UniServer G7 series at H3C NAVIGATE Global Summit 2024 during GITEX Global.
General Computing Flagship: Triple Advantages for Robust Computing Support
The implementation of large-scale language models today requires powerful computing power to support training and inference processes. H3C has launched the universal computing flagship H3C UniServer R4900/R4700 G7,featuring three key advantages: high-efficient computing power,agile deployment,and multiple layers of protection. These strengths collectively enhance the performance of general,scientific,and AI computing.
- High-efficient Computing Power
The server supports 2 Intel® Xeon® processors with up to 288 cores,representing a 125% improvement over the previous generation. It includes support for 32 DDR5 6400MT/s memory modules,offering a 14% increase in memory speed. Additionally,within a 2U space,it can support up to 6 dual-width GPUs,delivering robust heterogeneous computing power.
- Agile Deployment
The server supports various storage interfaces,including E3.S,E1.S,U.2,SAS,and SATA,allowing flexible media selection based on data types. It offers up to 21 PCIe 5.0 interfaces to meet the most demanding business needs. Additionally,it supports PCIe 5.0 standard slots,dedicated Mezz slots,and OCP 3.0 slots for flexible business expansion.
- Multiple Layers of Protection
The system supports a full lifecycle fault diagnosis system with an accuracy rate of 96%. It features intelligent memory repair technology,reducing unexpected downtime by 40%. Additionally,it supports secure encryption for password certificate storage and secure transmission of sensitive information.
Heterogeneous Computing Platforms: Continuous Innovation Accelerates AI Computing
The diversity and complexity of AI large model and applications require diversified computing power. H3C's newly launched heterogeneous computing platform,the H3C UniServer R5300/5500 series,achieves efficient acceleration of GPU cluster performance through OAM interconnection standardization. It also breaks the memory wall by implementing memory resource pooling based on CXL 2.0 technology.
The R5300 series supports over 95% of the CUDA operator library,making underlying adaptations more efficient and agile,and enabling seamless ecosystem migration. It's also compatible with the latest OCP OAI open hardware standards,promoting computing power openness and evolution. Through an open ecosystem and diverse computing capabilities,it builds a sustainably evolving intelligent computing capacity that flexibly meets the varying computational demands of different AI workloads.
Next-generation AI Storage: Born for AI
Data storage is a crucial component in the collection,preprocessing,training,and inference of AI large model. It determines the capacity of stored data,efficiency of data reading during training and inference,and data reliability. The H3C next-generation AI storage,the Polaris X20000 series,completely reconstructs the storage software stack through collaborative software and hardware innovation. This provides the super performance and extreme stability needed for AI large model training while further simplifying the AI data storage infrastructure.
- Performance Acceleration: The Polaris X20000 series significantly reduces ineffective I/O and optimizes the I/O path through a comprehensive reconstruction of the storage software stack,achieving outstanding performance of 80GB/s bandwidth and over 1 million IOPS per node. The large-scale cluster performance scales near linearly,meeting the high-performance demands of intelligent computing applications for massive small file concurrency and hundreds of TB-level bandwidth requirements.
- Architecture Simplification: The Polaris X20000 series addresses diverse data processing in AI intelligent computing by seamlessly integrating underlying protocols without performance loss. A single logical resource pool supports file,object,and big data protocols simultaneously,allowing cross-protocol access without data format conversion. This enables data to remain in place from collection to model training and distribution throughout the entire AI large model training process.
- Ultimate Stability: The Polaris X20000 series supports multi-tenant data and resource isolation,ensuring stable operations in multi-tenant environments without interference. Data resources are isolated to further ensure security. Business upgrade times are significantly reduced,with both planned and unplanned deployments,upgrades,and expansions being seamless to upper-layer applications. Faults are reported within milliseconds,with detection times accelerated by 10 times,enabling second-level switching and uninterrupted training tasks.
In the AI era,all industries are urgently demanding for high-performance,large-capacity,and low-latency storage. The H3C Next-generation AI Storage Polaris Series,offering an efficient AI storage solution through performance acceleration,intelligent O&M,data pooling,intelligent acceleration,compute-storage integration,and smart assistants,helping users meet the data storage challenges of the AI era.
- Performance Acceleration: Accelerates the transition to SSD all-flash storage to meet the demands for millions of small file concurrency and TB/s bandwidth throughput in intelligent computing.
- Intelligent O&M: Enhances storage reliability and availability with AI technology.
- Data Pooling: Enables seamless data flow across clusters,clouds,and media without relocation.
- Intelligent Acceleration: Utilizes deep learning to profile businesses and users,improving cache hit rates and performance.
- Compute-storage Integration: Develop an end-to-end unified solution with direct GPU data access; support operator offloading to enhance intelligent computing application efficiency.
Computing power is a new quality productive force in the digital economy era and a foundational element for empowering industrial digital transformation. In the future,H3C will uphold "Dedication,For A Smarter Future",continually deepening the "AI in ALL" technology strategy. By refining intelligent computing products,solutions,and services,we aim to empower and energize the intelligent upgrades across various industries,supporting the development of new productive forces.