Benchmark tests on identical 4-core 16GB hardware show PostgreSQL 13 delivers 2x SELECT, 4-5x INSERT, and 5-6x UPDATE throughput over MySQL 8.0, with hot-row updates 8x faster, making PostgreSQL the preferred choice for high-performance workloads despite MySQL's larger ecosystem.
The article analyzes why heterogeneous CPU-accelerator architecture is essential for AI for Science (AI4S), detailing ten frontier applications on Sugon 8000's 100,000-card cluster that demonstrate superior performance in weather forecasting, molecular dynamics, quantum-supercomputing collaboration, and rocket plume simulation, overcoming traditional CPU-only bottlenecks.
This analysis examines how AI workloads are reshaping high-performance networking, covering commodity cluster dominance, Ethernet surpassing InfiniBand in performance share, scale-up interconnects like NVLink and Slingshot, the Ultra Ethernet Consortium's protocol innovations for tail latency, and DPU offloading benchmarks showing 46% Alltoall speedup.
The article analyzes the definition, technical traits, and three classifications of supernodes (SuperNode, SuperPoD, SuperCluster), explains how they boost large‑model training and inference efficiency, and examines the evolution of domestic supernode architectures, interconnects, and ecosystem design.
AI TrainingGPU interconnectHigh-Performance Computing
Huawei’s Atlas 850E wind‑cooled supernode delivers 14.27 PFLOPS, 768 GB HBM, 4 TB/s bandwidth and VCE phase‑change cooling for inference in standard data centers, while the Atlas 950 SuperPoD provides a flagship 1 EFLOPS FP8 training platform with 256 TB unified memory, 1.72 PB/s interconnect, sub‑3 µs RTT, full liquid cooling and scalability up to 8 192 NPU cards.
The article explains what an AI workstation is, its key components such as GPUs or NPUs, large memory, high bandwidth and efficient cooling, how it bridges desktop PCs and rack servers, and classifies workstations by form factor and compute level while referencing a 2026 global AI compute report.
At WAIC, China’s first fully domestic 100,000‑GPU AI supercluster, the Shuguang 8000, was shown to run at full capacity within its first week—processing over 150,000 jobs daily and peaking above 500,000—thanks to intelligent scheduling, high‑density cabinets, proprietary RDMA interconnects, top‑ranked storage and immersion cooling, while offering open access programs for researchers and enterprises.
AI supercomputingHigh-Performance ComputingShuguang 8000
China’s Sugon ParaStor F9000 all‑flash distributed storage system has claimed the top spots on both the full‑node and 10‑node IO500 production benchmarks, delivering up to 247% bandwidth gains and powering large‑scale AI, scientific simulation, and autonomous‑driving workloads across a ten‑thousand‑card supercomputing cluster.
The Ling Sheng supercomputer topped the TOP500 with a 2.19 EFLOPS all‑CPU design, eschewing GPUs, achieving 84.4% scaling efficiency across millions of cores, high energy efficiency, and demonstrating a full‑stack domestic alternative that reshapes the global high‑performance computing landscape.
AI accelerationCPU architectureHigh-Performance Computing
The Huawei AI Data Center Reference Design offers a standardized, integrated, high‑performance compute infrastructure for large‑model training and inference, built on GB/T 50174, featuring modular GPU/HBM servers, 20–50 kW per rack, leaf‑spine 100/200/400 Gbps networking, liquid cooling, redundant power, and intelligent management, with a downloadable package for replication.
The article explains what AI supernodes are, how they differ from traditional server clusters, and why their bus‑level interconnect, global memory pooling, peer‑to‑peer compute and integrated liquid‑cooled racks deliver up to 15× bandwidth gains, 4× inference concurrency, and significant cost reductions, while comparing the approaches of Nvidia, Huawei and other Chinese vendors and outlining future scaling challenges.
The article provides a detailed technical analysis of Huawei's Atlas 900A3 SuperPoD (CM384) supernode wiring architecture, covering cabinet composition, NPU/CPU counts, 400 G optical interconnect design, cable installation practices, scalability to 24 Pods, and the performance benefits for AI workloads.
400G optical interconnectAtlas 900A3Data center wiring
Isaak Freeman, a former MIT PhD student, argues that humanity must embrace AI‑driven brain emulation—estimating that tens of thousands of H100 GPUs could simulate a human brain within a decade, but highlighting massive data‑acquisition, memory‑wall, and connectivity challenges that demand a multi‑decade, multi‑billion‑dollar effort.
A Chinese Academy of Sciences team introduced the MatRIS‑MoE model and the Janus training framework, enabling a 11.5 billion‑parameter universal machine‑learning interatomic potential to be trained on two exascale systems at 1.2 EFLOPS, compressing weeks‑long training into a few hours.
AI for ScienceExascale trainingHigh-Performance Computing
This article explains the fundamentals of Remote Direct Memory Access (RDMA), its low‑latency, zero‑copy and kernel‑bypass mechanisms, programming interfaces, and real‑world applications in data‑center networks, high‑performance computing, and distributed storage, providing developers with practical guidance and code examples.
The article examines how the rise of large‑model AI training reintroduces the need for gang scheduling in Kubernetes, contrasting the rigid resource requirements of HPC‑style workloads with cloud‑native elasticity, and outlines the historical evolution, current implementations, and future directions for achieving more flexible, high‑throughput compute orchestration.
AI TrainingGang SchedulingHigh-Performance Computing
AState is a general‑purpose state data management system for reinforcement‑learning tasks that tackles low IO efficiency, slow weight synchronization, and state‑recovery challenges, achieving sub‑10‑second weight sync for trillion‑parameter models through a three‑layer architecture, zero‑redundancy transfers, and hardware‑aware co‑design, with the code openly available on GitHub.
Awex is a high‑performance Python framework that synchronizes training and inference weights for trillion‑parameter reinforcement‑learning models in seconds, using unified conversion, metadata management, and NCCL/RDMA transfer plans, dramatically reducing RL training latency and supporting diverse parallel strategies.
High-Performance ComputingPythondistributed training
This article explains how AI model training has evolved from single‑GPU workloads to massive distributed training using MPI for CPU‑centric communication and NCCL for GPU‑centric communication, covering their histories, core concepts, programming interfaces, topology discovery, protocol choices, and performance testing on multi‑GPU clusters.
AI distributed trainingGPU communicationHigh-Performance Computing
This article explains the fundamentals of Remote Direct Memory Access (RDMA), its low‑latency, zero‑copy architecture, core principles, programming interfaces, and how it transforms data‑center networking, high‑performance computing, and distributed storage by bypassing the CPU and kernel.
This article systematically examines the demand, technology stack, and industry landscape of large‑scale AI compute clusters, highlighting the limitations of traditional copper interconnects and presenting device‑level and chip‑level optical interconnect solutions—including OCS, pluggable modules, silicon photonics, VCSEL, and micro‑LED—while outlining current challenges and future directions.
The scaleX640 supernode, unveiled at the Wuzhen World Internet Conference, packs 640 AI accelerators into a single rack, delivering unprecedented compute density, energy efficiency, open ecosystem compatibility, and reliability features that enable massive AI model training and inference at scale.
AI hardwareHigh-Performance ComputingOpen architecture
This article examines the challenges of building scalable, low‑latency, and cost‑effective network architectures—such as Clos/Fat‑Tree, Spine‑Leaf, Dragonfly, and Torus—for massive GPU clusters used in training trillion‑parameter AI models, comparing multi‑rail and single‑rail designs and highlighting real‑world implementations from Tencent and Alibaba.
This curated reading list presents must‑read books covering cloud, edge, distributed, high‑performance, parallel, heterogeneous, quantum, and AI computing, offering expert editorial insights, author backgrounds, and publication details to help readers grasp core concepts and advance their technical expertise.
This article explains the fundamentals of high‑performance computing (HPC), covering serial and parallel processing, the roles of CPUs and GPUs, system architectures, FLOPS metrics, current supercomputer capabilities, and the scale needed to reach the next exa‑FLOPS era.
This article compares four cutting‑edge high‑performance processors—Fujitsu A64FX, NVIDIA H100, AMD MI250X, and Intel PonteVecchio—examining their architectures, parallelism strategies, domain‑specific accelerators, supported data types, performance metrics, and power consumption to inform future E‑level computing designs.
This guide explains why Java struggles with high‑performance or data‑intensive workloads, introduces GPU acceleration with CUDA, compares integration options such as JCuda, JNI, and JNA, walks through a practical encryption use case with performance benchmarks, and provides production‑grade best practices for memory, threading, testing, security, and deployment.
This article examines the architectural differences between Scale‑Out and Scale‑Up networking, compares PCIe, NVLink, UALink, Infiniband and RoCE, and explains why high‑bandwidth, low‑latency GPU interconnects like NVLink are essential for modern AI and HPC workloads.
AI accelerationGPU interconnectHigh-Performance Computing
This article introduces four cutting‑edge AI networking technologies—SUE, OISA, ALS, and ETH+—detailing their backgrounds, architectural designs, and performance enhancements that enable ultra‑high bandwidth, low‑latency, and scalable interconnects for modern AI compute clusters.
This article, part of the 2025 AI Network Technology Whitepaper, classifies AI high‑performance networking into Scale‑Up, Scale‑Out, and frontier breakthroughs, then dives deep into NVLink’s evolution, technical features, NVSwitch’s full‑mesh architecture, and the newly opened NVLink Fusion ecosystem.
AI networkingGPU interconnectHigh-Performance Computing
The UPN512 whitepaper details Alibaba Cloud's next‑generation AI infrastructure network, explaining the shift from dense to MoE models, the rise of train‑and‑inference integration, xPU scale‑up challenges, and how high‑radix Ethernet with LPO/NPO optical interconnects delivers ultra‑high bandwidth, low latency, cost‑effective, and reliable large‑scale AI compute clusters.
Huawei’s 2025 Full‑Connection Conference unveiled the Atlas 950 and Atlas 960 SuperPoD supernodes, detailing their massive card counts, unprecedented compute, memory and bandwidth capabilities, and explaining how their full‑stack hardware‑software design dramatically accelerates large‑model AI training and inference.
The article analyzes Nvidia's 2023 Blackwell GPU series and GB200 NVL72 architecture, detailing their advanced 3‑4nm manufacturing, redesigned CUDA cores, next‑gen ray‑tracing and DLSS upgrades, massive compute and memory bandwidth gains, NVLink Gen5 improvements, and the diverse GB200 product configurations for high‑performance AI workloads.
This article explains the logical and physical architecture of AI compute centers, detailing the division into access, security, network, management, out‑of‑band, AI compute cluster, and general compute zones, and describes the four network planes—parameter, sample, business, and management—required for high‑performance AI workloads.
The article explains how AI large‑model training demands massive GPU resources and how carefully designed network architectures—such as Clos/Fat‑Tree, Spine‑Leaf, multi‑rail versus single‑rail connections, Dragonfly, and Torus—impact performance, scalability, cost, and reliability, guiding the selection of optimal data‑center networks.
The CCF‑Ant Research Fund 2025, now open for its first batch, invites global university and institute researchers to apply by August 25 2025 for up to 50 projects spanning data security, hardware‑software co‑design, supercomputing, and artificial intelligence, with detailed topics, eligibility rules, and submission channels provided.
This article outlines the high‑performance networking requirements for AI data center training, compares InfiniBand and RoCE solutions, discusses their advantages in bandwidth, latency, scalability and cost, and provides design guidelines for building scalable, low‑latency, non‑blocking AI‑centric network architectures.
This article traces the evolution of global supercomputing—from early US and Japanese initiatives to Europe’s coordinated investments—and details China’s rapid development of successive supercomputer generations, highlighting landmark systems such as SGI Power Challenge XL, Dawning‑2000, DeepComp series, the “元” platform and the “东方” machine, as well as homegrown high‑performance software like HPSEPS and HPLES.
From powering cutting‑edge 3A games and operating system kernels to accelerating scientific simulations, high‑frequency trading, and embedded IoT devices, C++ remains the go‑to language for high‑performance, low‑level control across diverse domains, thanks to its speed, portability, and fine‑grained memory management.
This article explains the technical definitions, core differences, and practical use cases of Scale‑Up and Scale‑Out networking in AI systems, highlighting how they impact latency, bandwidth, and cost, and illustrates their combined application through NVIDIA's NVL72 supernode case study.
AI infrastructureGPU networkingHigh-Performance Computing
Huawei’s CloudMatrix 384, built from 384 Ascend 910C chips and a multi‑to‑multi topology, delivers up to 300 PFLOP BF16 performance—nearly twice that of Nvidia’s GB200 NVL72—while exposing supply‑chain dependencies on foreign fabs, higher power consumption, and a rapid push to scale China’s domestic semiconductor capabilities.
This article explores how Remote Direct Memory Access (RDMA) technology transforms high‑performance computing, artificial intelligence, and cloud storage by eliminating data copies, bypassing the kernel, and offloading protocols to hardware, while reviewing key metrics, product ecosystems, real‑world use cases, challenges, and future trends.
In March 2024 Nvidia unveiled the Blackwell GPU family and the GB200 NVL72 architecture, featuring 3‑4 nm processes, redesigned CUDA cores, next‑gen ray‑tracing, upgraded DLSS, massive FP16/FP8 compute gains, 8 TB/s memory bandwidth, and NVLink Gen5, while also presenting complex power, cooling, and packaging challenges for large‑scale AI deployments.
The article explains why standard GEMM libraries may fall short for special matrix shapes, introduces NVIDIA’s open‑source CUTLASS library, details its hierarchical tiling architecture, and walks through a complete device‑API example that customizes tile sizes and data layouts to achieve near‑hand‑written kernel performance on modern GPUs.
Volcengine’s vRDMA brings high‑performance, low‑latency RDMA acceleration to cloud VPCs, combining self‑developed congestion control, elastic ENI integration, and compatibility with HPC, AI, and big‑data workloads to deliver up to 320 Gbps bandwidth and microsecond‑level latency.
The article provides an in‑depth technical overview of Remote Direct Memory Access (RDMA), covering its zero‑copy, kernel‑bypass, and protocol‑offload features, hardware and software ecosystems, and its impact on high‑performance computing, artificial intelligence, cloud storage, finance, and edge computing.
This article provides an in‑depth overview of Shenwei chips, covering their development history, core technical advantages such as a self‑designed instruction set and high‑performance computing capabilities, the current product line‑up, and their applications in supercomputing, cloud data centers, security, and embedded systems.
CPU architectureHigh-Performance ComputingServer processors
This article explains how Baidu’s Kunlun supernode, built on high‑density liquid‑cooled cabinets and a modular 1U 4‑card design, breaks traditional 8‑card limits, boosts compute density four‑fold, improves power and cooling efficiency, and provides a scalable foundation for large‑model AI training and inference.
AI infrastructureGPU clusterHigh-Performance Computing
This article provides a comprehensive technical analysis of InfiniBand architecture, its protocol stack, comparison with Ethernet‑based RDMA solutions like RoCE and iWARP, and an overview of Omni‑Path, highlighting performance advantages, design trade‑offs, and practical limitations.
The article explains how large‑scale AI model training and inference require massive data handling, describes the RapidFS storage acceleration cluster deployed on a 30,000‑card Kunlun chip system with hundreds of domestic CPU servers, and presents performance tests showing linear throughput scaling up to over 1 TiB/s, demonstrating the impact of high‑performance storage on compute efficiency.
This article analyses the emergence of AI‑focused storage, detailing its ultra‑high bandwidth, concurrency, scale and low‑latency characteristics, the architectural shift from layered to fused designs, the specific performance and data‑management demands of training and inference, and a three‑phase roadmap for future storage innovations.
AI storageGPU accelerationHigh-Performance Computing
The article provides a detailed technical analysis of Huawei's CloudMatrix 384 AI supernode—its 384 Ascend 910C chips, 300 PFLOP BF16 performance, massive memory and bandwidth, power consumption, scale‑up and scale‑out optical networking, and how it compares to Nvidia's GB200 NVL72 in architecture, cost, and energy efficiency.
The article explains how RapidFS, a near‑compute storage acceleration solution built on BOS object storage, delivers up to 10 TiB/s throughput for massive AI model training, detailing its architecture, deployment on a 30,000‑card Kunlun cluster, and performance test results that show linear scaling from 20 to 70 nodes.
The 18th China Electronics Information Conference will be held in Chengdu from April 17‑21, 2025, featuring the DATA+AI forum that gathers leading academicians and industry experts to discuss data‑AI integration, with detailed speaker biographies, presentation titles, and abstracts covering topics such as large‑model inference, cloud‑edge ultrasound diagnostics, and the future of databases in the AI era.
This article examines NVIDIA's NVSwitch technology, explaining why it was needed, how it builds on NVLink to overcome PCIe bottlenecks, tracing its evolution from Pascal to the third‑generation design, and detailing its architectural features, scalability, full‑duplex bandwidth, non‑blocking communication, and optimized network topologies for high‑performance AI and HPC systems.
AI hardwareGPU interconnectHigh-Performance Computing
As AI models grow to trillion‑parameter scales, training them demands massive GPU clusters whose performance is increasingly limited by network bandwidth; this article examines why traditional PCIe interconnects become bottlenecks and how NVIDIA's NVLink and NVSwitch technologies dramatically improve multi‑GPU communication and overall system efficiency.
France will christen its upcoming 2025 supercomputer after Alice Recoque, a trailblazing 1970s engineer, highlighting both the nation's high‑performance computing ambitions and a symbolic push for gender diversity in a traditionally male‑dominated field.
Alice RecoqueFrench technologyHigh-Performance Computing
Migrating a Java computing layer to Rust yielded dramatic performance gains—30% lower CPU usage, 70% less memory—and greater stability, as the authors explain how Rust’s ownership, borrowing, lifetimes, and concurrency, combined with optimized data handling, FFI integration, Tokio async, Docker deployment, and monitoring, outweigh the steep learning curve and ecosystem gaps.
The 2024 report of Alibaba Cloud's Infrastructure Network team details AI‑driven network breakthroughs, high‑performance protocol stacks, large‑scale monitoring systems, numerous top‑conference paper acceptances, open‑source ecosystem initiatives, and extensive industry outreach, highlighting the evolving AI infra landscape.
AI infrastructureConference PapersHigh-Performance Computing
This article explains RDMA technology, covering its core principles, programming model with Verbs API, various communication modes, and its impact on data‑center networking, high‑performance computing, and distributed storage, highlighting its low‑latency, zero‑copy advantages over traditional TCP/IP.
The article explains how the newly approved national cloud‑HPC standard, co‑created by Alibaba Cloud and the China Electronics Standardization Institute, addresses resource limits, reduces costs, and guides industry adoption across sectors such as automotive, semiconductor design, and weather forecasting.
This article provides a detailed technical analysis of multi‑GPU server architectures, covering component breakdowns, NVSwitch networking, bandwidth calculations, and the differences between NVIDIA A100, A800, and H100 configurations for large‑scale AI workloads.
AI hardwareGPU architectureHigh-Performance Computing
This article breaks down the core components of high‑performance GPU servers—including PCIe switch chips, the evolution of NVLink from version 1.0 to 4.0, NVSwitch architecture, HBM memory tiers, and the nuances of bandwidth units—providing a comprehensive technical foundation for large‑scale model training.
In a keynote at CCF HPC China 2024, Alibaba Cloud’s VP explains how diversified high‑performance computing workloads, elastic cloud resources, and the proprietary CIPU architecture are driving the shift to a data‑plus‑AI era across industries such as automotive, life‑science, and large‑model training.
This article examines the evolution from the OSI and TCP/IP models to RDMA‑based technologies, compares traditional three‑tier and leaf‑spine architectures, analyzes NVIDIA SuperPOD designs, and evaluates Ethernet, InfiniBand, and RoCE switches to guide high‑throughput, low‑latency data‑center networking decisions.
This paper presents the implementation and practice of generative retrieval based on Yuan large model in Tencent Advertising, addressing three key challenges: user intent capture, model alignment in advertising domain, and high-performance platform design under ROI constraints.
This article analyzes Baidu's Baige 4.0 AI infrastructure, detailing its four‑layer architecture, XMAN 5.0 hardware, HPN network, BCCL communication library, and AIAK inference upgrades, and explains how these innovations address large‑model training and inference challenges while boosting performance, utilization, and cost efficiency.
AI infrastructureGPU accelerationHigh-Performance Computing
The report examines the 2023 global HPC market—covering on‑premise servers, cloud services, storage, compute engines, and interconnect technologies—showing total spending of $297 billion, forecasting growth to $437 billion by 2028, and highlighting key hardware trends, cloud adoption rates, and emerging AI‑driven workloads.
The article analyzes NVIDIA's DGX SuperPOD architectures across three GPU generations—H100, GH200, and GB200—detailing their NVLink/NVSwitch topologies, bandwidth calculations, scalability limits, and the practical challenges of constructing 256‑GPU and 576‑GPU supercomputing clusters.
This article traces the evolution of business intelligence platforms from early report‑centric tools to modern AI‑enhanced, search‑driven solutions, detailing the architectural layers, high‑performance data analysis design, multi‑level aggregation, hot‑cold data tiering, and large‑model applications that enable zero‑threshold data consumption and intelligent insights.
Business IntelligenceHigh-Performance Computingartificial intelligence
High Bandwidth Memory (HBM) is a DRAM technology that uses stacked chips, TSV, and micro‑bump interconnects to deliver ultra‑high data rates, lower power consumption, and compact form factor, addressing the bandwidth, latency, power, space, thermal, and complexity challenges of traditional 2D memory in GPUs, AI, HPC, and data‑center workloads.
High‑performance computing demands ultra‑low latency and massive scale, prompting a shift from traditional CLOS designs to alternative topologies such as Fat‑Tree, Dragonfly, and Torus, each offering distinct trade‑offs in bandwidth, scalability, routing complexity, and cost‑effectiveness for modern data‑center and HPC environments.
This article defines distributed storage, presents China's market size and growth rates from 2021 to 2022, breaks down the market into file, block, and object segments, highlights the top adopting industries, and outlines nine key application scenarios driving rapid expansion.
This article analyzes the challenges and solutions for constructing a super‑large GPU training cluster, outlining five fundamental design principles, a four‑layer plus one‑domain architecture, and practical considerations for hardware, networking, and operational reliability in AI workloads.
With AI models growing to billions of parameters, the choice of high‑performance interconnect—InfiniBand or RoCEv2—directly impacts training speed, scalability, latency, and operational complexity, and this article analyzes their architectures, performance metrics, vendor ecosystems, and suitability for large‑scale AI clusters.
This article analyzes the architectural upgrades required for ultra‑large AI clusters, covering single‑GPU performance, super‑node scaling, DPU‑based heterogeneous computing, power‑efficiency, high‑throughput storage, and robust high‑speed networking to support trillion‑parameter model training and inference.
This article provides a comprehensive technical overview of large‑scale GPU server architectures, detailing the component topology of 8‑GPU A100/A800 and H100/H800 nodes, explaining storage network cards, NVSwitch interconnects, bandwidth calculations, and the trade‑offs between RoCEv2 and InfiniBand for AI workloads.
The latest Top500 ranking shows the United States dominating with the two fastest supercomputers, while China, despite having powerful hundred‑exaflop systems, has stopped reporting its machines, reflecting a strategic shift amid the tech cold war between the two nations.
This article provides a comprehensive overview of the core components and terminology of large‑scale GPU computing, covering GPU server architecture, PCIe interconnects, NVLink generations, NVSwitch, high‑bandwidth memory (HBM), and bandwidth unit considerations for AI and HPC workloads.
This article provides an in‑depth technical analysis of RoCE v2, covering its architecture, NIC requirements, and detailed comparisons with InfiniBand across physical layers, protocol stacks, switching, congestion handling, routing, and topology, while also highlighting the UEC alliance’s new transport protocol initiative.
Based on TrendForce data, AI server shipments are projected to grow at a 12.2% CAGR through 2027, while advances in PCIe switching, retiming chips, and high‑speed GPU interconnects such as NVLink and NVSwitch are reshaping the architecture and performance of next‑generation AI compute platforms.
AI serversGPU interconnectHigh-Performance Computing
This article provides a detailed technical breakdown of modern multi‑GPU server nodes, covering component composition, storage network cards, NVSwitch interconnects, bandwidth calculations, and the architectural differences between NVIDIA A100/A800 and H100/H800 configurations for AI training workloads.
This article provides a comprehensive breakdown of high‑performance GPU server infrastructure, covering PCIe generations, NVLink evolution, NVSwitch and NVLink switches, HBM memory technologies, and bandwidth measurement units, helping readers understand the hardware connections and performance considerations essential for large‑scale model training.
The article examines the challenges of ultra‑large‑scale HPC networking, compares traditional CLOS with Fat‑Tree, Dragonfly, and Torus topologies, explains their bandwidth and latency characteristics, presents scalability formulas, and evaluates routing algorithms and practical trade‑offs for each design.
This article examines the key GPU/TPU cluster networking options—NVLink, InfiniBand, RoCE Ethernet, and emerging DDC full‑scheduling fabrics—detailing their latency, loss‑less transmission, congestion control, cost, power, and scalability considerations for large‑scale AI training deployments.
This article presents a comprehensive overview of the challenges, key technologies, and future applications of a unified storage platform built on Alluxio for China's national supercomputing internet, detailing its architecture, data flow strategies, deployment status, and industry use cases across multiple sectors.
This article provides a comprehensive overview of InfiniBand technology, covering its history, architecture, packet format, layer functions, switching mechanisms, and performance advantages over Ethernet, while highlighting its rapid growth and future prospects in HPC environments.
NetEase Cloud Music’s RTA advertising system delivers real‑time, personalized ads at massive scale by using isolated Nginx clusters, layered decoupling, asynchronous Netty/Redis processing, and optimized storage with hash‑based key compression and Protostuff serialization, while supporting automated audience selection and in‑app attribution to boost user acquisition.
The article defines high‑performance computing, presents China's dominant share in the Top500 supercomputer rankings, details recent global and Chinese HPC market growth rates, forecasts market size to 2026, and cites multiple industry reports as sources.
The ACM announced that a paper based on China’s 1.5 exaflops Oceanlite supercomputer has been shortlisted for the 2023 Gordon Bell Prize, highlighting its novel turbulent‑flow code, the SW26010 Pro processor architecture, other global contenders, and geopolitical implications voiced by Jack Dongarra.
ExascaleGordon Bell PrizeHigh-Performance Computing
The Hyperion Research ISC23 HPC Market Update briefing highlights modest 2022‑2023 growth, forecasts global HPC spending reaching $33 billion in 2023 and $52 billion by 2026, outlines ten key 2023 predictions—including AI regulation, cloud‑driven HPC expansion, and emerging DPU/IPU markets—while emphasizing the continuing talent shortage and the strategic importance of AI across high‑performance computing.
The article examines how Alibaba Cloud’s Predictable Network, InfiniBand versus Ethernet trade‑offs, and the HPN high‑performance network design together address the extreme bandwidth, latency, scalability and reliability requirements of modern large‑model AI training workloads in cloud data centers.
This article explains the fundamentals of Direct Memory Access (DMA) and Remote Direct Memory Access (RDMA), compares their data transfer mechanisms with traditional networking, and outlines RDMA's advantages, protocols, ecosystem, and real‑world adoption in high‑performance computing and data centers.
The article explains how AIGC models demand massive GPU compute, why network bandwidth and latency become the critical limiting factors, and how the Distributed Disaggregated Chassis (DDC) architecture addresses these challenges with scalable, high‑throughput networking solutions.
This article explains the principles of DMA and RDMA, compares RDMA protocols such as InfiniBand, RoCE, and iWARP, outlines their performance advantages, and reviews the key standards bodies, open‑source communities, hardware vendors, and real‑world adoption in high‑performance data centers.
The article details Baidu's development of a massive high‑performance GPU/IB cluster, its architectural design, the challenges of training trillion‑parameter models, and how the integrated AI stack—spanning hardware, framework, and resource management—overcomes compute, memory, and communication bottlenecks to accelerate large‑model training.
Tencent Cloud's new HCC high‑performance computing cluster triples previous generation performance with 3.2 TB/s server bandwidth, Xingsha servers and NVIDIA H800 GPUs delivering up to 1979 TFlops, while its Xingmai 3.2 T ETH RDMA network, TB‑level storage via COS + GooseFS, and multi‑form access (bare metal, cloud servers, containers, functions) enable efficient large‑model training.