{"id":4065,"date":"2023-03-27T06:26:49","date_gmt":"2023-03-27T06:26:49","guid":{"rendered":"https:\/\/www.fibermall.com\/blog\/?p=4065"},"modified":"2024-06-14T03:17:14","modified_gmt":"2024-06-14T03:17:14","slug":"fibermall-delivers-hpc-networking-solutions-for-aigc","status":"publish","type":"post","link":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm","title":{"rendered":"FiberMall Delivers HPC Networking Solutions for AIGC"},"content":{"rendered":"\n<p>AIGC (AI-Generated Content) has been developing rapidly recently, and the iteration rate is exponentially exploding. Among them, the launch of GPT-4 and ERNIE Bot has drawn great attention to its business value and application scenarios. With the development of AIGC, the scale of training model parameters has gone from hundreds of billions to trillions of levels, and the scale of underlying GPU support has also reached trillion card levels. The resulting network scale keeps increasing, and the communication between network nodes faces increasing challenges. In this context, how to improve the AI server computing power and networking communication capability and take into account the cost has become one of the important research directions in the current AI field.<\/p>\n\n\n\n<p>FiberMall launched the industry&#8217;s advanced &#8220;Smart Speed&#8221; DDC (Distributed Disaggregated Chassis) high-performance network solution to address the relationship between AIGC computing power, GPU utilization and network, as well as the challenges faced by mainstream HPC networking, to help AIGC business computing power soar.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png\" alt=\"Diagram of FiberMall's DDC product connection\" class=\"wp-image-4066\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png 1055w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection-300x165.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection-1024x565.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection-768x424.png 768w\" sizes=\"(max-width: 1055px) 100vw, 1055px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Diagram of FiberMall&#8217;s DDC product connection<\/b><\/strong><\/span><\/p>\n\n\n\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_76 ez-toc-wrap-left counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Relationship_Between_AIGC_Arithmetic_Power_GPU_Utilization_and_Network\" >Relationship Between AIGC Arithmetic Power, GPU Utilization and Network<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Relationship_Between_Training_Time_and_GPU_Utilization_of_ChatGPT\" >Relationship Between Training Time and GPU Utilization of ChatGPT<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Relationship_Between_Network_Efficiency_and_GPU_Utilization\" >Relationship Between Network Efficiency and GPU Utilization<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#The_Challenges_of_Mainstream_HPC_Networking\" >The Challenges of Mainstream HPC Networking<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#IB_Networking_is_Expensive_and_Closed\" >IB Networking is Expensive and Closed<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#PFC_and_ECN_may_Trigger_a_Speed_Drop\" >PFC and ECN may Trigger a Speed Drop<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Unbalanced_ECMP_may_Cause_Congestion\" >Unbalanced ECMP may Cause Congestion<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Challenges_of_AI_Clustering_with_Box_Switches\" >Challenges of AI Clustering with Box Switches<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#DDC_Products_Born_to_Support_AIGC\" >DDC Products Born to Support AIGC<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#DDC_supports_ultra-large_scale_deployment\" >DDC supports ultra-large scale deployment<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#More_balanced_load_based_on_VOQCell_mechanism_lower_packet_loss_rate\" >More balanced load based on VOQ+Cell mechanism, lower packet loss rate<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#No_Deadlock_under_PFC_Single-hop_Deployment\" >No Deadlock under PFC Single-hop Deployment<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-13\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#NCC-free_design_with_distributed_OS_to_enhance_reliability\" >NCC-free design with distributed OS to enhance reliability<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-14\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Test_Comparison_Results\" >Test Comparison Results<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-15\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#FiberMall_Equipment_Introduction\" >FiberMall Equipment Introduction<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-16\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Conclusion\" >Conclusion<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-17\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\/#Related_Posts\" >Related Posts<\/a><\/li><\/ul><\/li><\/ul><\/nav><\/div>\n<h2 class=\"wp-block-heading\" id=\"h-relationship-between-aigc-arithmetic-power-gpu-utilization-and-network\"><span class=\"ez-toc-section\" id=\"Relationship_Between_AIGC_Arithmetic_Power_GPU_Utilization_and_Network\"><\/span><span style=\"font-size: 14pt;\"><strong><b>Relationship Between AIGC Arithmetic Power, GPU Utilization and Network<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-relationship-between-training-time-and-gpu-utilization-of-chatgpt\"><span class=\"ez-toc-section\" id=\"Relationship_Between_Training_Time_and_GPU_Utilization_of_ChatGPT\"><\/span><span style=\"font-size: 12pt;\"><strong><b>Relationship Between Training Time and GPU Utilization of ChatGPT<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Taking ChatGPT as an example, in terms of arithmetic power, the total arithmetic power consumption for training on Microsoft Azure AI supercomputing infrastructure (a high-bandwidth cluster of 10,000 V&nbsp;100 GPUs) is about 3,640 PF-days (one thousand trillion calculations per second, running for 3,640 days), here is a formula to convert how long it takes to train 10,000 V&nbsp;100s.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ChatGPT-computing-power-and-training-schedule.png\" alt=\"ChatGPT computing power and training schedule\" class=\"wp-image-4067\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ChatGPT-computing-power-and-training-schedule.png 1080w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ChatGPT-computing-power-and-training-schedule-300x88.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ChatGPT-computing-power-and-training-schedule-1024x300.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ChatGPT-computing-power-and-training-schedule-768x225.png 768w\" sizes=\"(max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>ChatGPT computing power and training schedule<\/b><\/strong><\/span><\/p>\n\n\n\n<p>Note: ChatGPT computing power requirements are available online and are provided here for reference only.&nbsp;In the article &#8220;AI and Compute&#8221;, OpenAI assumes a utilization rate of 33%, while a group of researchers at NVIDIA, Stanford and Microsoft have achieved utilization rates of 44% to 52% for training large language models on distributed systems.<\/p>\n\n\n\n<p>It can be seen that the main factors affecting the training time of a model are the GPU utilization, and the GPU cluster processing power. These key metrics are in turn closely related to network efficiency. Network efficiency is an important factor that affects GPU utilization in AI clusters. In AI clusters, GPUs are usually the core resource of compute nodes because they can efficiently handle large-scale deep learning tasks. However, GPU utilization is influenced by several factors, among which network efficiency is a key factor.<\/p>\n\n\n\n<figure class=\"wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"wp-block-embed__wrapper\">\n<div class=\"ast-oembed-container \" style=\"height: 100%;\"><iframe title=\"Network Planning in HPC Solutions | FiberMall\" width=\"500\" height=\"281\" src=\"https:\/\/www.youtube.com\/embed\/Rtf1zcFrvRc?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe><\/div>\n<\/div><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-relationship-between-network-efficiency-and-gpu-utilization\"><span class=\"ez-toc-section\" id=\"Relationship_Between_Network_Efficiency_and_GPU_Utilization\"><\/span><span style=\"font-size: 12pt;\"><strong><b>Relationship Between Network Efficiency and GPU Utilization<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>The network plays a critical role in AI training, and AI clusters usually consist of multiple compute and storage nodes, which need to communicate and exchange data frequently. If the network is inefficient, the communication between these nodes will become slow, which will directly affect the computing power of the AI cluster.<\/p>\n\n\n\n<p>Inefficient networks can lead to the following problems, which can reduce GPU utilization.<\/p>\n\n\n\n<p><strong><b>Increased data transfer time<\/b><\/strong>: In an inefficient network, the data transfer time will increase. GPU utilization will decrease when GPUs need to wait for data transfer to complete before they can perform calculations.<\/p>\n\n\n\n<p><strong><b>Network bandwidth bottleneck<\/b><\/strong>: In an AI cluster, GPUs usually need to exchange data with other compute nodes frequently. If the network bandwidth is insufficient, GPUs will not get enough data for computation, resulting in lower GPU utilization.<\/p>\n\n\n\n<p><strong><b>Unbalanced task scheduling<\/b><\/strong>: In an inefficient network, tasks may be assigned to different compute nodes from the GPUs. This may cause the GPU to wait idle when a large amount of data transfer is required, thus reducing GPU utilization.<\/p>\n\n\n\n<p>In order to improve GPU utilization, network efficiency needs to be optimized. This can be achieved by using faster network techniques, optimizing the network topology, and rationalizing the bandwidth allocation. In the training model, the parallelism of distributed training: data parallelism, tensor parallelism and flow parallelism determines the communication model between the data processed by GPUs. The efficiency of communication between models is influenced by several factors:<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Factors-affecting-communication.png\" alt=\"Factors affecting communication\" class=\"wp-image-4068\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Factors-affecting-communication.png 1080w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Factors-affecting-communication-300x181.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Factors-affecting-communication-1024x617.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Factors-affecting-communication-768x463.png 768w\" sizes=\"(max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Factors affecting communication<\/b><\/strong><\/span><\/p>\n\n\n\n<p>Among them, bandwidth and device forwarding latency are limited by hardware, end-processing latency is influenced by the choice of technology (TCP or RDMA), <span style=\"color: #0000ff;\"><a style=\"color: #0000ff;\" href=\"https:\/\/www.fibermall.com\/blog\/what-is-rdma.htm\" target=\"_blank\" rel=\"noopener\">RDMA<\/a><\/span>&nbsp;will be lower, and queuing and retransmission are influenced by network optimization and technology choice.<\/p>\n\n\n\n<p>Based on the quantitative model: GPU utilization = iterative computation time within GPU \/ (iterative computation time within GPU + overall network communication time) the following conclusions are drawn:<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Graph-of-bandwidth-throughput-and-GPU-utilization.png\" alt=\"Graph of bandwidth throughput and GPU utilization\" class=\"wp-image-4069\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Graph-of-bandwidth-throughput-and-GPU-utilization.png 1080w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Graph-of-bandwidth-throughput-and-GPU-utilization-300x101.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Graph-of-bandwidth-throughput-and-GPU-utilization-1024x343.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Graph-of-bandwidth-throughput-and-GPU-utilization-768x257.png 768w\" sizes=\"(max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Graph of bandwidth throughput and GPU utilization&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; <\/b><\/strong><\/span><span style=\"font-size: 10pt;\"><strong><b>Graph of dynamic latency and GPU utilization<\/b><\/strong><\/span><\/p>\n\n\n\n<p>It can be seen that network bandwidth throughput and dynamic latency (congestion\/packet loss) have a significant impact on GPU utilization.<\/p>\n\n\n\n<p>Based on the composition of the total communication latency:<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Composition-of-total-communication-latency.png\" alt=\"Composition of total communication latency\" class=\"wp-image-4070\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Composition-of-total-communication-latency.png 1312w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Composition-of-total-communication-latency-300x119.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Composition-of-total-communication-latency-1024x407.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Composition-of-total-communication-latency-768x305.png 768w\" sizes=\"(max-width: 1312px) 100vw, 1312px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Composition of total communication latency<\/b><\/strong><\/span><\/p>\n\n\n\n<p>Static latency has a smaller impact, so it is more important to focus on how to reduce dynamic latency, which can effectively improve the utilization of GPU to achieve the goal of improving computing power.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-the-challenges-of-mainstream-hpc-networking\"><span class=\"ez-toc-section\" id=\"The_Challenges_of_Mainstream_HPC_Networking\"><\/span><span style=\"font-size: 14pt;\"><strong><b>The Challenges of Mainstream HPC Networking<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-ib-networking-is-expensive-and-closed\"><span class=\"ez-toc-section\" id=\"IB_Networking_is_Expensive_and_Closed\"><\/span><span style=\"font-size: 12pt;\"><strong><b>IB Networking is Expensive and Closed<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p><span style=\"color: #333300;\">Infiniband<\/span> networking is the most effective solution for current high-performance networks, using ultra-high bandwidth and Credit-based mechanisms to ensure no congestion and ultra-low latency, but it is also the most expensive solution. It is also the most expensive solution. It is several times more expensive than traditional Ethernet networking with the same bandwidth. At the same time, <a href=\"https:\/\/www.fibermall.com\/collection\/infiniband.htm\" target=\"_blank\" rel=\"noopener\"><span style=\"color: #0000ff;\">Infiniband<\/span><\/a> technology is closed, and there is only one mature supplier in the industry, which makes it impossible for end users to achieve a second source of supply.<\/p>\n\n\n\n<p>Therefore, most users in the industry will choose the traditional Ethernet networking solution.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-pfc-and-ecn-may-trigger-a-speed-drop\"><span class=\"ez-toc-section\" id=\"PFC_and_ECN_may_Trigger_a_Speed_Drop\"><\/span><span style=\"font-size: 12pt;\">PFC and ECN may Trigger a Speed Drop<\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>The current mainstream networking solution for high-performance networks is based on RoCE v2 to build RDMA-enabled networks. Two important collocation technologies are PFC and ECN, both of which are created to avoid congestion in the link.<\/p>\n\n\n\n<p>Under multi-stage PFC networking, it will target the switch ingress congestion and backpressure to the source server to suspend transmission step by step to relieve network congestion and avoid packet loss; however, this solution may face the risk of PFC Deadlock causing RDMA traffic to stop forwarding under multi-stage networking.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-PFC-working-mechanism-1.png\" alt=\"Schematic diagram of PFC working mechanism\" class=\"wp-image-4072\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-PFC-working-mechanism-1.png 1080w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-PFC-working-mechanism-1-300x173.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-PFC-working-mechanism-1-1024x591.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-PFC-working-mechanism-1-768x443.png 768w\" sizes=\"(max-width: 1080px) 100vw, 1080px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Schematic diagram of PFC working mechanism<\/b><\/strong><\/span><\/p>\n\n\n\n<p>While ECN generates a RoCEv2 CNP packet directly to notify the source of speed reduction based on destination-side awareness of congestion at the switch egress, the source server receives the CNP message and precisely reduces the sending rate of the corresponding QP to relieve congestion while avoiding indiscriminate speed reduction.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-ECN.png\" alt=\"Schematic diagram of ECN\" class=\"wp-image-4073\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-ECN.png 797w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-ECN-300x114.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Schematic-diagram-of-ECN-768x292.png 768w\" sizes=\"(max-width: 797px) 100vw, 797px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Schematic diagram of ECN<\/b><\/strong><\/span><\/p>\n\n\n\n<p>Both of these technologies are designed to solve congestion, but may be frequently triggered by possible congestion in the network. Eventually, the source end will pause or slow down the transmission speed, and the communication bandwidth will be reduced. The GPU utilization rate is greatly affected, which reduces the computing power of the entire high-performance network.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-unbalanced-ecmp-may-cause-congestion\"><span class=\"ez-toc-section\" id=\"Unbalanced_ECMP_may_Cause_Congestion\"><\/span><span style=\"font-size: 12pt;\"><strong><b>Unbalanced ECMP may Cause Congestion<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>In AI training calculation, there are two main models, All-Reduce and All-to-All, both of which require frequent communication from one GPU to multiple GPUs.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/AI-training-calculation-models.png\" alt=\"AI training calculation models\" class=\"wp-image-4074\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/AI-training-calculation-models.png 1268w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/AI-training-calculation-models-300x128.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/AI-training-calculation-models-1024x436.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/AI-training-calculation-models-768x327.png 768w\" sizes=\"(max-width: 1268px) 100vw, 1268px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><strong><span style=\"font-size: 10pt;\">AI training calculation models<\/span><\/strong><\/p>\n\n\n\n<p>In traditional networking, ToR and Leaf devices adopt the routing +ECMP networking mode. ECMP performs hash load routing based on streams. In an extreme case, one ECMP link is full due to one elephant stream, while other ECMP links are relatively idle, resulting in uneven load.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Traditional-ECMP-deployment-diagram.png\" alt=\"Traditional ECMP deployment diagram\" class=\"wp-image-4075\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Traditional-ECMP-deployment-diagram.png 1267w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Traditional-ECMP-deployment-diagram-300x161.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Traditional-ECMP-deployment-diagram-1024x548.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Traditional-ECMP-deployment-diagram-768x411.png 768w\" sizes=\"(max-width: 1267px) 100vw, 1267px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Traditional ECMP deployment diagram<\/b><\/strong><\/span><\/p>\n\n\n\n<p>In a test environment with 8 ECMP links simulated internally, the test results are as follows:<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ECMP-traffic-test-results.png\" alt=\"ECMP traffic test results\" class=\"wp-image-4076\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ECMP-traffic-test-results.png 902w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ECMP-traffic-test-results-300x153.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/ECMP-traffic-test-results-768x393.png 768w\" sizes=\"(max-width: 902px) 100vw, 902px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>ECMP traffic test results<\/b><\/strong><\/span><\/p>\n\n\n\n<p>As can be seen, flow-based ECMP causes more obvious occupancy of certain links (ECMP1-5 and 1-6) and idleness (ECMP1-0 to 1-3 are idle). In both All-Reduce and All-to-All models, it is easy for a route to become congested due to uneven load on ECMP. Once the congestion causes retransmission, it increases the overall total communication latency and reduces GPU utilization.<\/p>\n\n\n\n<p>Therefore, the research community has proposed rich solutions such as phost, Homa, NDP, 1RMA, and Aeolus. They address incast to varying degrees, and also address load balancing and low latency request\/response traffic. But they also bring new challenges. Often, these studied solutions require end-to-end problem solving, with large changes to hosts, NICs, and networks, which is costly for the average user.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-challenges-of-ai-clustering-with-box-switches\"><span class=\"ez-toc-section\" id=\"Challenges_of_AI_Clustering_with_Box_Switches\"><\/span><span style=\"font-size: 12pt;\"><strong><b>Challenges of AI Clustering with Box Switches<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Some Internet companies look to box switches with DNX chips supporting VOQ technology to solve the problem of low bandwidth utilization due to load imbalance, but they also face several challenges as follows.<\/p>\n\n\n\n<p>Average scalability. The frame size limits the maximum number of ports. If you want to do a larger scale cluster, you need to expand multiple frames horizontally, which also generates multi-level PFC and ECMP links. So the frame is only suitable for small-scale deployment.<\/p>\n\n\n\n<p>Large power consumption of the device. The number of line card chips, Fabric chips, fans, etc. in the frame is large, and the power consumption of single device is great, easily more than 20,000 watts, some even more than 30,000 watts, with high power requirements for the cabinet.<\/p>\n\n\n\n<p>The number of single device ports is large, and the failure domain is large.<\/p>\n\n\n\n<p>So for the above reasons, box devices are only suitable for small-scale deployment of AI computing clusters.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-ddc-products-born-to-support-aigc\"><span class=\"ez-toc-section\" id=\"DDC_Products_Born_to_Support_AIGC\"><\/span><span style=\"font-size: 14pt;\"><strong><b>DDC Products Born to Support AIGC <\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>DDC is a distributed decoupled frame device solution, using almost the same chip and key technologies as traditional frame switches, but the DDC architecture is simple to support elastic expansion and rapid iteration of functions, easier to deploy, and low power consumption per machine.<\/p>\n\n\n\n<p>As shown in the figure below, the service line card becomes the NCP role as the front end, and the switch board becomes the NCF role as the back end. The original connector components between the two are now replaced by fiber optic cables, and the management engine of the original frame device becomes the centralized\/distributed management component of the NCC in the DDC architecture.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/DDC-product-connectivity-diagram.png\" alt=\"DDC product connectivity diagram\" class=\"wp-image-4077\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/DDC-product-connectivity-diagram.png 1400w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/DDC-product-connectivity-diagram-300x167.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/DDC-product-connectivity-diagram-1024x570.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/DDC-product-connectivity-diagram-768x427.png 768w\" sizes=\"(max-width: 1400px) 100vw, 1400px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>DDC product connectivity diagram<\/b><\/strong><\/span><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-ddc-supports-ultra-large-scale-deployment\"><span class=\"ez-toc-section\" id=\"DDC_supports_ultra-large_scale_deployment\"><\/span><span style=\"font-size: 12pt;\"><strong><b>DDC supports ultra-large scale deployment<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>The advantage of DDC architecture over box architecture is that it can provide flexible scalability, and the network scale can be flexibly selected according to the size of AI cluster.<\/p>\n\n\n\n<p>In the single POD network, 96 NCPs are used as accesses, of which 36 200G interfaces on the downlink of NCPs are responsible for connecting NICs of AI computing clusters. The upstream total of 40<a href=\"https:\/\/www.fibermall.com\/store-20654-200g-qsfp56.htm\" target=\"_blank\" rel=\"noopener\"><span style=\"color: #0000ff;\"> 200G interfaces<\/span><\/a> can connect 40 NCFs, NCF provides 96 200G interfaces, and the upstream and downstream bandwidth of this scale is 1.1:1. The whole POD can support 3456 200G network interfaces, and according to the calculation of one server with 8 GPUs, 432 AI computing servers can be supported.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Single-POD-network-architecture-diagram.png\" alt=\"Single POD network architecture diagram\" class=\"wp-image-4078\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Single-POD-network-architecture-diagram.png 1268w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Single-POD-network-architecture-diagram-300x117.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Single-POD-network-architecture-diagram-1024x400.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Single-POD-network-architecture-diagram-768x300.png 768w\" sizes=\"(max-width: 1268px) 100vw, 1268px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Single POD network architecture diagram<\/b><\/strong><\/span><\/p>\n\n\n\n<p>In multi-level POD networking, on-demand construction based on POD can be realized. Because the NCF equipment in this scenario POD has to sacrifice half of the SerDes for connecting the NCF of the second level, so at this time the single POD uses 48 NCPs as access, with a total of 36 200G interfaces in the downlink, and can support 1728 200G interfaces in a single POD. By increasing the POD horizontally to realize the scale expansion, the overall maximum can support more than 10,368 200G network ports.<\/p>\n\n\n\n<p>NCP uplink 40 200G to 40 NCFs in POD, NCFs in POD use 48 200G interfaces downstream, and 48 200G interfaces are divided into 16 groups to uplink to NCFs in the second level. 40 planes are used for NCFs in the second level, and each plane is designed with 3 units, corresponding to 40 NCFs in POD.<\/p>\n\n\n\n<p>The entire network achieves an overdrive ratio of 1:1:1 within the POD and a convergence ratio of 1:1 between the POD and the second-stage NCF.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/overdrive.png\" alt=\"overdrive\" class=\"wp-image-4079\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/overdrive.png 1269w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/overdrive-300x210.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/overdrive-1024x717.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/overdrive-768x538.png 768w\" sizes=\"(max-width: 1269px) 100vw, 1269px\" \/><\/figure>\n\n\n\n<p>The 200G network port is compatible with <a href=\"https:\/\/www.fibermall.com\/store-21962-100g-nic.htm\" target=\"_blank\" rel=\"noopener\"><span style=\"color: #0000ff;\">100G NIC<\/span><\/a> access, and in special cases, it is compatible with 25\/50G NIC using 1 in 2 or 1 in 4 cables.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-more-balanced-load-based-on-voq-cell-mechanism-lower-packet-loss-rate\"><span class=\"ez-toc-section\" id=\"More_balanced_load_based_on_VOQCell_mechanism_lower_packet_loss_rate\"><\/span><span style=\"font-size: 12pt;\"><strong><b>More balanced load based on VOQ+Cell mechanism, lower packet loss rate<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Relying on the Cells forwarding mechanism after splitting for dynamic load balancing, it realizes the stability of delay and reduces the bandwidth peak difference of different links.<\/p>\n\n\n\n<p>The forwarding process is shown in the figure:<\/p>\n\n\n\n<p>First the sender receives packets from the network and sorts them into VOQs for storage. Before sending the packets a Credit message is sent to determine if the receiver has enough cache space to handle them.<\/p>\n\n\n\n<p>If so, the packets are sliced into Cells and dynamically load balanced to intermediate Fabric nodes. These Cells are reassembled and stored at the receiving end and then forwarded to the network.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/forwarding-process.png\" alt=\"forwarding process\" class=\"wp-image-4080\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/forwarding-process.png 1268w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/forwarding-process-300x185.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/forwarding-process-1024x631.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/forwarding-process-768x473.png 768w\" sizes=\"(max-width: 1268px) 100vw, 1268px\" \/><\/figure>\n\n\n\n<p>Cells are packet-based slicing techniques, typically 64-256 bytes in size.<\/p>\n\n\n\n<p>The sliced Cells are forwarded according to the cell destination query in the reachability table and are sent using a polling mechanism. The advantage of this is that the load of sliced Cells will be fully utilized for each uplink and the amount of data transmitted on all uplinks will be approximately equal, compared to the ECMP mode of selecting a particular path after hashing by flow.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/cell-based.png\" alt=\"cell based\" class=\"wp-image-4081\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/cell-based.png 1268w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/cell-based-300x146.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/cell-based-1024x498.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/cell-based-768x374.png 768w\" sizes=\"(max-width: 1268px) 100vw, 1268px\" \/><\/figure>\n\n\n\n<p>If the receiver is temporarily incapable of processing the message, the message will be temporarily stored in the VOQ on the sender side and will not be forwarded directly to the receiver side resulting in packet loss. Each DNX chip can provide an on-chip OCB cache and an off-chip 8GB HBM cache, which is equivalent to caching about 150ms of data for a 200G port. Credit messages are sent only when they are clearly acceptable at the other end. With such a mechanism, making full use of the cache can significantly reduce packet loss, or even not generate packet loss. With less data retransmission, the overall communication latency is more stable and lower, thus bandwidth utilization can be improved, and thus service throughput efficiency can be enhanced.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/credit-based-fabric.png\" alt=\"credit based fabric\" class=\"wp-image-4082\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/credit-based-fabric.png 1268w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/credit-based-fabric-300x196.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/credit-based-fabric-1024x668.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/credit-based-fabric-768x501.png 768w\" sizes=\"(max-width: 1268px) 100vw, 1268px\" \/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-no-deadlock-under-pfc-single-hop-deployment\"><span class=\"ez-toc-section\" id=\"No_Deadlock_under_PFC_Single-hop_Deployment\"><\/span><span style=\"font-size: 12pt;\"><strong><b>No Deadlock under PFC Single-hop Deployment<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>According to the logic of DDC, all NCPs and NCFs can be seen as one device. Therefore, after deploying RDMA domain in this network, there is only 1 level of PFC at the interface targeting the server, which will not generate multi-level PFC suppression and deadlock as in traditional networks. In addition, according to the data forwarding mechanism of DDC, ECN can be deployed at the interface, and once the internal Credit and cache mechanism cannot support the burst traffic, CNP messages can be sent to the server side to request speed reduction (usually under the communication model of AI, All-to-All and All-Reduce+Cell slicing can balance the traffic as much as possible, and it is difficult to have (1 port is filled up, so ECN can be unconfigured in most cases).<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/No-deadlock-under-PFC-single-hop-deployment.png\" alt=\"No deadlock under PFC single-hop deployment\" class=\"wp-image-4083\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/No-deadlock-under-PFC-single-hop-deployment.png 1265w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/No-deadlock-under-PFC-single-hop-deployment-300x169.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/No-deadlock-under-PFC-single-hop-deployment-1024x578.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/No-deadlock-under-PFC-single-hop-deployment-768x433.png 768w\" sizes=\"(max-width: 1265px) 100vw, 1265px\" \/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-ncc-free-design-with-distributed-os-to-enhance-reliability\"><span class=\"ez-toc-section\" id=\"NCC-free_design_with_distributed_OS_to_enhance_reliability\"><\/span><span style=\"font-size: 12pt;\"><strong><b>NCC-free design with distributed OS to enhance reliability<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>In the management and control plane, in order to solve the impact of management network failure and single point of failure of NCC, we eliminate the centralized control plane of NCC and build a distributed OS, configuring management devices through standard interfaces (Netconf, GRPC, etc.) by SDN operation and maintenance controllers, and each NCP and NCF is independently managed with independent control and management planes.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-test-comparison-results\"><span class=\"ez-toc-section\" id=\"Test_Comparison_Results\"><\/span><span style=\"font-size: 12pt;\"><strong><b>Test Comparison Results<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>From the theoretical point of view, DDC has many advantages such as supporting elastic expansion and rapid iteration of functions, easier deployment, and low power consumption of single machine; however, from the practical point of view, traditional networking also has advantages such as more brands and product lines available on the market, and can support larger scale clusters and other advantages brought by mature technology. Therefore, when customers are faced with project requirements, they can refer to the following comparison and test results to determine whether to choose a higher performance DDC or a traditional network for larger scale deployment:<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Comparison-result-between-traditional-networking-and-DDC-test.png\" alt=\"Comparison result between traditional networking and DDC test\" class=\"wp-image-4084\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Comparison-result-between-traditional-networking-and-DDC-test.png 1269w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Comparison-result-between-traditional-networking-and-DDC-test-300x145.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Comparison-result-between-traditional-networking-and-DDC-test-1024x494.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Comparison-result-between-traditional-networking-and-DDC-test-768x370.png 768w\" sizes=\"(max-width: 1269px) 100vw, 1269px\" \/><\/figure>\n\n\n\n<p class=\"has-text-align-center\"><span style=\"font-size: 10pt;\"><strong><b>Comparison result between traditional networking and DDC test<\/b><\/strong><\/span><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-fibermall-equipment-introduction\"><span class=\"ez-toc-section\" id=\"FiberMall_Equipment_Introduction\"><\/span><span style=\"font-size: 14pt;\"><strong><b>FiberMall Equipment Introduction<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Based on the deep understanding of customer needs, FiberMall has been the first to launch two deliverable products, 200G NCP switch and 200G NCF switch.<\/p>\n\n\n\n<p><strong><b>NCP: FM-S6930-36DC40F1 Switch<\/b><\/strong><\/p>\n\n\n\n<p>This switch is 2U high and provides 36 200G panel ports, 40 200G Fabric inline ports, 4 fans and 2 power supplies.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCP-FM-S6930-36DC40F1-Switch.png\" alt=\"NCP FM-S6930-36DC40F1 Switch\" class=\"wp-image-4085\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCP-FM-S6930-36DC40F1-Switch.png 1268w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCP-FM-S6930-36DC40F1-Switch-300x145.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCP-FM-S6930-36DC40F1-Switch-1024x493.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCP-FM-S6930-36DC40F1-Switch-768x370.png 768w\" sizes=\"(max-width: 1268px) 100vw, 1268px\" \/><\/figure>\n\n\n\n<p><strong><b>NCF: FM-X56-96F1 Switch<\/b><\/strong><\/p>\n\n\n\n<p>This switch is 4U high, providing 96 200G inline ports, 8 fans and 4 power supplies.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter is-resized\"><img decoding=\"async\" src=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCF-FM-X56-96F1-Switch.png\" alt=\"NCF FM-X56-96F1 Switch\" class=\"wp-image-4086\" style=\"width:800px\" width=\"800\" srcset=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCF-FM-X56-96F1-Switch.png 1096w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCF-FM-X56-96F1-Switch-300x233.png 300w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCF-FM-X56-96F1-Switch-1024x797.png 1024w, https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/NCF-FM-X56-96F1-Switch-768x598.png 768w\" sizes=\"(max-width: 1096px) 100vw, 1096px\" \/><\/figure>\n\n\n\n<p>FiberMall will continue to develop and launch 400G port form factor products in the future.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-conclusion\"><span class=\"ez-toc-section\" id=\"Conclusion\"><\/span><span style=\"font-size: 14pt;\"><strong><b>Conclusion<\/b><\/strong><\/span><span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>FiberMall, as an industry leader, has been committed to providing high quality, high reliability network equipment and solutions to meet the increasing demand of customers for Smart Computing Center. While launching the &#8220;Smart Speed&#8221; DDC solution, FiberMall is also actively exploring and developing end-network optimization solutions in traditional networking. By making full use of server intelligent NICs and network equipment protocol optimization, the entire network bandwidth utilization can be improved to help customers usher in the AIGC smart computing era faster.<\/p>\n<style>\r\n\r\n        .lwrp.link-whisper-related-posts{\r\n            \r\n            margin-top: 40px;\nmargin-bottom: 30px;\r\n        }\r\n        .lwrp .lwrp-title{\r\n            \r\n            \r\n        }\r\n        .lwrp .lwrp-description{\r\n            \r\n            \r\n\r\n        }\r\n        .lwrp .lwrp-list-container{\r\n        }\r\n        .lwrp .lwrp-list-multi-container{\r\n            display: flex;\r\n        }\r\n        .lwrp .lwrp-list-double{\r\n            width: 48%;\r\n        }\r\n        .lwrp .lwrp-list-triple{\r\n            width: 32%;\r\n        }\r\n        .lwrp .lwrp-list-row-container{\r\n            display: flex;\r\n            justify-content: space-between;\r\n        }\r\n        .lwrp .lwrp-list-row-container .lwrp-list-item{\r\n            width: calc(100% - 20px);\r\n        }\r\n        .lwrp .lwrp-list-item:not(.lwrp-no-posts-message-item){\r\n            \r\n            list-style: decimal;\r\n        }\r\n        .lwrp .lwrp-list-item img{\r\n            max-width: 100%;\r\n            height: auto;\r\n        }\r\n        .lwrp .lwrp-list-item.lwrp-empty-list-item{\r\n            background: initial !important;\r\n        }\r\n        .lwrp .lwrp-list-item .lwrp-list-link .lwrp-list-link-title-text,\r\n        .lwrp .lwrp-list-item .lwrp-list-no-posts-message{\r\n            \r\n                \r\n        }\r\n        @media screen and (max-width: 480px) {\r\n            .lwrp.link-whisper-related-posts{\r\n                \r\n                \r\n            }\r\n            .lwrp .lwrp-title{\r\n                \r\n                \r\n            }\r\n            .lwrp .lwrp-description{\r\n                \r\n                \r\n            }\r\n            .lwrp .lwrp-list-multi-container{\r\n                flex-direction: column;\r\n            }\r\n            .lwrp .lwrp-list-multi-container ul.lwrp-list{\r\n                margin-top: 0px;\r\n                margin-bottom: 0px;\r\n                padding-top: 0px;\r\n                padding-bottom: 0px;\r\n            }\r\n            .lwrp .lwrp-list-double,\r\n            .lwrp .lwrp-list-triple{\r\n                width: 100%;\r\n            }\r\n            .lwrp .lwrp-list-row-container{\r\n                justify-content: initial;\r\n                flex-direction: column;\r\n            }\r\n            .lwrp .lwrp-list-row-container .lwrp-list-item{\r\n                width: 100%;\r\n            }\r\n            .lwrp .lwrp-list-item:not(.lwrp-no-posts-message-item){\r\n                \r\n                \r\n            }\r\n            .lwrp .lwrp-list-item .lwrp-list-link .lwrp-list-link-title-text,\r\n            .lwrp .lwrp-list-item .lwrp-list-no-posts-message{\r\n                \r\n                    \r\n            }\r\n        }<\/style>\r\n<div id=\"link-whisper-related-posts-widget\" class=\"link-whisper-related-posts lwrp\">\r\n            <h3 class=\"lwrp-title\">Related Posts<\/h3>    \r\n        <div class=\"lwrp-list-container\">\r\n                                            <ul class=\"lwrp-list lwrp-list-single\">\r\n                    <li class=\"lwrp-list-item\"><a href=\"https:\/\/www.fibermall.com\/blog\/25g-sfp28-cwdm.htm\" class=\"lwrp-list-link\"><span class=\"lwrp-list-link-title-text\">SFP28 25G CWDM Transceiver Module: The Future of Ethernet Connectivity<\/span><\/a><\/li><li class=\"lwrp-list-item\"><a href=\"https:\/\/www.fibermall.com\/video\/qsfp-40g-lr4-psm-transceiver.htm\" class=\"lwrp-list-link\"><span class=\"lwrp-list-link-title-text\">QSFP+ 40G LR4 PSM Transceiver\uff1awhat\u2019s so special?<\/span><\/a><\/li><li class=\"lwrp-list-item\"><a href=\"https:\/\/www.fibermall.com\/blog\/managed-switch.htm\" class=\"lwrp-list-link\"><span class=\"lwrp-list-link-title-text\">Managed Switch vs. Unmanaged Switch: 5 Key Differences You Need to Know<\/span><\/a><\/li><li class=\"lwrp-list-item\"><a href=\"https:\/\/www.fibermall.com\/blog\/mellanox-switch.htm\" class=\"lwrp-list-link\"><span class=\"lwrp-list-link-title-text\">Mellanox Switch: Your Guide to High-Performance Ethernet Solutions<\/span><\/a><\/li><li class=\"lwrp-list-item\"><a href=\"https:\/\/www.fibermall.com\/blog\/coherent-optical-transceiver.htm\" class=\"lwrp-list-link\"><span class=\"lwrp-list-link-title-text\">Revolutionizing Networks: The Rise of Coherent Optical Transceivers<\/span><\/a><\/li>                <\/ul>\r\n                        <\/div>\r\n<\/div>","protected":false},"excerpt":{"rendered":"<p>AIGC (AI-Generated Content) has been developing rapidly recently, and the iteration rate is exponentially exploding. Among them, the launch of GPT-4 and ERNIE Bot has drawn great attention to its business value and application scenarios. With the development of AIGC, the scale of training model parameters has gone from hundreds of billions to trillions of [&hellip;]<\/p>\n","protected":false},"author":4,"featured_media":4066,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"site-sidebar-layout":"default","site-content-layout":"default","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"default","adv-header-id-meta":"","stick-header-meta":"default","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":"","_wpscppro_dont_share_socialmedia":false,"_wpscppro_custom_social_share_image":0,"_facebook_share_type":"default","_twitter_share_type":"default","_linkedin_share_type":"default","_pinterest_share_type":"default","_linkedin_share_type_page":"default","_instagram_share_type":"default","_medium_share_type":"","_threads_share_type":"","_google_business_share_type":"","_selected_social_profile":[{"id":"skM9ewvR8O","platform":"linkedin","platformKey":0,"name":"Jason Xue","type":"person","thumbnail_url":"https:\/\/media.licdn.com\/dms\/image\/C5603AQErPqKD0j6qBg\/profile-displayphoto-shrink_100_100\/0\/1599138392315?e=1723075200&v=beta&t=joEkh1OeKQ0F-QpAPv4xxQyBdGlHyccIQZauRSs6RvU","share_type":"default"}],"_wpsp_enable_custom_social_template":false,"_wpsp_social_scheduling":{"enabled":false,"datetime":null,"platforms":[],"status":"template_only","dateOption":"today","timeOption":"now","customDays":"","customHours":"","customDate":"","customTime":"","schedulingType":"absolute"},"_wpsp_active_default_template":true},"categories":[2,29],"tags":[],"class_list":["post-4065","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","category-networking"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v20.13 (Yoast SEO v25.8) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>FiberMall Delivers HPC Networking Solutions for AIGC | FiberMall<\/title>\n<meta name=\"description\" content=\"FiberMall launched the industry&#039;s advanced &quot;Smart Speed&quot; DDC (Distributed Disaggregated Chassis) high-performance networking solution.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"FiberMall Delivers HPC Networking Solutions for AIGC\" \/>\n<meta property=\"og:description\" content=\"AIGC (AI-Generated Content) has been developing rapidly recently, and the iteration rate is exponentially exploding. Among them, the launch of GPT-4 and\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\" \/>\n<meta property=\"og:site_name\" content=\"fibermall.com\" \/>\n<meta property=\"article:published_time\" content=\"2023-03-27T06:26:49+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-06-14T03:17:14+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1055\" \/>\n\t<meta property=\"og:image:height\" content=\"582\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Mia\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Mia\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"17 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\"},\"author\":{\"name\":\"Mia\",\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/person\/56a37c7b0e1e8c61823cd6ef48e8dce3\"},\"headline\":\"FiberMall Delivers HPC Networking Solutions for AIGC\",\"datePublished\":\"2023-03-27T06:26:49+00:00\",\"dateModified\":\"2024-06-14T03:17:14+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\"},\"wordCount\":2741,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png\",\"articleSection\":[\"Blog\",\"Networking\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\",\"url\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\",\"name\":\"FiberMall Delivers HPC Networking Solutions for AIGC | FiberMall\",\"isPartOf\":{\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png\",\"datePublished\":\"2023-03-27T06:26:49+00:00\",\"dateModified\":\"2024-06-14T03:17:14+00:00\",\"description\":\"FiberMall launched the industry's advanced \\\"Smart Speed\\\" DDC (Distributed Disaggregated Chassis) high-performance networking solution.\",\"breadcrumb\":{\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage\",\"url\":\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png\",\"contentUrl\":\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png\",\"width\":1055,\"height\":582,\"caption\":\"Diagram of FiberMall's DDC product connection\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/www.fibermall.com\/blog.htm\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"FiberMall Delivers HPC Networking Solutions for AIGC\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#website\",\"url\":\"https:\/\/www.fibermall.com\/blog.htm\/\",\"name\":\"fibermall.com\",\"description\":\"Optical Communication Expert\",\"publisher\":{\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.fibermall.com\/blog.htm\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#organization\",\"name\":\"fibermall.com\",\"url\":\"https:\/\/www.fibermall.com\/blog.htm\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2024\/12\/cropped-Fiber-Mall-Logo.jpg\",\"contentUrl\":\"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2024\/12\/cropped-Fiber-Mall-Logo.jpg\",\"width\":200,\"height\":67,\"caption\":\"fibermall.com\"},\"image\":{\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/logo\/image\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/person\/56a37c7b0e1e8c61823cd6ef48e8dce3\",\"name\":\"Mia\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/2e83831e57346d5d451c45e43aa96d14fc758d8f72f7304810e123b36585382b?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/2e83831e57346d5d451c45e43aa96d14fc758d8f72f7304810e123b36585382b?s=96&d=mm&r=g\",\"caption\":\"Mia\"},\"description\":\"Optical Transmission Consultant\",\"sameAs\":[\"https:\/\/www.fibermall.com\/\"],\"url\":\"https:\/\/www.fibermall.com\/blog.htm?author=4\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"FiberMall Delivers HPC Networking Solutions for AIGC | FiberMall","description":"FiberMall launched the industry's advanced \"Smart Speed\" DDC (Distributed Disaggregated Chassis) high-performance networking solution.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm","og_locale":"en_US","og_type":"article","og_title":"FiberMall Delivers HPC Networking Solutions for AIGC","og_description":"AIGC (AI-Generated Content) has been developing rapidly recently, and the iteration rate is exponentially exploding. Among them, the launch of GPT-4 and","og_url":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm","og_site_name":"fibermall.com","article_published_time":"2023-03-27T06:26:49+00:00","article_modified_time":"2024-06-14T03:17:14+00:00","og_image":[{"width":1055,"height":582,"url":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png","type":"image\/png"}],"author":"Mia","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Mia","Est. reading time":"17 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#article","isPartOf":{"@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm"},"author":{"name":"Mia","@id":"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/person\/56a37c7b0e1e8c61823cd6ef48e8dce3"},"headline":"FiberMall Delivers HPC Networking Solutions for AIGC","datePublished":"2023-03-27T06:26:49+00:00","dateModified":"2024-06-14T03:17:14+00:00","mainEntityOfPage":{"@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm"},"wordCount":2741,"commentCount":0,"publisher":{"@id":"https:\/\/www.fibermall.com\/blog.htm\/#organization"},"image":{"@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage"},"thumbnailUrl":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png","articleSection":["Blog","Networking"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm","url":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm","name":"FiberMall Delivers HPC Networking Solutions for AIGC | FiberMall","isPartOf":{"@id":"https:\/\/www.fibermall.com\/blog.htm\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage"},"image":{"@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage"},"thumbnailUrl":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png","datePublished":"2023-03-27T06:26:49+00:00","dateModified":"2024-06-14T03:17:14+00:00","description":"FiberMall launched the industry's advanced \"Smart Speed\" DDC (Distributed Disaggregated Chassis) high-performance networking solution.","breadcrumb":{"@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#primaryimage","url":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png","contentUrl":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2023\/03\/Diagram-of-FiberMalls-DDC-product-connection.png","width":1055,"height":582,"caption":"Diagram of FiberMall's DDC product connection"},{"@type":"BreadcrumbList","@id":"https:\/\/www.fibermall.com\/blog\/fibermall-delivers-hpc-networking-solutions-for-aigc.htm#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.fibermall.com\/blog.htm"},{"@type":"ListItem","position":2,"name":"FiberMall Delivers HPC Networking Solutions for AIGC"}]},{"@type":"WebSite","@id":"https:\/\/www.fibermall.com\/blog.htm\/#website","url":"https:\/\/www.fibermall.com\/blog.htm\/","name":"fibermall.com","description":"Optical Communication Expert","publisher":{"@id":"https:\/\/www.fibermall.com\/blog.htm\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.fibermall.com\/blog.htm\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.fibermall.com\/blog.htm\/#organization","name":"fibermall.com","url":"https:\/\/www.fibermall.com\/blog.htm\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/logo\/image\/","url":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2024\/12\/cropped-Fiber-Mall-Logo.jpg","contentUrl":"https:\/\/www.fibermall.com\/blog\/wp-content\/uploads\/2024\/12\/cropped-Fiber-Mall-Logo.jpg","width":200,"height":67,"caption":"fibermall.com"},"image":{"@id":"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/person\/56a37c7b0e1e8c61823cd6ef48e8dce3","name":"Mia","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.fibermall.com\/blog.htm\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/2e83831e57346d5d451c45e43aa96d14fc758d8f72f7304810e123b36585382b?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/2e83831e57346d5d451c45e43aa96d14fc758d8f72f7304810e123b36585382b?s=96&d=mm&r=g","caption":"Mia"},"description":"Optical Transmission Consultant","sameAs":["https:\/\/www.fibermall.com\/"],"url":"https:\/\/www.fibermall.com\/blog.htm?author=4"}]}},"_links":{"self":[{"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/posts\/4065","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4065"}],"version-history":[{"count":5,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/posts\/4065\/revisions"}],"predecessor-version":[{"id":9176,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/posts\/4065\/revisions\/9176"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=\/wp\/v2\/media\/4066"}],"wp:attachment":[{"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4065"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4065"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.fibermall.com\/blog.htm\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4065"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}