Part of the Technology photoes in this website are created by rawpixel.com - www.freepik.com

Optimizing CXL Memory Pooling: Performance and Compatibility Enhancements

3620

The article "CXL™ Memory Form-Factor Comparison," authored by Timothy Pezarro and Torry Steed, explores the demand for innovative memory solutions to meet the requirements of modern applications, including AI and machine learning. The authors introduce the Compute Express Link™ (CXL™) standard and delve into Type 3 Memory modules, focusing on emerging form factors like EDSFF and AIC. The article compares EDSFF, offering smaller form factors with reduced capacity and power consumption, and AIC, which boasts larger capacities and power requirements but requires internal installation. In conclusion, the article highlights the adaptability of these form factors for CXL devices, enabling seamless integration into existing computer systems without the need for extensive redesign. This balancing of size and capacity caters to the demands of contemporary computer systems.

 

Echoing this article, H3 strives to ensure the compatibility of their composable CXL memory pooling solution with various CXL modules, trying to approach the levels of their direct-attached performances. H3's overarching strategy revolves around prioritizing universal applicability. H3 aims to ensure that the CXL switch system can seamlessly accommodate different CXL modules, regardless of their specific type, to embrace the diversity of CXL modules available in the market. Moreover, H3's CXL memory expansion box mainly adheres to the NVMe standard form factor, ensuring alignment with established industry standards. It not only guarantees compatibility but also streamlines integration with existing systems.

 

At present, H3 actively engages in testing CXL memory modules available in the market, particularly those with E3.S and AIC form factors. These tests involve partnerships with prominent manufacturers, including Smart Modular (Microchip Technology), Samsung (Montage Technology), and Intel for Emerald Rapid and Granite Rapid CPUs. These collaborations aim to enhance their understanding of compatibility and performance. For instance, H3 has conducted memory performance tests on the Samsung CXL memory expander, providing valuable statistics below as a comparison baseline. These partnerships underline H3's commitment to fostering strong industry relationships within the ecosystem.

CXL via flex bus (if PCle GEN 5)           

  

Based on the provided data, the 1:1 Reads-Writes bandwidth test result with one thread (25,722.37 MB/s) is close to the ideal theoretical value of 32 GB/s for an x8 link. It indicates good bandwidth performance. The switch-attached latency measurement of 591 ns compared to the direct-attached test result is described as "ordinary" and suggests that there is room for improvement in terms of reducing latency. Lower latency is generally desirable in high-performance computing and data transfer applications. Therefore, further system development and optimization efforts are recommended to enhance overall performance.

 

In summary, H3's strategy emphasizes compatibility through extensive testing of CXL modules, validating their recognition and data alignment with original specifications. This approach establishes a critical benchmark for future testing, especially in switch-attached modules, enhancing overall compatibility and performance. H3 also works hard to improve performance to meet evolving market demands, all while nurturing valuable industry partnerships.

 

 

References:

Pezarro, Timothy, and Torry Steed. " CXL™ Memory Form-Factor Comparison: Examination of Form-Factors for this Growing Standard.", CXL Consortium, Sep. 5, 2023, https://www.computeexpresslink.org/post/cxl-memory-form-factor-comparison-examination-of-form-factors-for-this-growing-standard?fbclid=IwAR22rsTtO31eeyLBY8Mkq1aujlHts_lw3LHjfXHvfgReUFDyH3Bb9JBsDO4.


category : CXL
tags :
修正document.querySelector('link[rel="canonical"]').href = url_now; setCanonical('https://www.h3platform.com/blog-detail/' + reserved_para); } if (blogNum == "0") { if (para_id == "26") { setTD("NVMe MR-IOV - Lower TCO of IT System|H3 Platform", " Falcon 5208 NVMe MR-IOV solution ensures SSD performance and flexibility,. With built-in PCIe fabric, it requires less hardware to achieve high-performance storage service in comparison to other NVMe-oF solutions. An MR-IOV solution also allows better utilization of expensive CPUs especially in virtual environments."); } else if (para_id == "29") { setTD("【CXL Storage】 CXL 2.0 / PCIe Gen 5 - The Future of Composable Infrastructure|H3 Platform", "H3 Platform has NVMe MR-IOV solution, increasing storage utilization. SR-IOV of the NVMe SSDs is enabled in the NVMe chassis. CXL device are general-purpose accelerators such as NIC and GPU. CXL specification is based on PCIe Gen 5, and CXL allows CPU to access shared memory on accelerator devices. Nowadays, CXL 2.0 introduces pooling capability to the CXL protocol, improving the composability of memory."); } else if (para_id == "30") { setTD("【PCIe Expansion Chassis】– Big Accelerator Memory-Enhancing GPU and Storage Efficiency with PCIe Expansion Solution|H3 Platform", "Nvidia recently released a report on the effectiveness of Big Accelerator Memory (BaM) architecture. BaM leverages GPUDirect RDMA, allowing GPU thread to communicate with SSDs using NVMe queues to ultimately reduce reliance on CPU."); } else if (para_id == "36") { setTD("【CXL memory expansion】– Memory Expansion for Breakthrough Performance|H3 Platform", "CXL memory have been widely discussed for its capability to enhance memory bandwidth and capacity, and these benefits are significant to the emerging AI/ML applications. "); } else if (para_id == "40") { setTD("Toward PCIe Gen 5 Composable Infrastructure as a Service|H3 Platform", "The two case examples above indicate H3's capability to realize device pooling potential and expand resource configuration flexibility. That might be why SC 22 invites H3 to share experiences in the panel session. H3 is ready for everything @SC22. We look forward to displaying H3's avant-garde PCIe Gen 5 CIaaS worldwide."); } else { setTD(strT); } setInternalLink(document.querySelector("div.editor-content"), { href: "/product-list/10", anchor: array_gpuchassis[urlID % array_gpuchassis.length] }, { href: "/product", anchor: array_product[urlID % array_product.length] }); setArticleSchema(); document.querySelectorAll("ul.breadcrumb a")[1].href = "https://www.h3platform.com/blog-list?category=10"; document.querySelectorAll("ul.breadcrumb a")[2].innerHTML = document.querySelector(".title-container h1").innerText; document.querySelectorAll("ul.breadcrumb a")[2].href = url_now; document.querySelectorAll("ul.breadcrumb a")[2].style.color = "#808285"; } else if (blogNum == "1") { if (para_id == "24") { setTD("Increase the Efficiency of Storage System with Multi-host NVMe SR-IOV solution|H3 Platform", "NVMe SR-IOV is the solution for NVMe SSD sharing the resource among multiple servers often limits SSD’s performance as the networking creates I/O bottleneck."); } else if (para_id == "25") { setTD("NVMe MR-IOV – High-Performance Storage Solution for Virtual Environment Deployments|H3 Platform", "Multi-host NVMe SR-IOV, or multi-root SR-IOV (MR-IOV) is the solution aims to improve SSD performance under virtual environments while ensuring high utilization and flexibility for the storage resources. H3 Platform's proposed MR-IOV solution extends the application of SR-IOV."); } else if (para_id == "50") { setTD("【PCIe Gen 5 NVMe chassis】PCIe Gen 5 NVMe MRIOV Solution for Storage Scalability|H3 Platform", "NVMe, a new generation of high-speed storage interface, has higher bandwidth and lower latency than the traditional SATA interface. NVMe Multi-Root IO Virtualization technology (NVMe MR-IOV) further scales up the NVMe resources to realize mass storage sharing and virtualization by allowing multiple virtual machines to visit the same pool of NVMe devices at the same time."); } else { setTD(strT); } setInternalLink(document.querySelector("div.editor-content"), { href: "/product-list/17", anchor: "NVMe MR-IOV Solution" }, { href: "/product", anchor: "Composable NVMe SSD" }); setArticleSchema(); document.querySelectorAll("ul.breadcrumb a")[1].href = "https://www.h3platform.com/blog-list?category=11"; document.querySelectorAll("ul.breadcrumb a")[2].innerHTML = document.querySelector(".title-container h1").innerText; document.querySelectorAll("ul.breadcrumb a")[2].href = url_now; document.querySelectorAll("ul.breadcrumb a")[2].style.color = "#808285"; } else if (blogNum == "2") { setTD(strT); setInternalLink(document.querySelector("div.editor-content"), { href: "/product-list/17", anchor: "NVMe MR-IOV Solution" }, { href: "/product", anchor: "Composable NVMe SSD" }); setArticleSchema(); document.querySelectorAll("ul.breadcrumb a")[1].href = "https://www.h3platform.com/blog-list?category=12"; document.querySelectorAll("ul.breadcrumb a")[2].innerHTML = document.querySelector(".title-container h1").innerText; document.querySelectorAll("ul.breadcrumb a")[2].href = url_now; document.querySelectorAll("ul.breadcrumb a")[2].style.color = "#808285"; } else if (blogNum == "3") { if (para_id == "73") { setTD("Composable Memory System: 210M IOPS, Reduce Bottlenecks|H3 Platform", "Composable memory systems deliver up to 210 million IOPS and remove memory bottlenecks using CXL. Features include dynamic memory pooling, real-time allocation, and improved resource use—helping data centers scale faster while reducing TCO."); } else if (para_id == "72") { setTD("CXL 2.0 Memory Pooling Breakthrough|Four Servers Sharing 2TB Achieve 210M IOPS and 120GB/s Bandwidth|H3 Platform", "Discover H3 Platform's latest advancement in CXL 2.0 memory pooling and memory sharing technology, enabling four servers to share 2TB of memory. Key highlights include achieving 210 million IOPS and 120GB/s bandwidth, significantly enhancing data access speeds and system performance. Explore the detailed test environment, methodologies, and results that showcase this innovative leap in server memory management."); } else if (para_id == "68") { setTD("What is CXL Memory Sharing? Unlocking Shared Memory for AI and HPC|H3 Platform", "Learn how CXL memory sharing is revolutionizing computing with enhanced scalability and efficiency. This blog dives into CXL shared memory, its applications in AI and HPC, and how it transforms disaggregated memory architecture. Explore CXL technologies, protocols, and their role in creating resilient memory management systems for distributed environments. Discover why CXL memory is the future of high-performance computing and data processing."); document.querySelector("main#blog-content img.cover").alt = document.querySelector("div.title-container h1").textContent; } else { setTD(strT); } setInternalLink(document.querySelector("div.editor-content"), { href: "/product-list/18", anchor: "CXL Memory Pooling Solution" }, { href: "/blog-detail/68", anchor: "CXL Memory Sharing Architecture" }); setArticleSchema(); document.querySelectorAll("ul.breadcrumb a")[1].href = "https://www.h3platform.com/blog-list?category=14"; document.querySelectorAll("ul.breadcrumb a")[2].innerHTML = document.querySelector(".title-container h1").innerText; document.querySelectorAll("ul.breadcrumb a")[2].href = url_now; document.querySelectorAll("ul.breadcrumb a")[2].style.color = "#808285"; } else if (blogNum == "4") { // 2025-1208 setTD(strT); /* setInternalLink(document.querySelector("div.editor-content"), { href: "/blog-detail/77", anchor: "AI Storage Fundamentals" }); */ setArticleSchema(); document.querySelectorAll("ul.breadcrumb a")[1].href = "https://www.h3platform.com/blog-list?category=15"; document.querySelectorAll("ul.breadcrumb a")[2].innerHTML = document.querySelector(".title-container h1").innerText; document.querySelectorAll("ul.breadcrumb a")[2].href = url_now; document.querySelectorAll("ul.breadcrumb a")[2].style.color = "#808285"; if (para_id == "77") { setFAQSchema(); } } var breads = [{ href: "/", anchor: "H3 Platform" }, { href: "/blog-list", anchor: "Blog" }, { href: url_now, anchor: document.querySelector(".title-container h1").innerText }]; setBreadCrumbSchema(breads); setSocialMediaMeta({ cond: "meta[property='og:title']", cont: strT }, { cond: "meta[property='og:url']", cont: url_now }, { cond: "meta[property='og:description']", cont: strD }); createTag("meta", { name: "thumbnail", content: document.querySelector("img.cover").src }); function checkData(obj) { for (var i = 0; i < obj.group.length; i++) { if (obj.group[i].blogID.includes(para_id)) { return i; } } } function setTD() { var metaTitle = document.querySelector("title"); var metaDes = document.querySelector("meta[name='description']"); if (arguments.length > 1) { if (!metaDes) { var des = document.createElement("meta"); des.name = "description"; document.getElementsByTagName("head")[0].appendChild(des); des.content = arguments[1]; } else { metaDes.content = arguments[1]; } metaTitle.innerHTML = arguments[0]; } else { metaTitle.innerHTML = arguments[0]; } } function createDetailContent(target, id, content) { var real_id = "jsContent" + id; target.innerHTML = '' + target.textContent + ''; var tag_article = document.createElement("article"); tag_article.style.display = "none"; tag_article.style.textAlign = "center"; tag_article.style.marginBottom = "1em"; tag_article.id = real_id; tag_article.innerHTML = content; target.parentNode.insertBefore(tag_article, target.nextElementSibling); } function show(id) { var t = document.querySelector("article#" + id); t.style.display = (t.style.display == "none") ? "" : "none"; } function addSchema(schema) { var scriptJSON = document.createElement("script"); scriptJSON.type = 'application/ld+json'; scriptJSON.innerHTML = JSON.stringify(schema); document.getElementsByTagName("head")[0].appendChild(scriptJSON); } function extend(obj, src) { for (var key in src) { if (src.hasOwnProperty(key)) obj[key] = src[key]; } } function setBreadCrumbSchema(breadContent) { var schemaData_bread = { "@context": "http://schema.org", "@type": "BreadcrumbList", "itemListElement": [] }; var itemListElement = []; for (var i = 0; i < breadContent.length; i++) { var item = { "@type": "ListItem", "position": i + 1, "item": { "@id": breadContent[i].href, "name": breadContent[i].anchor } }; itemListElement.push(item); } extend(schemaData_bread.itemListElement, itemListElement); addSchema(schemaData_bread); } function setSocialMediaMeta() { for (var i = 0; i < arguments.length; i++) { document.querySelector(arguments[i].cond).content = arguments[i].cont; } } function createTag(tagName) { var tag_head = document.getElementsByTagName("head")[0]; var tag = document.createElement(tagName); for (var i = 1; i < arguments.length; i++) { for (attr in arguments[i]) { tag.setAttribute(attr, arguments[i][attr]); } } tag_head.appendChild(tag); } function setInternalLink(target) { var tagDiv = document.createElement("div"); tagDiv.style.marginTop = "2.5em"; tagDiv.style.textAlign = "left"; tagDiv.style.color = "#231F20"; var strLink = ""; for (var i = 1; i < arguments.length; i++) { strLink += '' + arguments[i].anchor + '|'; } tagDiv.innerHTML = 'Product Info:' + strLink.substring(0, strLink.length - 1); target.appendChild(tagDiv); } function count_url(url) { var url_to_id = 0; for (var i = 0; i < url.length; i++) { url_to_id += url.charCodeAt(i); } return url_to_id; } function getParameter(name, url) { name = name.replace(/[\[\]]/g, "\\$&"); var regex = new RegExp("[?&]" + name + "(=([^&#]*)|&|#|$)"); var results = regex.exec(url); if (!results) { return null; } if (!results[2]) { return ' '; } return decodeURIComponent(results[2].replace(/\+/g, " ")); } function setCanonical(url_path){ var canonical_check = document.querySelector("link[rel=canonical]"); if(!canonical_check){ var link_seo = document.createElement("link"); link_seo.rel = "canonical"; link_seo.href = url_path; var head_place = document.getElementsByTagName("head")[0]; head_place.appendChild(link_seo); } else{ canonical_check.href = url_path; } } function setArticleSchema() { var imgElem = document.querySelector("div.editor-content img"); var imgUrl = imgElem ? imgElem.src : "https://www.h3platform.com/img/blog/blog-banner.jpg"; var timeText = document.querySelector("time").textContent.trim(); var PublishDate = formatDateToISO(timeText); var schemaData_Article = { "@context": "https://schema.org", "@type": "Article", "headline": document.querySelector(".title-container h1").innerText, "image": imgUrl, "datePublished": PublishDate, "author": { "@type": "Organization", "name": "H3 Platform", "url": "https://www.h3platform.com/about" }, "publisher": { "@type": "Organization", "name": "H3 Platform", "logo": { "@type": "ImageObject", "url": "https://www.h3platform.com/img/logo.png" } }, "description": document.querySelector("div.editor-content").innerText.substring(0, 300) + " ..." }; addSchema(schemaData_Article); } function formatDateToISO(timeText) { var dateObj = new Date(timeText); var yyyy = dateObj.getFullYear(); var mm = String(dateObj.getMonth() + 1).padStart(2, '0'); var dd = String(dateObj.getDate()).padStart(2, '0'); var fixedTime = "09:00:00"; var timezone = "+08:00"; return `${yyyy}-${mm}-${dd}T${fixedTime}${timezone}`; } function setFAQSchema() { var schemaData_FAQ = { "@context": "http://schema.org", "@type": "FAQPage", "mainEntity": [] }; var questionList = []; for (var i = 0; i < document.querySelectorAll(".FAQ_Schema_Q").length; i++) { var item = { "@type": "Question", "name": document.querySelectorAll(".FAQ_Schema_Q")[i].textContent.trim(), "acceptedAnswer": { "@type": "Answer", "text": document.querySelectorAll(".FAQ_Schema_A")[i].textContent.trim() } }; questionList.push(item); } extend(schemaData_FAQ.mainEntity, questionList); addSchema(schemaData_FAQ); }