支援AI加速器品牌坐二望三,聯想推出8U、8加速模組的伺服器
<div class="field field-name-field-image field-type-image field-label-hidden"><div class="field-items"><div class="field-item even"><img src="https://s4.itho.me/sites/default/files/lian_xiang_thinksystem_sr680a_v3_main.jpg" width="960" height="420" alt="" /></div></div></div><div class="field field-name-body field-type-text-with-summary field-label-hidden"><div class="field-items"><div class="field-item even"> <p>今年3月Nvidia舉行的年度GTC大會期間,聯想在旗下的伺服器產品線ThinkSystem系列,推出兩款8U尺寸8-GPU組態的2路伺服器系統:SR680a V3與SR685a V3,均可搭配多種資料中心GPU:Nvidia HGX H100 8-GPU、HGX H200 8-GPU,或<a href="https://www.amd.com/zh-tw/products/accelerators/instinct/mi300/platform.html">AMD Instinct MI300X Platform(8個OAM外形的Instinct MI300X)</a>;</p> <p><a href="https://www.intel.com/content/www/us/en/newsroom/resources/computex-2024.html">在6月台北國際電腦展期間</a>,英特爾展出多款採用<a href="https://www.ithome.com.tw/review/162428">Gaudi 3加速器</a>的伺服器,聯想SR680a V3正是其中一款機型;</p> <p><a href="https://s4.itho.me/sites/default/files/images/2024-06-Intel-Computex-Keynote-03-web.jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/2024-06-Intel-Computex-Keynote-03-web.jpg" style="width: 298px;" /></a> <a href="https://s4.itho.me/sites/default/files/images/2024-06-Computex-Lenovo SE680a v3-P1420391-web.jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/2024-06-Computex-Lenovo%20SE680a%20v3-P1420391-web.jpg" style="width: 298px;" /></a></p> <p>到了9月舉行的2024 Lenovo TW Accelerate 通路展望大會,聯想再度展出SR680a V3,而在現場擺設的產品資訊規格的標示看板當中,他們增列尚未推出的Blackwell架構次世代GPU產品B100。</p> <p><img alt="" src="https://s4.itho.me/sites/default/files/images/Lenovo%20AI%20Server-2024-09-13-web.jpg" style="width: 298px;" /> <a href="https://s4.itho.me/sites/default/files/images/SR680a-2024-09-13-2.jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/SR680a-2024-09-13-2.jpg" style="width: 298px;" /></a></p> <p><img alt="" src="https://s4.itho.me/sites/default/files/images/Lenovo%20AI%20servers-2024-09-13-1-680.jpg" style="width: 600px;" /></p> <p>聯想這兩款8U、8-GPU伺服器的主要差異,在於採用不同廠商推出的運算平臺。例如:SR680a V3搭配英特爾第五代Xeon Scalable系列處理器,可安裝32支DDR5-5600記憶體;SR685a V3搭配AMD第四代EPYC系列處理器,可安裝24支DDR5-4800記憶體。</p> <p>若以企業級軟體相容性而言,<a href="https://docs.nvidia.com/certification-programs/nvidia-certified-systems/index.html">這兩款機型均通過Nvidia-Certified Systems的驗證</a>,可支援Nvidia彙整的AI整合軟體套餐Nvidia AI Enterprise,能用於裸機、虛擬化,以及雲端環境的部署。</p> <p><a href="https://s4.itho.me/sites/default/files/images/Qualified System Catalog-SR680a v3.jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/Qualified%20System%20Catalog-SR680a%20v3.jpg" style="width: 298px;" /></a> <a href="https://s4.itho.me/sites/default/files/images/Qualified System Catalog-SR685a v3.jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/Qualified%20System%20Catalog-SR685a%20v3.jpg" style="width: 298px;" /></a></p> <p>在今年公開的AI效能測試項目當中,聯想也提交這兩款8U尺寸8-GPU伺服器的紀錄。例如,在8月底揭露的MLPerf Inference 4.1,以及11月中揭露的MLPerf Training 4.1,SR680a V3與SR685a V3都各有一筆紀錄,搭配的GPU均為內建141 GB HBM3e記憶體的Nvidia H200。</p> <p><a href="https://s4.itho.me/sites/default/files/images/MLPerf Inference v4_1-Lenovo.jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/MLPerf%20Inference%20v4_1-Lenovo.jpg" style="width: 298px;" /></a> <img alt="" src="https://s4.itho.me/sites/default/files/images/MLPerf%20Training%20v4_1-Lenovo.jpg" style="width: 298px;" /></p> <p>關於Nvidia H200的支援,SR685a V3在產品指南文件加入相關敘述是在7月,10月底宣告支援AMD第四代EPYC處理器96核心款式9654;而在AMD Instinct MI300X Platform的搭配上,AMD在2024第二季認證伺服器型錄,列入SR685a V3。至於Intel Gaudi 3的搭配選項,截至11月底為止,聯想仍未在SR680a V3與SR685a V3產品指南與技術規格,列入相關描述。</p> <p><a href="https://s4.itho.me/sites/default/files/images/AMD Instinct MI300X Platform-Accelerated Servers-2024-Q2(1).jpg"><img alt="" src="https://s4.itho.me/sites/default/files/images/AMD%20Instinct%20MI300X%20Platform-Accelerated%20Servers-2024-Q2(1).jpg" style="width: 500px;" /></a></p> <p>以產品設計概念而言,聯想這批AI加速伺服器主打可支援多種AI加速器,令人想到市面上另一款更早推出的產品Dell PowerEdge XE9680,這款伺服器採用6U尺寸機箱,最初可搭配8個Nvidia A100或H100,後續擴增支援更多加速器選項,截至目前為止,XE9680可搭配Nvidia HGX H100 8-GPU、HGX H200 8-GPU,或AMD Instinct MI300X Platform,也能改用Intel Gaudi 3 HLB-325(8個OAM外形的Gaudi 3)。</p> <p>事實上,Dell XE9680這些AI加速器搭配是陸續提供的。首先,以Nvidia的產品而言,<a href="https://www.ithome.com.tw/review/156425">XE9680去年可支援Nvidia HGX H100 8-GPU與HGX A100 8-GPU</a>,<a href="https://www.dell.com/zh-tw/blog/dell-poweredge-xe9680-ai-acceleration-announcements-at-nvidia-gtc/">Dell今年3月宣布支援HGX H200</a>,10月宣布搭配HGX H200與HGX H100的HPL Benchmark效能測試數據;第二,是AMD的產品,<a href="https://infohub.delltechnologies.com/en-au/p/silicon-diversity-deploy-genai-on-the-poweredge-xe9680-with-amd-instinct-mi300x-accelerators/">Dell今年5月宣布可搭配AMD Instinct MI300X Platform</a>,以及Dell Validated Design for Generative AI with AMD,7月XE9680搭載AMD Instinct MI300X開始出貨,Dell Validated Design for Generative AI with AMD也正式供應,<a href="https://infohub.delltechnologies.com/fr-fr/p/amd-instinct-mi300x-accelerators-on-poweredge-xe9680-serving-cohere-s-command-r-model-using-vllm/">10月揭露用於大型語言模型Command R+的AI處理效能數據</a>;至於Intel的產品,<a href="https://www.dell.com/zh-tw/blog/experience-choice-in-ai-with-poweredge-and-gaudi3/">Dell在今年4月宣布XE9680將可搭載Intel Gaudi 3</a>,<a href="https://www.dell.com/zh-tw/blog/exciting-dell-intel-ai-updates-from-dell-technologies-world/">5月舉行的Dell年度用戶大會預告將在下半年推出</a>,<a href="https://infohub.delltechnologies.com/en-us/p/introducing-dell-poweredge-xe9680-with-intel-gaudi-3-ai-accelerator/">9月宣布這樣的搭配將在第三季量產</a>,到了9月底英特爾正式發表Xeon 6 P-core處理器的同時,<a href="https://www.intel.com/content/www/us/en/newsroom/news/next-generation-ai-solutions-xeon-6-gaudi-3.html#gs.idk9ou">也預告兩款10月供應Gaudi 3的伺服器產品</a>,其中之一正是Dell XE9680。</p> <p>反觀聯想,根據在他們網站公布的<a href="https://lenovopress.lenovo.com/lp1909-thinksystem-sr680a-v3-server">SR680a V3</a>與<a href="https://lenovopress.lenovo.com/lp1909-thinksystem-sr680a-v3-server">SR685a V3</a>產品指南文件,最近一次改版日期是11月18日,上面並未提到Intel Gaudi 3,僅在文末的商標使用聲明提到Intel Gaudi,或許未來會對這部分支援明確交代相關資訊。</p> <p><span style="color:#008080;"><strong>產品資訊</strong></span></p> <p><span style="color:#0000FF;"><strong>聯想ThinkSystem </strong><strong>SR680a V3與SR685a V3</strong></span><br /> ●原廠:聯想<br /> ●建議售價:廠商未提供<br /> ●機箱尺寸:8U<br /> ●處理器:SR680a V3搭配2顆Intel第五代Xeon Scalable系列(最高搭配350瓦TDP機型、64核心),SR685a V3搭配2顆AMD第四代EPYC系列(最高搭配400瓦TDP機型、96核心)<br /> ●記憶體:SR680a V3搭配32個DDR5-5600插槽,最大可擴充至4 TB,SR685a V3搭配24個DDR5-4800插槽,最大可擴充至3 TB<br /> ●儲存配置:16臺2.5吋NVMe固態硬碟+2臺M.2 NVMe固態硬碟<br /> ●搭配GPU:Nvidia HGX H100 8-GPU,8個SXM5形式GPU,或是Nvidia HGX H200 8-GPU,8個SXM5形式GPU,或是AMD Instinct MI300X 8-GPU,8個OAM形式GPU<br /> ●I/O擴充介面:10個PCIe 5.0 x16全高半長介面卡<br /> ●電源供應器:8臺2600瓦(N+N備援)</p> <p>【註:規格與價格由廠商提供,因時有異動,正確資訊請洽廠商】</p> </div></div></div>