Expanded Manufacturing Footprint is Targeted to Bring Global Liquid-Cooled Rack Capacity to More than Double Today's 1,000 AI SuperClusters Shipped Per Month
SAN JOSE, Calif., June 17, 2024 /PRNewswire/ -- Supermicro, Inc. (NASDAQ: SMCI), a Total IT Solution Provider for AI, Cloud, Storage, and 5G/Edge, continues to expand in Silicon Valley with new campuses as the demand for liquid-cooled data center increases. The new facilities will be part of the new liquid-cooled ecosystem, reducing the time needed to deliver to customers worldwide. These new facilities will focus on delivering entire plug-and-play liquid-cooled solutions, from systems to racks to water towers. With AI factories becoming more prevalent, liquid-cooled data centers are critical to meet these increasing customer demands for AI-focused workloads. Liquid-cooled data centers increase the amount of AI compute performance per watt, resulting in more performance per data center. In addition, electricity consumption and environmental impact are reduced, and operational expenses can be up to 40% lower compared to traditional, air-cooled data centers.
"Supermicro has the highest performing generative deep learning and inferencing AI platform and clusters which benefit from liquid-cooled technologies," said Charles Liang, president and CEO of Supermicro. "Many data center owners are looking for electricity saving Direct Liquid-Cooled solutions. Supermicro is developing building block liquid-cooled solutions for AI factories and the HPC market. We anticipate that liquid-cooled data centers will grow from historically less than 1% to an expected 15% and up to 30% of all data center installations in the next two years. This expansion positions us to capture the majority share of that growth. New data centers will run more efficiently, reducing their carbon footprint, and can be virtually free, compared to air-cooled data centers, with significant operational savings realized through lower electricity use over time."
To learn more about Supermicro's Building Block liquid cooling solutions, click here
"We are thrilled that Supermicro is expanding in San Jose and leading the charge to reduce the impact of data centers," said Matt Mahan, Mayor of San Jose, CA. "Supermicro represents the best of Silicon Valley as a company that continues to push the boundaries of both technology and sustainability."
Supermicro, as a total solution provider, supplies all the components needed for an efficient and optimized liquid-cooled solution. From the Supermicro optimized cold plates to the coolant distribution manifolds (CDMs) to the redundant coolant distribution units (CDUs) and even the external cooling tower, the entire end-to-end solution is designed and tested by Supermicro engineers, which results in a higher quality solution, enabling organizations to get more productive sooner.
Many Supermicro servers are explicitly designed for high-performance NVIDIA GPUs for AI processing, such as the very high-density 4U-8GPU systems, the Intel CPU-based SYS-421GE-TNHR2-LCC, or the AMD CPU-based AS -4125GS-TNHR2-LCC. In addition, many Supermicro servers, such as the popular 8U-8GPU systems, are liquid-cooled ready, as well as NVIDIA-based 1U ARS-111GL-NHR-LCC, which includes the Grace Hopper Superchip. Liquid-cooled servers include the 8U-20-node SuperBlade, which provides the highest CPU and GPU computing density available today. The 2U-4 node Supermicro BigTwin and the 4U-8 node FatTwin multi-node servers are optimized for liquid cooling and are in high volume production today.
Supermicro continues to work closely with data center operators to match the right server technology to the demanding workloads, which are unique to each organization and cloud service provider. Many of these application-optimized servers are designed to be liquid-cooled, which reduces the overall operational cost of the data center.
Many customers have limited experience handling liquid inside data centers, requiring vendors to perform the installation and acceptance testing along with future maintenance and warranty service. On-site support is critical, and Supermicro's service organization is highly trained in liquid-cooling deployments.
擴大製造足跡的目標是使全球液冷機架容量達到當今每月出貨的1,000個人工智能超級集群的兩倍以上
加利福尼亞州聖何塞,2024年6月17日 /PRNewswire/ — 隨着對液冷數據中心需求的增加,人工智能、雲、存儲和5G/Edge的整體IT解決方案提供商超微電腦(納斯達克股票代碼:SMCI)繼續在硅谷擴張新園區。新設施將成爲新的液冷生態系統的一部分,從而縮短向全球客戶交付所需的時間。這些新設施將專注於提供完整的即插即用液冷解決方案,從系統到機架再到水塔。隨着人工智能工廠變得越來越普遍,液冷數據中心對於滿足客戶對以人工智能爲中心的工作負載不斷增長的需求至關重要。液冷數據中心可提高每瓦特的人工智能計算性能,從而提高每個數據中心的性能。此外,與傳統的風冷數據中心相比,電力消耗和環境影響降低,運營費用最多可降低40%。
超微電腦總裁兼首席執行官Charles Liang表示:“超微電腦擁有性能最高的生成式深度學習和推理人工智能平台和集群,這些平台和集群受益於液冷技術。”“許多數據中心所有者正在尋找節電的直接液體冷卻解決方案。超微電腦正在爲人工智能工廠和 HPC 市場開發積木液冷解決方案。我們預計,在未來兩年中,液冷數據中心將從歷史上的不到1%增長到預期的15%,佔所有數據中心安裝量的30%。這種擴張使我們能夠佔據增長的大部分份額。與風冷數據中心相比,新的數據中心將更高效地運行,減少其碳足跡,並且幾乎可以免費使用,隨着時間的推移降低用電量,可以節省大量的運營成本。”
加利福尼亞州聖何塞市長馬特·馬漢說:“我們很高興超微電腦在聖何塞擴張業務並牽頭減少數據中心的影響。”“作爲一家不斷突破技術和可持續發展界限的公司,超微電腦代表了硅谷的精華。”
作爲整體解決方案提供商,超微電腦提供高效和優化的液冷解決方案所需的所有組件。從超微電腦優化的冷板到冷卻劑分配歧管 (CDM) 再到冗餘冷卻劑分配單元 (CDU) 甚至外部冷卻塔,整個端到端解決方案均由超微電腦工程師設計和測試,從而產生更高質量的解決方案,使組織能夠更快地提高生產力。
許多超微電腦服務器都是專門爲用於人工智能處理的高性能 NVIDIA GPU 設計的,例如密度極高的 4U-8GPU 系統、基於英特爾 CPU 的 SYS-421GE-TNHR2-LCC 或基於 AMD CPU 的 AS -4125GS-TNHR2-LCC。此外,許多超微電腦服務器,例如流行的8U-8GPU系統,以及基於NVIDIA的1U ARS-111GL-NHR-LCC,包括Grace Hopper Superchip,都已做好液冷準備。液冷服務器包括8U-20節點的SuperBlade,它提供了當今最高的CPU和GPU計算密度。2U-4節點的Supermicro BigTwin和4U-8節點的FatTwin多節點服務器針對液體冷卻進行了優化,目前正在大批量生產。
超微電腦繼續與數據中心運營商密切合作,將正確的服務器技術與要求苛刻的工作負載相匹配,這些工作負載是每個組織和雲服務提供商所獨有的。這些針對應用程序進行了優化的服務器中有許多是爲液體冷卻而設計的,這降低了數據中心的總體運營成本。
許多客戶在數據中心內處理液體的經驗有限,需要供應商進行安裝和驗收測試以及未來的維護和保修服務。現場支持至關重要,超微電腦的服務組織在液體冷卻部署方面接受過嚴格的培訓。