Revolutionary cooling technology: Supermicro unveils NVIDIA HGX H100 8.4-GPU server with liquid cooling and 40% lower power costs

0

Supermicro is a leading provider of comprehensive end-to-end IT solutions for cloud computing, artificial intelligence/machine learning, storage and 5G/Edge technologies. The company has expanded its data centre portfolio and proudly presents its NVIDIA HGX H100 liquid-cooled rack-scale solutions. Numerous benefits are achieved thanks to Supermicro’s innovative liquid cooling technologies. The installation of a complete data centre is significantly accelerated, the performance of the servers is increased and the operating costs are significantly reduced. In addition, these technologies lead to a significant improvement in the Power Usage Effectiveness (PUE) value of data centres. When Supermicro liquid cooling solutions are used, the energy savings of a data centre can be estimated at around 40% compared to conventional air-cooled solutions. In addition, direct cooling costs can be reduced by up to an impressive 86% compared to existing data centres. These groundbreaking solutions set new standards for efficient and high-performance data centres.

Supermicro remains an industry leader, providing high-performance solutions for AI workloads and modern data centres worldwide, including energy-efficient GPU servers with liquid cooling technology

Charles Liang, Supermicro’s president and CEO, notes that Supermicro continues to lead the industry in supporting the demanding requirements of AI workloads and modern data centres worldwide. He points out that their innovative GPU servers, which use liquid cooling technology, significantly reduce data centre power requirements. Given the enormous amount of energy required for today’s fast-growing, large-scale AI models, optimising total cost of ownership (TCO) and environmental cost of ownership (TCE) is critical for data centre operators. Supermicro has extensive experience in designing and building full racks of high-performance servers. These GPU systems have been designed from the ground up for rack-scale integration with liquid cooling to ensure superior performance, efficiency and ease of deployment. This allows customers’ requirements to be met with a short lead time.

According to an official announcement from Supermicro, the latest Supermicro product families, including the Intel and AMD server product lines, are capable of delivering AI-optimised racks that are both based on standard engineering templates and can be customised to meet specific user requirements. Supermicro leads the industry in offering the most comprehensive line of high-performance servers and storage systems to successfully handle the most demanding compute-intensive projects. Rack-scale integration gives customers the confidence and flexibility to seamlessly connect racks to their network and achieve faster productivity without having to worry about managing the technology themselves.

According to the information provided, the state-of-the-art liquid-cooled GPU server includes either two Intel or AMD CPUs and eight or four interconnected NVIDIA HGX H100 Tensor Core GPUs. It was pointed out that the use of liquid cooling reduces data centre power consumption by up to 40%, thereby reducing operating costs. In addition, it was reported that both the Intel and AMD systems are significantly more powerful and efficient than the previous generation of systems with NVIDIA HGX GPUs. It was highlighted that these systems offer up to 30 times the performance and efficiency in today’s large Transformer models, with faster GPU-GPU interconnect speeds and PCIe 5.0-based networking and storage.

Supermicro’s NVIDIA H100 SXM5 Tensor Core GPU servers provide a groundbreaking solution for AI model processing. These servers are equipped with advanced NVIDIA Tensor Cores that deliver exceptional computing performance. With their state-of-the-art architecture, Supermicro’s NVIDIA H100 SXM5 Tensor Core GPU servers are capable of efficiently processing today’s largest AI models. With their support, researchers and developers can run complex AI algorithms in real time and gain new insights. Supermicro’s NVIDIA H100 SXM5 Tensor Core GPU servers are a powerful platform for AI research and enable the development of cutting-edge AI applications.

  • SYS-821GE-TNHR – (Dual Intel Xeon skalierbare CPUsder 4. Generation, NVIDIA HGX H100 8 GPUs, 8U)
  • AS -8125GS-TNHR – (Dual AMD EPYC CPUsder 4. Generation, NVIDIA HGX H100 8 GPUs, 8U)

Neben herkömmlichen Servern hat Supermicro auch eine Reihe von GPU-Servern entwickelt, die speziell für KI-Anwendungen konzipiert sind. Diese GPU-Server bieten eine hohe Flexibilität und können für schnelles KI-Training, umfangreiches KI-Inferencing oder KI-verknüpfte HPC-Workloads angepasst werden. Besonders bemerkenswert sind die Systeme, die mit vier NVIDIA H100 SXM5 Tensor Core GPUs ausgestattet sind. Dank dieser leistungsstarken GPUs können Unternehmen komplexe KI-Aufgaben schnell und effizient bewältigen.

  • SYS-421GU-TNXR – (Dual Intel Xeon skalierbare CPUsder 4. Generation, NVIDIA HGX H100 4 GPUs, 4U)
  • SYS-521GU-TNXR – (Dual Intel Xeon skalierbare CPUsder 4. Generation, NVIDIA HGX H100 4 GPUs, 5U)

Supermicro stellt eine fortschrittliche Flüssigkeitskühlungslösung auf Rack-Ebene vor, die für eine breite Palette von Supermicro-Servern entwickelt wurde. Diese Lösung bietet eine beeindruckende Kühlleistung von bis zu 80 kW für CPUs und GPUs mit hoher TDP. Die Coolant Distribution Unit (CDU) gewährleistet eine effiziente und gleichmäßige Verteilung der Kühlflüssigkeit. Durch die Verwendung von redundanter Stromversorgung und austauschbaren Flüssigkeitskühlpumpen wird eine kontinuierliche Kühlung der Server sichergestellt, selbst bei Strom- oder Pumpenausfällen. Die auslaufsicheren Anschlüsse bieten den Kunden zusätzliche Gewissheit und ermöglichen eine zuverlässige und unterbrechungsfreie Flüssigkeitskühlung aller Systeme.

Die Entwicklung und Integration von Rack-Scale-Systemen sind für Systemanbieter von großer Bedeutung geworden. Mit dem Aufstieg von KI und HPC als immer wichtigeren Technologien in Unternehmen müssen Konfigurationen von der Serverebene bis hin zum gesamten Rechenzentrum optimiert und angepasst werden, um maximale Leistung zu erzielen. Die Experten von Supermicro, einem führenden Anbieter von Systemen und Racks, arbeiten eng mit ihren Kunden zusammen, um deren Anforderungen eingehend zu erforschen. Sie verfügen über das erforderliche Fachwissen und beeindruckende Fertigungskapazitäten, um eine große Anzahl von Racks weltweit an ihre Kunden zu liefern. Durch diese enge Zusammenarbeit können sie maßgeschneiderte Lösungen entwickeln, die den spezifischen Anforderungen ihrer Kunden gerecht werden.

Leave A Reply