Lenovo Pushing the Boundaries of Scientific Discovery for High-Performance Computing at SC24 Using NVIDIA Accelerated Computing

Lenovo, the world's #1 supercomputing provider, is at the forefront of high-performance computing (HPC) and AI innovation. At Supercomputing 2024 (SC24), Lenovo is highlighting powerful, energy-efficient platforms designed to help scientists, engineers, and researchers advance scientific discovery.

This document details the technology behind these accelerated computing platforms, developed in collaboration with NVIDIA, focusing on performance, flexibility, and efficiency for various HPC needs.

Lenovo ThinkSystem SC777 V4 Neptune

Featuring the NVIDIA GB200 Grace Blackwell NVL4 Superchip, the Lenovo ThinkSystem SC777 V4 Neptune is engineered for high-performance computing and accelerated computing for intensive simulations and hybrid AI. It is suitable for technical computing, grid deployments, and analytics workloads in research, life sciences, energy, engineering, and financial simulation.

Description of Figure 1: A detailed internal view of the Lenovo ThinkSystem SC777 V4, showcasing its advanced components and liquid cooling system. The image displays multiple server trays with integrated processors and GPUs, connected by intricate copper water loops, highlighting the system's dense and powerful architecture.

The ThinkSystem SC777 V4 integrates into a standard 19" rack cabinet with the ThinkSystem N1380 Neptune 13U enclosure. An enclosure can hold up to eight SC777 V4 compute trays. A standard rack can accommodate three enclosures, totaling 24 NVIDIA GB200 NVL4 compute nodes, which translates to 48 processors and 96 GPUs within a single rack footprint.

Lenovo Neptune liquid cooling technologies utilize high-quality materials, including custom copper water loops and patented CPU cold plates, for full system water-cooling. Unlike systems using lower-quality FEP plastic, Neptune employs durable stainless steel and reliable EPDM hoses. Liquid cooling can increase performance by up to 10% by enabling continuous turbo mode and reduce data center energy consumption by up to 40%. The fanless system also virtually eliminates noise.

The ThinkSystem SC777 V4 offers a flexible design supporting various system configurations and networking options, ensuring seamless integration into existing data center infrastructures. It supports high-performance NVMe storage and high-speed, low-latency networking with the latest NVIDIA Quantum InfiniBand and NVIDIA Spectrum-X Ethernet platforms, making it a comprehensive solution for HPC and hybrid AI workloads.

NVIDIA Grace Blackwell GB200 NVL4 Superchip

The NVIDIA GB200 Grace Blackwell NVL4 Superchip, at the core of the ThinkSystem SC777 V4 Neptune, delivers revolutionary performance for converged HPC workloads. It combines four NVLink-connected Blackwell GPUs and two Grace CPUs over NVLink-C2C.

• Form Factor: 4 Blackwell GPUs + 2 Grace CPUs
• NVLink Domain: 4-GPU NVLink Domain
• Large Coherent Memory: Up to 768GB
• Performance: 2X Scientific Computing Performance vs. previous generation

Lenovo ThinkSystem SR675 V3

The Lenovo ThinkSystem SR675 V3 is a versatile 3U rack system designed to support up to eight double-wide and single-wide GPUs, including the NVIDIA H200 NVL and L40S Tensor Core GPUs, or the NVIDIA HGX H200 4-GPU offering. It features NVLink and Lenovo Neptune Air hybrid liquid-to-air cooling. The server is based on AMD EPYC 9004 Series processors and the new 5th Gen AMD EPYC 9005 Series processors.

Description of Figure 2: A front view of the Lenovo ThinkSystem SR675 V3 server, configured to support eight double-wide H200 NVL GPUs. The front panel displays numerous ventilation grills, indicating a high-density compute environment.

The SR675 V3 is a cutting-edge server solution for enterprises and data centers aiming to maximize AI and deep learning performance. Its GPU-optimized architecture offers unmatched scalability, housing up to 8 NVIDIA GPUs within a streamlined 3U chassis. This enables organizations to handle demanding AI workloads efficiently while conserving valuable rack space.

Engineered as a future-ready platform, the SR675 V3 is built to support the evolving needs of AI, machine learning, and big data analytics. Its high bandwidth and efficiency ensure readiness for increasingly complex and computationally intensive AI tasks.

Lenovo's integration of 8 x 600W NVIDIA GPUs within a single 3U system provides flexibility and capacity for advancing AI technologies, making it a valuable investment for businesses requiring robust, scalable infrastructure for digital transformation.

For mainstream scientific computing applications across higher education, healthcare, life sciences, energy, and financial services, the ThinkSystem SR675 V3 with 8x NVIDIA H200 NVL and accelerated CUDA software delivers powerful performance and efficient scaling. It also offers flexibility in configuration choices to meet diverse HPC workload needs.

NVIDIA H200 NVL

The H200 NVL is optimized for air-cooled, flexible HPC solutions with a 4-GPU NVLink domain in a PCIe form factor. It offers significant improvements over the previous generation H100.

• Connectivity: Up to four GPUs NVLink connected
• Memory: 141GB HBM3e
• Performance Increase: 1.5X memory increase and 1.2X bandwidth increase over H100 NVL, delivering up to 1.3X HPC performance.
• NVLink Speed: 2X faster 4th generation NVLink speed at 900 GB/s with new multi-node scalability up to 256 GPUs for 57.6 TB/s bandwidth.
• Software: Includes a 5-year subscription to NVIDIA AI Enterprise.

Description of Figure 3: A front view of the Lenovo ThinkSystem SR675 V3 server, highlighting its design with multiple bays for GPU modules. The server chassis is shown with a clean, professional aesthetic typical of enterprise hardware.

Conclusion

Lenovo's innovations, combined with NVIDIA GB200 Grace Blackwell NVL4 and H200 NVL platforms, represent a significant advancement in high-performance computing and AI capabilities, pushing the boundaries of scientific computing.

Visit Lenovo booth #2201 at SC24 to learn more about how "Smarter creates cooler HPC" with Lenovo and NVIDIA, and to explore the full product portfolio.

Author

Patrick Moakley is the Director of Marketing for HPC & AI at Lenovo Infrastructure Solutions Group. He has over 30 years in IT sales and marketing of x86 systems at Lenovo and IBM and holds a Bachelor's degree in Economics from the University of Wisconsin-Madison.

Notices

Lenovo may not offer all products, services, or features discussed in this document in all countries. Please consult your local Lenovo representative for current availability. Lenovo does not imply endorsement of non-Lenovo websites referenced for convenience. Use of those websites is at your own risk.

Lenovo provides this publication "as is" without warranty of any kind. Changes are periodically made to the information herein. Lenovo may make improvements and/or changes in the product(s) and/or program(s) described at any time without notice.

The products described are not intended for use in implantation or other life support applications where malfunction may result in injury or death. The information contained herein does not affect or change Lenovo product specifications or warranties.

Contact for Licensing Inquiries:

Lenovo (United States), Inc.
8001 Development Drive
Morrisville, NC 27560
U.S.A.
Attention: Lenovo Director of Licensing

Comments and Feedback:

Use the online Contact us review form found at: https://lenovopress.lenovo.com/LP2070

Send comments via email to: comments@lenovopress.com

This document, LP2070, was created or updated on November 18, 2024.

Trademarks

Lenovo and the Lenovo logo are trademarks or registered trademarks of Lenovo. AMD and AMD EPYC™ are trademarks of Advanced Micro Devices, Inc. Other company, product, or service names may be trademarks or service marks of others.

Lenovo Trademarks: Lenovo®, Neptune®, ThinkSystem®

© Copyright Lenovo 2024. All rights reserved.