For at least four years, Lenovo has been expressing its big ambitions for high performance computing, and the company is clearly making progress towards its goals. This is the eighth year since China-based Lenovo bought the Intel x86-based server business from IBM, an acquisition that marked the company’s entry into HPC, and despite certain sales restrictions. of the US federal market sector imposed on the company, Lenovo has progressed among the leading suppliers of HPC servers.
We recently spoke with Noam Rosen, Director EMEA, HPC & AI at Lenovo Data Center Group, about the company’s recent HPC accomplishments that she will be showcasing at the ISC, accomplishments that include:
- The company’s execution of its HPC strategy, he said, of “continuing to deliver a diverse, state-of-the-art HPC portfolio that supports our vision, what we call ‘Exascale to Every Scale ‘”.
- Strong financial performance: Rosen said in its latest quarterly earnings report, Lenovo’s HPC revenue grew 32% year-over-year. “We came in last year thinking it would be a pivotal year for us in HPC,” said Rosen. “We saw significant headwinds…” he said, not the least of which was COVID-19, as well as a disruption in supplier shipments. “However, our team (EMEA), in particular in HPC, had a very rich year with exceptional results in all financial areas, turnover, gross margin…, despite all the challenges.
On the server and contract earnings front, Rosen cited Lenovo’s work with AMD and Intel.
In March, the company’s Data Center group launched HPC servers using AMD’s new 7nm EPYC 7003 Series processors – ThinkSystem and ThinkAgile hyperconverged infrastructure (HCI) solutions to support edge computing to the cloud.
In addition, the Lenovo-AMD tandem is teaming up to provide what is expected to be the most powerful supercomputer in the Netherlands for SURF, the ICT cooperative for education and research. The € 20 million project, which started earlier this year, is intended to support scientists at more than 100 educational and research institutions dealing with complex calculations in the fields of meteorology, astrophysics, medical and social sciences, and materials and earth sciences, such as climate change research.
The system, said Rosen, “will be a blend of the current generation of AMD EPYC as well as next generation hot water-cooled EPYC processors, which will be considered the fourth generation of AMD EPYC next year. . It will also include a next-generation high-speed NVME interconnect from Nvidia… with 3PFLOPS of Nvidia accelerated partitioning for artificial intelligence and machine learning workloads, as well as nearly 10 petabytes of Lenovo high-performance storage. This will become the first supercomputer in the Netherlands that will achieve 14PFLOPS. “
As for Intel, Lenovo launched ThinkSystem servers using the 3rd Intel Xeon Scalable data center chips, codenamed “Ice Lake”, when the chips were introduced in April.
“We have already been able to complete major deployments on Ice Lake, which were among the first in Europe for HPC,” said Rosen. “We are very proud of it. “
Working in partnership with Intel, Lenovo and the Leibniz Supercomputing Center (LRZ) of the Bavarian Academy of Sciences and Letters, earlier this month announced phase two of LRZ’s SuperMUC-NG supercomputer, which will provide integrated solutions to the community of LRZ users using AI. to implement advanced simulations, modeling and data analysis. This is a second phase of expansion of SuperMUC-NG, which is part of the Gauss Center for Supercomputing (GCS).
“Today we are seeing a significant shift in the way HPC and AI workloads converge,” said Rosen, “we see a lot of powerful synergy between the two at LRZ.” Phase two of SuperMUC-NG, he said, will focus on the AI capabilities of the system incorporating hundreds of nodes equipped with the next generation of Intel’s “Sapphire Rapides” Xeon server processors based on the. 3rd refinement of the 10nm process, as well as Intel’s next “Ponte”. GPU Vecchio ”, integrated in water-cooled nodes with Lenovo Neptune technology. The system will also use high performance Intel DAOS storage.
“It’s very well suited for highly scalable and data-intensive workloads, especially AI applications. DAOS is an open source, software-defined object store designed from the ground up for massively distributed non-volatile memory. This is an essential element of a multi-year strategic partnership that we have with Intel for LRZ.
Lenovo’s water cooling technology “allows us to deliver up to 2PFLOPS of computing power directly with the technology currently available,” said Rosen. “We are among the first to offer this type of direct hot water liquid cooling for GPUs in the market, especially in Europe. “
Other winning customers over the past year and more include a 17PFLOPS supercomputer at the Karlsruhe Institute of Technology (KIT) in Germany. Called HoreKa, the system was put online last fall, a contract which would be in the order of 15 million euros. The hybrid system has nearly 60,000 next-generation Intel Xeon Scalable processor cores and 220 terabytes of main memory as well as 740 Nvidia A100 Tensor Core GPUs. A non-blocking NVIDIA Mellanox InfiniBand HDR network with 200 Gb / s per port is used for communication between nodes.
Last July, Lenovo announced that it was rolling out a liquid-cooled Neptune supercomputer at the Max Planck Society, a $ 20 million project that included a 100,000-core Neptune made up of Lenovo ThinkSystem servers with Intel processors and Nvidia Tesla A100 GPUs, software and operational support, increasing the company’s overall peak performance in its HPC complex at 12 PFLOPS. The HPC cluster is water-cooled in two stages – in the first stage, the essential components of the boards are cooled with copper pipes circulating liquid; secondly, the racks are cooled using the Lenovo rear door heat exchanger (water cooled rear doors) to dissipate 100% of the heat lost by convection in the water.
Rosen also discussed a joint project Lenovo is engaged in with Israel-based AI chipmaker Next Silicon that is creating a buzz. This project will build a test bed at the University of Birmingham with new accelerator technology from NextSilicon as part of a co-design partnership. The project will evaluate the performance of the main codes used by British Research and Innovation Researchers (UKRI), with a focus on the evaluation of classes of algorithms used in high-performance computing and data science, as well as other HPC applications.
The testbeds are part of the ExCALIBUR program, a £ 45.7million initiative of the UK Government’s Strategic Priorities Fund led by the Met Office, UK Research and Innovation (UKRI) with the UK Atomic Energy Authority ( UKAEA) and is the UK’s Exascale program through the end of 2024.
“We believe that the trends in supercomputing for the next decade and beyond will be determined by disaggregation,” said Rosen, “and made possible by improved network and even greater (processor) specialization. will also be very important, it is obviously the management of power, energy efficiency, cooling and the reduction of operating costs of the systems.
“Just to give you one, an example of innovative technology that we are working to bring to market with a specific partner is… Next Silicon…”, he said, “to integrate different compute technologies at the exascale level for different organizations. We also work closely with some of the very promising tech startups… Next Silicon is a good example.
He said that Lenovo and Next Silicon are optimizing that company’s first-generation processors for upcoming Lenovo HPC systems and conducting an early access program with intensive data centers to develop application porting, tuning and equipment evaluation. “This allows us to host Next Silicon technology at some of our joint customers and prepare the best system solutions for next year so that they are ready for production. “