Lenovo provided servers for the Skoltech supercomputer center

On June 30, 2021, Lenovo announced that its Infrastructure Solutions Group (Lenovo ISG) division provided servers for the computing cluster of the Skolkovsky Institute of Science and Technology (Skoltech). Lenovo ThinkSystem SD530 servers formed the basis of a 60-node system for implementing a computer prediction method for crystal structures. This method of computer prediction of USPEX crystal structures was developed in the Laboratory of Computer Design of Materials under the leadership of Professor Artem Oganov. In addition to crystal structures, USPEX makes it possible to predict the structures of low-sized materials: nanoparticles, polymers, surfaces, grain boundaries and 2D crystals.

These tasks required a non-standard supercomputer center, which allows you to solve a large number of parallel tasks. For each structure, 4-5 calculation stages are required. Each calculation is about 8 thousand tasks. And each of the 25-30 users of the system conducts dozens of such calculations monthly.

Accordingly, the main problem was to maximize the number of cores and system performance for a large number of short tasks. When designing the cluster, it was necessary to get the maximum number of cores while minimizing the budget. To solve these problems, Lenovo ThinkSystem SD530 servers were selected.

The developers of the system managed to save on the interconnect: within the node, communications between the cores are carried out very quickly, but the speed of communication between the nodes was medium. The limitation of the amount of RAM has also resulted in some savings. The few laboratory tasks that require large memory capacity are solved on a general institute cluster and on supercomputers of the Academy of Sciences and Nizhny Novgorod University.

The Lenovo ThinkSystem SD530 server contains up to two processors Intel Xeon processors per node, up to 2.75TB RAM with Intel Optane DC Persistent Memory, up to 24 SFF SATA SAS drives\Hot Swap bays - six per node. The chassis in the 2U housing is designed for installation in a rack and includes four independent computing units with the possibility of using graphics accelerators.

Each SD530 ThinkSystem node consists of a 2U modular D2 chassis with up to four SD530 server modules. Each server contains two Intel Xeon Processor Scalable processors. With support for GPU modules as well as the Lenovo Neptune Thermal Transfer Module, the server can be adapted to different workloads.

Modules non-volatile pamyatiintel Optane DC Persistent Memory allow to create the universal level of data storage intended for support of resource-intensive workloads. They allow you to significantly increase server memory capacity, reduce latency, and significantly increase IOPS I/O performance.

ThinkSystem SD530 also supports the LiCO platform which facilitates management of the workloads and resources of a cluster connected with high-performance calculations and artificial intelligence.

Skoltech's HPC Team, which built and tuned this cluster, together with Professional Services Lenovo specialists and with the support of Intel Performance Application engineers, managed to achieve high performance of the system with the 10Gbps Ethernet interconnect, which affected the computing capabilities on the cluster - researchers can quickly obtain the results of calculations of new materials.

The software used by the laboratory is optimized for multi-core processors, further optimization is achieved at the stage of compiling software. In addition, Lenovo engineers optimized performance when installing the system hardware platform.

{{quote 'author
= noted Artem Oganov.|"Solving our problems requires a large number of fast computing cores. They are very well paralleled across 32 processor cores. Further scaling leads to loss. And we have a lot of such tasks, each task includes many small calculations that take 2-4 hours, "-}}

The laboratory continues to engage in high-temperature superconductivity, and a number of high-temperature superconductors have been predicted by its researchers and obtained experimentally. Among the results are thorium hydride, yttrium hydride and a number of other compounds. Scientists are also actively involved in thermoelectric materials. They created their own program for calculating thermoelectric properties, and there are already interesting results. Now there are several materials in development for which they clarify theoretical predictions by studying the high-pressure phases, etc. The tools needed for this are constantly being improved.