Intel Borealis test system for the Aurora supercomputer goes online

The Aurora supercomputing system is a supercomputer project of the U.S. Department of Energy’s Argonne National Laboratory. It will be built on the basis of Intel Sapphire Rapids and Ponte Vecchio to achieve exaFLOPs per second (ExaFLOP) computing power. The project has been delayed by problems with Intel’s research and development.

Recently, the Intel Borealis test system deployed by Intel for the Aurora supercomputing system has finally been launched. It will be used to run system performance evaluation, stress testing, and debugging technology and architecture before it is fully launched. It has finally taken a milestone step. Intel also released a video of the Borealis test system, and those who are interested can click to watch it on the official website.

Aurora will have about 10,000 blade servers, each with two Xeon Max CPUs and six Ponte Vecchio GPUs. The architecture of the Intel Borealis test system is the same as that of the Aurora supercomputing system, and the configuration and scalability settings are the same, except that there are only 128 blade servers. At the same time, the CPU temporarily uses the fourth-generation Intel Xeon Scalable processor, not a model with stronger performance and HBM. It is understood that its nodes are eight HPE Slingshot-11 interconnection systems, using a dragonfly topology.

Intel has previously stated that the Aurora supercomputing system is designed to handle high-performance computing, AI/ML, and big data analysis workloads, and can achieve a peak computing power of 2 ExaFLOP. The entire Aurora supercomputing system occupies about the size of two basketball courts. When it is officially launched, it is expected to become one of the few supercomputers reaching the Exascale level.