Google and Intel have announced a collaboration to boost High Performance Computing (HPC) performance on Google Cloud using the Intel® MPI Library. This is particularly significant for computationally demanding tasks such as molecular dynamics simulations, computational geoscience, weather forecasting, and Electronic Design Automation (EDA).
The article highlights the use of the optimized Intel MPI Library, specifically version 2021.11, which has been fine-tuned to work seamlessly with Google Cloud's 3rd Gen VMs and Titanium technology. By leveraging the network offload capabilities of the Intel Infrastructure Processing Unit (IPU) E2000, the optimized MPI Library enables faster and more efficient communication between compute nodes, leading to improved performance for HPC applications.
Tests conducted using Siemens SimcenterTM STAR-CCM+TM software, a popular application used in computational fluid dynamics, demonstrate significant performance improvements when using the Intel MPI Library 2021.11. For instance, some benchmarks, such as LeMans_100M_Coupled and AeroSuvSteadyCoupled106M, exhibited superlinear scaling for certain VM counts, likely due to the increased available cache.
Furthermore, when compared to the previous version (Intel MPI 2021.7), the 2021.11 version showed marked improvements in parallel scalability and absolute performance for nearly all benchmarks. These efficiency gains - and thus faster time-to-solution and lower costs - are already present at just two VMs (up to 1.06x improvement) and grow dramatically at larger VM counts (between 2.42x and 5.27x at 32 VMs).
These findings underscore the substantial impact of software optimizations like MPI libraries on HPC performance. By leveraging the latest advancements in both hardware and software, organizations can achieve faster time-to-solution, improved resource utilization, and drive innovation across various scientific and engineering domains.
