Abstract:HPC world is dominated by x86 ISA CPUs. This monoculture is not necessarily justified by best performance evaluation, but may inherit from e.g. SW related restrictions on the choice of HW platforms. To avoid running (further) into path dependency, alternate HW platforms need to be evaluated for performance compared to existing HPC setup. As a result, it may turn out alternate HW platforms are more efficient for HPC. In any case, even if performance differences are low, avoiding path dependencies that stem from HW choice restrictions simplifies switching to different HW platforms in future, should suitable systems evolve. Moreover, broadening the perspective to generic HW platforms may trigger cooperation and wield influence on HW platform development, resulting in HW/SW co-design advantages.

What problem does this paper attempt to address?

The problems that this paper attempts to solve mainly focus on evaluating the advantages and disadvantages of high - performance computing (HPC) servers based on the ARM architecture in terms of computing performance and energy efficiency compared to existing x86 - architecture HPC systems. Specifically, the paper focuses on the following aspects: 1. **Performance Evaluation**: By using the GROMACS molecular dynamics simulation software as a benchmarking tool, evaluate the performance differences of running the same task on different hardware platforms. This includes the measurement of running time and the comparison of performance indicators (such as ns/day). 2. **Energy Efficiency**: In addition to performance, the paper also pays special attention to the energy consumption of different systems when performing the same task, that is, energy efficiency. By measuring the power consumption of different systems and combining it with the running time to calculate the energy required for each nanosecond of calculation, the energy efficiency of the system is evaluated. 3. **Avoiding Path Dependence**: The paper points out that the dominance of the x86 architecture in the current HPC field may not be entirely based on the optimal performance choice, but is affected by factors such as software compatibility. Therefore, evaluating HPC systems based on the ARM architecture helps break this path dependence and provides more flexibility for future hardware selection. 4. **Hardware/Software Co - design**: By evaluating the performance of different hardware platforms, the co - design of hardware and software can be promoted, thereby driving the development of HPC technology. For example, through this evaluation, certain specific hardware characteristics that have a significant impact on performance or energy efficiency can be discovered, thus guiding the design direction of future hardware. In summary, the main purpose of this paper is to provide a comprehensive understanding of the advantages and disadvantages of ARM - architecture servers compared to traditional x86 - architecture servers in the HPC field through detailed performance and energy efficiency evaluations, thereby promoting the diversified development of HPC technology.

Computational Performance and Energy Efficiency of ARM based HPC servers

Towards Green Computing: A Survey of Performance and Energy Efficiency of Different Platforms using OpenCL

Exploration of Performance and Energy Trade-offs for Heterogeneous Multicore Architectures

Future Computing Platforms for Science in a Power Constrained Era

Rethinking High Performance Computing Platforms: Challenges, Opportunities and Recommendations

Comprehensive analysis of energy efficiency and performance of ARM and RISC-V SoCs

Designing a High Performance Parallel Personal Cluster

Heterogeneous architectures for computational intensive applications: A cost-effectiveness analysis

Crossing the Architectural Barrier: Evaluating Representative Regions of Parallel HPC Applications

Energy-aware operation of HPC systems in Germany

Understanding HPC Benchmark Performance on Intel Broadwell and Cascade Lake Processors

Direct N-body application on low-power and energy-efficient parallel architectures

Performance on HPC Platforms Is Possible Without C++

ARM Wrestling with Big Data: A Study of Commodity ARM64 Server for Big Data Workloads

Parallelizing Workload Execution in Embedded and High-Performance Heterogeneous Systems

Evaluating HPX and Kokkos on RISC-V using an Astrophysics Application Octo-Tiger

Parallel Performance of ARM ThunderX2 for Atomistic Simulation Algorithms

Evaluating ARM and RISC-V Architectures for High-Performance Computing with Docker and Kubernetes

Redefining Computing: Rise of ARM from consumer to Cloud for energy efficiency

HW/SW Co-Optimization for Stencil Computation: Beginning with a Customizable Core

FPGA‐based HPC accelerators: An evaluation on performance and energy efficiency