.png){kind=icon}
.png){kind=icon}
Return
Evaluating RISC-V Processor as an Alternative for High Performance Computing
.png){kind=icon}
PRE
AI
DOI：10.1007/978-3-032-07612-0_40.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The current landscape of High-Performance Computing (HPC) systems is predominantly dominated by x86-based processors, as reflected by the Top 500 list, which showcases their processing power and software ecosystem maturity. However, ARM architectures are steadily gaining traction, driven by their energy efficiency and increasing support in the software ecosystem. Recently, the open-source RISC-V architecture has emerged as a potential alternative, offering flexibility. As RISC-V gains momentum, it is essential to assess its performance. This paper presents an in-depth evaluation of the performance of the RISC-V-based SOPHGO SG2042 processor, comparing it against prominent high-performance processors, including Intel Sapphire Rapids (SPR) and other widely used architectures, through a series of applications based on the Berkeley dwarfs. We assess the processor's performance across a diverse range of computational tasks, covering areas such as dense and sparse linear algebra, spectral methods, graph traversal, and more. We also compare performance in terms of memory & network bandwidth. Our results reveal that while the SG2042 demonstrates strong performance in specific tasks such as Combination Logic and Dynamic Programming, it lags behind in most other applications along with network and memory bandwidth efficiency. However, it shows notable improvements in scalability for some of the benchmarks. The paper provides valuable insights into the potential of RISC-V-based systems for high-performance computing, highlighting both their strengths and limitations. As RISC-V continues to mature, this research provides an objective analysis of the architecture's current performance and its potential for future competitiveness in the high-performance processor landscape.
Keywords:
RISC-V
High performance Computing
Berkeley Dwarfs
Performance evaluation
Journal
H
IF:
0
Papers:
54
Citations:
0