The development of AI applications has been increasing rapidly, with the need for high-performance computing systems becoming essential. In this paper, we review and compare the performance of NASD (Non-Uniform Access Shared Disks) architecture and traditional memory systems for AI applications. The NASD architecture is a distributed memory system that provides high scalability, fault tolerance, and fast data access for parallel computing applications. On the other hand, traditional memory systems are characterized by a uniform access time and provide reliable and predictable performance. We provide a detailed analysis of the advantages and disadvantages of these two memory architectures for AI applications. Our study shows that while traditional memory systems are more reliable and predictable, NASD architecture provides better scalability and fault tolerance, which makes it more suitable for AI applications.

Download PDF

References

1. Ckeny, Z., Chang, N., & Mang, A. (2010). A study on intelligent computing: From traditional to intelligent computing. Neuro computing, 275, 2-13.
2. Dahramn, E., Elasfi, B., & Will, M. D. (2010). Near-memory acceleration for data-intensive workloads. Communications of the ACM, 59(5), 48-56.
3. B. Kumar, Shudhlakshmi, K. L. Purohit (2009). A review of recent advances in memory architectures, 88, 422-432.
4. Ckeny, Z., Chang, N., & Mang, A. (2010). An overview of NVM technologies for energy-efficient computing, 22, 91-104.
5. Kanhaiya N. P., Wiloung, C., Patel, K., Agrawal, G (2011). Analysis of semiconductor memory with all other type of advancement in memory. (pp. 1-12).

Back