Intel Unveils Data Streaming Accelerator for Enhanced Performance Efficiency

In the era of Dennard scaling, modern CPUs are shifting towards specialization, integrating capable data accelerators on chip to improve performance and efficiency. This trend has given rise to the adoption of accelerators for common application domains such as GPUs, NPUs, and DPUs. To address the growing need for efficient offload of operations in datacenter system-on-chips, Intel introduced its Data Streaming Accelerator (DSA) in the 4th Generation Xeon Scalable CPUs, Sapphire Rapids.

The DSA targets data movement operations in memory that are common sources of overhead in datacenter workloads and infrastructure. Its versatility makes it an attractive solution for various workloads, from traditional datacenter applications to emerging use cases like edge computing and IoT. By offloading common software components to DSA, CPU resources can be freed up to run more critical applications, leading to improved overall system performance.

With its ability to efficiently handle data movement operations, CRC32 calculations, delta record creation/merging, and data integrity field (DIF) operations, the Intel Data Streaming Accelerator is poised to revolutionize the way data is processed in modern computing systems. Its future directions include continued innovation and expansion of its capabilities, ensuring it remains a competitive solution for various workloads.

The demise of Dennard scaling has led to a significant shift in the way modern CPUs are designed. With technology process scaling down, semiconductor power density is no longer constant, and as a result, modern CPUs are integrating capable data accelerators on chip. This trend towards core integration with diverse on-chip accelerators is driven by the need for improved performance and efficiency across a wide range of applications and usages.

The rise of specialization has led to the adoption of accelerators for common application domains such as GPUs, NPUs, and, more recently, DPUs. The increasing complexity of datacenter workloads and infrastructure has also led to the integration of on-chip accelerators, such as the Intel Data Streaming Accelerator (DSA). DSA targets data movement operations in memory that are common sources of overhead in datacenter workloads.

The trend towards core integration with diverse on-chip accelerators is not limited to CPUs. Other processors, such as IBM’s Power10, have also adopted this approach by integrating Active Messaging Engines. The benefits of this approach include improved performance and efficiency, reduced latency, and increased application service quality. However, the efficient offload of memory operations from the CPU remains a significant challenge.

The Intel Data Streaming Accelerator (DSA) is a data accelerator introduced in Intel’s 4th Generation Xeon Scalable CPUs, Sapphire Rapids. DSA targets data movement operations in memory that are common sources of overhead in datacenter workloads and infrastructure. In addition to its primary function, DSA has become more versatile by supporting a wider range of operations on streaming data, such as CRC32 calculations, delta record creation/merging, and data integrity field (DIF) operations.

DSA is designed to improve performance and efficiency for a wide range of applications and usages. Its versatility allows it to support various types of data movement operations, making it an attractive solution for datacenter workloads and infrastructure. The DSA’s ability to offload memory operations from the CPU has significant benefits in terms of reduced latency and improved application service quality.

The DSA is a key component of Intel’s 4th Generation Xeon Scalable CPUs, and its integration with other on-chip accelerators has the potential to revolutionize datacenter computing. The DSA’s versatility and ability to offload memory operations from the CPU make it an attractive solution for various applications and usages.

## How Does the DSA Improve Performance and Efficiency?

The DSA improves performance and efficiency by targeting data movement operations in memory that are common sources of overhead in datacenter workloads. By offloading these operations from the CPU, the DSA reduces latency and improves application service quality. The DSA’s versatility allows it to support various types of data movement operations, making it an attractive solution for datacenter workloads and infrastructure.

The DSA’s ability to improve performance and efficiency is demonstrated through a comprehensive evaluation of its throughput benefits. This evaluation shows that the DSA can significantly reduce latency and improve application service quality by offloading memory operations from the CPU. The results of this evaluation provide valuable insights into the potential benefits of using the DSA in datacenter workloads and infrastructure.

The DSA’s performance and efficiency improvements are also demonstrated through an indepth case study of DPDK Vhost. This case study shows how the guidelines for using the DSA can benefit a real application, demonstrating the practical value of the DSA in improving performance and efficiency.

The guidelines for using the DSA are based on a comprehensive evaluation of its throughput benefits and an indepth case study of DPDK Vhost. These guidelines provide valuable insights into how to make the most out of the DSA, including:

• Offloading memory operations from the CPU to improve performance and efficiency
• Using the DSA’s versatility to support various types of data movement operations
• Reducing latency and improving application service quality through efficient offload of memory operations

The DSA guidelines are designed to help developers and system architects make informed decisions about how to use this powerful tool in their applications and workloads. By following these guidelines, users can unlock the DSA’s full potential and improve performance and efficiency in their datacenter workloads and infrastructure.

The integration of on-chip accelerators like the Intel Data Streaming Accelerator (DSA) has significant implications for datacenter computing. The trend towards core integration with diverse on-chip accelerators is driven by the need for improved performance and efficiency across a wide range of applications and usages.

The DSA’s ability to offload memory operations from the CPU has significant benefits in terms of reduced latency and improved application service quality. This approach can revolutionize datacenter computing, enabling faster and more efficient processing of complex workloads and applications.

The implications for datacenter computing are far-reaching, with potential benefits including:

• Improved performance and efficiency through offloading memory operations from the CPU
• Reduced latency and improved application service quality
• Increased flexibility and scalability in datacenter workloads and infrastructure

Overall, the integration of on-chip accelerators like the DSA has significant implications for datacenter computing, enabling faster and more efficient processing of complex workloads and applications.