The surge in data transmission has led to the expansion of the server and communications markets. Customer requirements across various sectors, including cloud providers and compute-intensive markets, are diversifying, increasing the complexity of business solutions. Intel's new processors are expected to contribute to customer-centric workload optimization.

On the 11th, Intel unveiled the 4th generation Xeon Scalable Processor (Sapphire Rapids) and Max series product line. At a press conference held at the Federation of Korean Industries building in Yeouido, Intel highlighted its new products, including 12 built-in accelerators that support optimization of individual workloads, security solutions, and next-generation interfaces that will lead a paradigm shift in the industry.

Intel emphasized that the new product will provide improved data center performance and enhanced capabilities for individual workloads such as AI, cloud, and supercomputing.

“The 4th generation Xeon processor is different in that it analyzes and optimizes workloads with a customer-centric approach,” said Na Seung-joo, managing director of Intel Korea. “It provides energy efficiency by embedding the accelerator within the CPU to improve performance and power optimization mode.”

Intel Accelerator Engine Increases CPU Utilization and Reduces Costs
The recently released Sapphire Rapids is said to have the most built-in accelerators of any existing CPU. Through the accelerator engine, it supports workload optimization by demonstrating a wide range of application utilization from hardware to software, including artificial intelligence, data analysis, networking and 5G, storage, and HPC.

Key acceleration technologies include: Advanced Matrix Extension (AMX), which optimizes AI inference and learning reinforcement with 2D tile operations; QuickAssist Technology (QAT), which optimizes fast encryption and bulk data compression; Dynamic Load Balancer (DLB), which optimizes queue scheduling and packet processing; In-Memory Analytics Accelerator (IAA), which optimizes database performance with small memory bandwidth; and Data Streaming Accelerator (DSA), which improves data movement speed.

Intel AMX is an accelerator that enhances deep learning training and inference performance. It can fine-tune AI performance on the CPU and execute small- to medium-sized deep learning models. It is ideal for workloads such as natural language processing, recommendation systems, image recognition, object detection, and machine learning.

DSA improves streaming data movement and transformation, increasing efficiency for storage, networking, and data-intensive workloads. Designed to offload the most common data movement tasks that generate overhead in data centers, Intel DSA helps accelerate data movement across the CPU, memory, cache, and all connected memory, storage, and network devices.

DLB improves system performance related to network data processing on multi-core Sapphire Rapids. It efficiently distributes network processing across multiple CPU cores and threads, and dynamically distributes network data across multiple CPU cores to accommodate changing system loads. Assuming the same processing volume, latency is reduced by 96%, said Executive Director Na.

QAT reduces the workload of processor cores and contributes to power savings by offloading encryption, decryption, and compression. “By utilizing QAT, we can achieve target performance with 84% fewer cores compared to existing software,” said Executive Director Na.

IAA helps database and analytics workloads run faster and more power-efficiently, and is ideal for data stores such as in-memory databases, open source databases, RocksDB, and ClickHouse.

“With these built-in accelerators, we can improve performance per watt by 2.9 times,” said Na Sang-mu. “We can enable power optimization mode to reduce power consumption by up to 20% for specific workloads without compromising performance, and we can monitor power consumption and carbon emissions through telemetry.”

Accelerators help achieve the best performance for each workload, and can increase utilization without increasing the number of CPU cores, contributing to cost savings and improved power efficiency.

■ Adoption of PCIe 5.0 and DDR5, signaling market expansion
With the initial batch of Sapphire Rapids and approximately 1,000 samples reportedly being supplied to customers for testing, expectations are rising that the adoption of next-generation interfaces, PCIe 5.0 and DDR5, will gradually bring about a paradigm shift in the relevant market.

Sapphire Rapids combines up to four Intel 7 process-based tiles in a single package.It supports increased memory bandwidth through DDR5 using Intel EMIB packaging technology and wider I/O bandwidth through PCIe 5.0 and CXL 1.1 interconnect.

It also features advanced security solutions, including security engines such as software guard extensions, crypto acceleration, total memory encryption, and platform firmware resilience.

Intel emphasized that its modular architecture allows it to offer a wide range of processors, with over 50 SKUs tailored to individual workloads, from mainstream general-purpose SKUs to purpose-built SKUs.

“As the 4th generation Xeon processor ramps up, we expect the DDR5 market to grow as well,” said Executive Director Na. “There are over 400 Xeon-based designs under development, which seems to be a testament to the interest in the 4th generation Xeon as well,” he said, confident of market leadership.

The next-generation processor, Emerald Rapids, is also scheduled for production later this year. Despite the delay, Emerald Rapids is said to have undergone a much shorter testing period, as it will be built on the same platform as the recently released Sapphire Rapids.