Author丨Lu Dongxue
Editor | Chen Caixian
The digitalization of enterprises is coming fiercely, and the infrastructure of intelligent technologies such as computing power and data required for digitalization is often neglected.
However, when cost reduction and efficiency increase become one of the guidelines for enterprise development, all aspects involved in the implementation of digital projects cannot be underestimated. The data and computing power driving algorithms and models have also become an important means to achieve “cost reduction and efficiency increase”.
For example, suppose the delivery time of a vision algorithm supplied to terminal hardware is six months, and the algorithm training and tuning is originally two months. The company deploys two algorithm engineers, and the annual salary of an algorithm engineer is 1 million. If the algorithm training The speed can be increased by 10 times, so the direct impact on the enterprise is four words: real money!
Of course, this is the goal that computing service providers have been working hard for. However, is it possible to increase the real-time reasoning speed of AI algorithms by 10 times on PyTorch?
Intel did it.
At the 4th Generation Intel® Xeon® Scalable Processor Launch Conference held on January 11, Intel officially launched the 4th Generation Intel® Xeon® Scalable Processor, Intel® Xeon® CPU Max series and Intel ® Data Center GPU Max Series.
The fourth-generation Xeon not only accelerates AI inferencing, but also drives dramatic leaps in data center performance, energy efficiency, and security, bringing new capabilities to AI, cloud, networking, edge, and the world’s leading supercomputers.
After many years of gaming with Nvidia, Intel has found another way to invest in the Roman road that helps companies move towards digitalization. It is committed to improving infrastructure such as computing and data, and has become the person who lays the foundation and paves the cement road for the digital road.
1
“Speed up” is the basic disk for stronger computing power
As the digital transformation of enterprises enters the deep-water area, the application of multi-cloud, edge computing, and artificial intelligence technologies within enterprises has gradually become the norm. A more cost-effective computing power solution.
However, we all know that the CPU is one of the core hardware that provides computing power. Only stacking the number of CPU cores can only strengthen the basic computing power, while the built-in acceleration can improve the performance of specific applications more efficiently. When faced with today’s brand-new enterprise computing power demands, the 4th Generation Intel® Xeon® Scalable processors emerged to provide performance acceleration for the fastest-growing workloads in artificial intelligence, data analysis, storage, and scientific computing.
The 4th generation Intel® Xeon® Scalable processors are manufactured using the Intel 7 process technology and feature a new chip architecture. Acceleration is ubiquitous by integrating high-performance cores, more cores, relevant accelerators for high-demand data center workloads in the industry, and industry-leading DDR5, CXL1.1, and PCIe 5.0.
In addition, 4th Generation Intel® Xeon® Scalable processors enable technologies that scale and adapt as workload requirements change:
-
Further boost network, storage, and compute performance while increasing CPU utilization by offloading heavy tasks to the Intel® Infrastructure Processing Unit (Intel® IPU);
-
Increased multi-socket bandwidth (up to 16 gigatransfers per second [GT/s]) with Intel® UPI 2.0;
-
Use Intel® Speed Select Technology (Intel® SST) to configure CPUs to meet specific workload needs;
-
Added L3 shared cache (LLC) (up to 100 MB LLC shared by all cores);
-
Security enhanced through hardware;
-
Eliminate the need for a separate RAID card with Intel® Virtual RAID on CPU (Intel® VROC).
It is worth mentioning that the 4th generation Intel® Xeon® Scalable processors have many built-in accelerators, which can efficiently cope with artificial intelligence, data analysis, network, storage and other high-demand workloads, which can be constantly changing and demanding Provides considerable computing performance in data center usage and is optimized for workloads. Compared with the previous generation, it can improve the average performance per watt of target workloads by 2.9 times, and can save up to 70 watts per CPU and reduce 52 % to 66% total cost of ownership (TCO), which certainly represents another major step forward for the industry in data center platforms.
The ubiquitous acceleration experience comes from the seven computing power artifacts provided by the 4th generation Intel® Xeon® Scalable processors:
-
Intel® Advanced Matrix Extensions (Intel® AMX): Dramatically boosts deep learning training and inference performance by up to 10x, making it ideal for workloads such as natural language processing, recommendation systems, and image recognition.
-
Intel® Dynamic Load Balancer (Intel® DLB): Enables efficient distribution of network processing across multiple CPU cores/threads, dynamically distributes network data across multiple CPU cores for processing based on changes in system load, and It can restore the order of network data packets processed simultaneously on multiple CPU cores, significantly improving the system-level processing performance of network workloads, and reducing latency by 96%.
-
Intel® Data Streaming Accelerator (Intel® DSA): Designed to offload common data movement tasks that incur the overhead of datacenter-scale deployments, and helps accelerate CPU, memory, and cache, as well as all attached memory, storage, and networking devices Data movement between data-intensive workloads for storage, networking, and data-intensive workloads by up to 1.7x.
-
Intel® In-Memory Analytics Accelerator (Intel® IAA): Improves query throughput while reducing memory footprint, ideal for in-memory databases, open source databases, and data stores such as RocksDB and ClickHouse, up to 3x performance.
-
Intel® Data Protection and Compression Acceleration Technology (Intel® QAT): A built-in accelerator that frees up processor cores by offloading encryption, decryption, and compression, allowing the system to support more clients or reduce power consumption. Thanks to the Intel® QAT Accelerator, 4th Generation Intel® Xeon® Scalable processors are the highest performing CPUs to compress and encrypt data for a single stream.
-
Intel® Security Technology: Protect workloads with Intel® Software Guard Extensions (Intel® SGX) and Intel® Total Memory Confidentiality (Intel® TME); accelerate cryptographic operations with Intel® Crypto Acceleration ; Intel® Memory Fault Management Technology (Intel® MRT) for predictive security protection; and Intel® Platform Firmware Resilience Technology (Intel® PFR) for platform secure boot.
-
Intel® Xeon® CPU Max series: Integrated high-bandwidth memory to accelerate scientific computing workloads without code changes.
2
Accelerator “blessing”
Adapt to more business scenarios
In addition to being fast, the processor also needs to adapt to more business scenarios. Whether for enterprises or developers, the practicability and universality of processors in business scenarios is very important.
At the new product launch conference of the 4th generation Intel® Xeon® Scalable processor, Intel shared workloads such as artificial intelligence, data analysis, network and storage, which gave us a lot of thinking and let us see its computing power Exploration and breakthrough in processing and application scenarios.
In terms of AI, compared with the previous generation, with built-in Intel® AMX, the 4th generation Intel® Xeon® Scalable processors have increased the performance of PyTorch real-time inference and training by 10 times, and combined the inference and training of a wide range of AI workloads. Training performance is taken to new heights. The Intel® Xeon® CPU Max family builds on these capabilities by extending them for natural language processing, processing large language models up to 20 times faster.
At the same time, with Intel’s AI software suite, developers can use the AI tools of their choice to increase productivity and speed up AI development. The suite is portable from the workstation, making it scalable in the cloud and all the way to the edge. At the same time, the suite has been validated against more than 400 machine learning and deep learning AI models, covering the most common AI application scenarios in all business segments.
For networking applications, the 4th Generation Intel® Xeon® includes a series of processor products optimized for high-performance, low-latency network and edge workloads. These processors will play a key building block as industries such as telecommunications, retail, manufacturing and smart cities are driving a more software-defined future.
For 5G core workloads, built-in accelerators help increase throughput and reduce latency, while upgrades to power management improve platform responsiveness and energy efficiency. Additionally, 4th Gen Intel® Xeon® delivers up to twice the virtual radio access network (vRAN) capacity compared to previous generations without increasing power consumption. This enables communication service providers to double the performance per watt, meeting their critical performance, scaling and energy efficiency needs.
In the field of data centers, in addition to the ultimate pursuit of performance, security and algorithm privacy protection also need to be strengthened. With the migration of data to the cloud, the emergence of scenarios such as artificial intelligence, machine learning, multi-party participation, and collaborative work, the demand for decentralized data and algorithm protection has become more significant. It is necessary to provide practical, effective, and implementable platform security capabilities. In order to meet the needs of customers from all walks of life for security solutions. The 4th generation Intel® Xeon® Scalable processors can help implement a zero-trust security strategy, providing new solutions for business deployment and risk insight through advanced hardware security technologies, including Intel® SGX, Intel® TME, Algorithm hardware acceleration, etc. provide customers with comprehensive security protection.
3
Processors for the Data Center of the Future
Solving cloud problems is “just needed”
At present, while the cloud computing market is developing rapidly, cloud services are moving towards subdivision and deepening, and the entire industry presents new development characteristics in terms of technology, application, and management. For example, with the high growth of cloud demand in traditional industries, government and enterprise industries, and more enterprises choosing to use the cloud in depth instead of simply migrating to the cloud, it not only brings more consumption of cloud basic resources and cloud platform products, but also affects Security, performance, and latency have put forward higher requirements, which have also made cloud native, distributed storage, private cloud, hybrid cloud, cloud security, computing power services, etc. become areas of concern at present, driving cloud service vendors to increase Large investments in cost performance, security, and technical optimization of cloud products will promote cloud services to enter a new stage of improving quality and efficiency.
As cloud migration progresses across the enterprise, the cloud service environment becomes more complex. A synchronized business and technology strategy driven by industry best practices can enhance the agility, cost-efficiency, and other benefits of cloud-first operations. Enterprises need to increase efficiency and improve the predictability of service levels, while accelerating innovation and better supporting business decision-making. Intel® Xeon® Scalable processors continue to optimize software and hardware for cloud computing applications, and have always had the advantages of high efficiency, flexibility, and reliability. The 4th Generation Intel® Xeon® Scalable processors are more powerful in terms of workload performance on the cloud through numerous built-in accelerators and software optimizations.
In the cloud-native direction, microservices are still the main driving factor for the internal transformation of most enterprises, especially in the ultra-large-scale platform or edge cloud. Run better and faster at scale. Quality of service, infrastructure overhead, and observability are issues that enterprises have been iterating on in terms of microservices. To solve this problem, cloud hardware and software must be optimized separately.
Intel has also paid attention to these challenges. The built-in accelerators of the fourth-generation Intel® Xeon® Scalable processors can accelerate data movement between cloud platforms and within cloud platforms, effectively guaranteeing service quality, infrastructure overhead, and data and process observability operational goals. Developers get hardware-assisted queuing through Intel® DLB to help load balance millions of incoming requests per second; fast distributed network communication through cryptographic acceleration through Intel® QAT; and accelerated data transfer through Intel® DSA .
At the same time, there will be better performance in terms of microservices when used in conjunction with Intel’s other data center technologies. For example, Intel® Infrastructure Processing Unit (Intel® IPU) can be configured to release more CPU core resources, improve application performance, and optimize the overall utilization of the data center. Intel® Ethernet 800 Series Network Adapters can also be configured to accelerate high-priority applications, packet processing, and latency-sensitive workloads.
From cloud computing, new media, audio and video, cloud games to mobile apps, the rapid development of multiple cloud applications is inseparable from the database. From traditional relational databases to cloud databases, databases are constantly evolving. At the same time, it is also playing an increasingly important role. On the one hand, as a key part of the IT infrastructure, the database plays a supporting role in the development of the business of the enterprise; database. How to improve database performance has also become an important exploration of cloud applications.
In order to achieve higher speed and capacity levels, optimize database performance, balance TCO, and meet requirements for CRM systems and other business-critical databases, the 4th generation Intel® Xeon® Scalable processors can be equipped with Intel® IAA. Compared with the previous generation of products, it can improve the performance of RocksDB by up to 3 times; improve the performance of ClickHouse DB by up to 59%. At the same time, you can also configure the Intel® Optane 
Persistent memory, increase the density of database instances per node, and get more memory available at a similar system construction cost.
Security is also a top priority in cloud computing. Data at rest or in transit needs protection, as does data in use. With the 4th Generation Intel® Xeon® Scalable processors, Intel offers the data center chip industry’s broadest confidential computing portfolio while protecting sensitive data and data subject to strict laws and regulations. As a chip maker that provides application isolation for data center computing, Intel, with Intel® SGX, provides hardware encryption for data in use in private cloud, public cloud, and cloud-to-end environments to achieve a smaller attack surface and Leverage the cost-savings and scalability of cloud computing and easily scale to run large applications or databases across distributed networks while better ensuring cloud security and data compliance.
In addition, as Intel’s new virtual machine isolation technology, Intel® Trust Domain Extensions (Intel® TDX) is very suitable for porting existing applications to confidential environments. This feature will be launched in Microsoft Azure, Alibaba Cloud, Google Cloud and IBM Cloud.
4
Looking at the Future Data Center from the Processor
sustainable development
In 2021, the Ministry of Industry and Information Technology issued the “Three-Year Action Plan for the Development of New Data Centers (2021-2023)”, which proposed computing power indicators for the first time, aiming to guide data centers from extensive rack-scale growth to high-quality development of computing power Evolution, to better promote the quality change and efficiency change of the data center.
The next year will be a critical year for the acceptance of the new data center’s implementation results. The effect of computing power improvement and its implementation in various industries, the promotion effect of the enterprise’s in-depth cloud usage and digital transformation in key areas will all become important indicators. In this context, the data center will continue to transform towards flexibility and agility, high performance, high security, high stability, and excellent user experience, and help enterprises adapt to the load requirements of more scenarios, and help enterprises through green computing power The enterprise achieves the goal of sustainable development.
In addition, relevant agencies predict that by 2035, 5G, the Internet, and artificial intelligence will bring a growth of 40 trillion US dollars to the global economy, and computing power is an important resource support and infrastructure behind it. Under the strategic background of global carbon peaking and carbon neutrality, “how to improve efficiency and reduce energy consumption” is a grand proposition, which also puts forward high requirements for manufacturers who support green and low-carbon development through technological innovation.
As the most sustainable data center processor Intel has released to date, 4th Generation Intel® Xeon® Scalable processors are a major step forward in helping businesses meet their sustainability and emissions reduction goals. In addition to built-in accelerators that help increase performance-per-watt for specific workloads, it has a range of features to manage power and performance, and uses more than 90 percent renewable electricity in its manufacturing process.
Built-in accelerators are a more efficient way to improve workload performance than adding CPU cores. By building many accelerators, it can provide platform-level power savings and alleviate the need for additional independent acceleration, helping customers achieve sustainable development goals. Additionally, new optimized power modes for 4th Generation Intel® Xeon® Scalable processors can deliver up to 20 percent socket power savings for certain workloads with less than a 5 percent impact on performance. Innovations in air and liquid cooling further reduce the overall energy consumption of the data center.
In addition, through the Intel® On Demand activation mode, Intel supports users to activate additional built-in accelerators and functions in the CPU in the field and in production. The upgrade can be completed on top of the basic configuration of the original SKU. Enterprises can flexibly transform or upgrade the CPU to Support new workloads and tap greater value, which greatly improves the enterprise’s input-output ratio.
It is not difficult to see that Intel will continue to work closely with customers and industry partners to provide differentiated solutions and systems on a large scale based on the brand-new 4th generation Intel® Xeon® Scalable Platform to help customers solve computing power challenges and jointly achieve sustainable Continuous development.
Future-oriented chip manufacturers should be results-oriented, adhere to the workload-oriented strategy, highly optimize software for specific workloads, match corresponding power consumption and performance for different workloads and requirements, and achieve ideal total cost of ownership. And Intel has undoubtedly given the perfect answer through the new fourth-generation Intel Xeon Scalable processor.
For more content , click below to follow:
Leifeng.com
This article is transferred from:https://www.leiphone.com/category/ai/mLx2TTX0k7RH1tfJ.html
This site is only for collection, and the copyright belongs to the original author.