Shulou(Shulou.com)11/24 Report--

Recently, MLCommons released the results of the industry standard MLPerf training v3.1 benchmark for AI model training. Among them, the fourth-generation Intel ®strong ®scalable processor with built-in Intel ®Advanced Matrix extension (Intel ®AMX) acceleration engine, as the only CPU that submitted MLPerf test results, showed strong performance in the test, further confirming Intel's commitment to accelerate the large-scale deployment of AI in cloud, network, edge, and end workloads.

This time, Intel submitted test results for ResNet50, RetinaNet, BERT and DLRM dcnv2. It is worth noting that in DLRM dcnv2, a newly submitted test model, the fourth-generation Intel Xeon scalable processor was trained in 227 minutes using only four nodes. For ResNet50, RetinaNet and BERT, the fourth-generation Intel Xeon scalable processors also demonstrate powerful out-of-the-box performance.

Fourth generation Xeon scalable processors have excellent performance in general AI workloads

In three tests conducted in June, September and November this year, Intel submitted a number of reasoning benchmarks based on the fourth generation Intel Xeon scalable processor. The results show that it includes vision, language processing, voice and audio translation models, as well as the larger DLRM v2 deep learning recommendation model, 6 billion parameter large language model and computer vision and natural language processing model ChatGPT-J. The fourth generation Intel Xeon processors have excellent performance for general purpose AI workloads. Intel also submitted test results in cooperation with OEM vendors, further demonstrating the scalability of its AI performance and the accessibility of general-purpose servers based on Intel Xeon processors, fully meeting the customer Service level Agreement (SLA).

The powerful performance of the fourth generation Intel Xeon scalable processors provides enterprises with "out-of-the-box" functionality. AI can be deployed on general-purpose systems for data preprocessing, model training and deployment, thus achieving an optimal combination of AI performance, efficiency, accuracy and scalability, avoiding the high cost and complexity of introducing dedicated AI systems, which covers multiple frameworks and end-to-end data science tools. And a broad ecosystem of smart solutions. Many enterprises can use generic CPU to train small and medium-sized deep learning models cost-effectively and sustainably on their existing enterprise IT infrastructure, especially for use cases where the training object is intermittent workloads.

So far, 1 million fourth-generation Xeon scalable processors have been shipped and have been widely used by local ecological partners in many fields because of their strong AI performance. In the next-generation BCC instance of Baidu Intelligent Cloud released in September, the fourth-generation Zhi Qiang scalable processor expands AI computing power with AMX accelerator, thus enabling Baidu Intelligent Cloud users to easily obtain native powerful AI capabilities on any instance. Thanks to the powerful acceleration ability of Intel AMX instruction set for matrix operations, the BF16 throughput of Tencent BERT model has been greatly improved, thus effectively optimizing the terminal experience of its AI users. Through the different precision data processing of AMX INT8 and BF16, the in-depth tuning of AVX-512, and the blessing of Intel software, Aliyun address standardization business and Taobao search customized recommendation business have achieved significant performance improvement.

At the same time, the fourth-generation Intel Xeon scalable processors have also helped independent software service providers such as AsiaInfo, user Friends, Kingdee and Neusoft to upgrade their businesses in areas such as OCR, helping them successfully cope with the diversified AI workload requirements from different application scenarios.

Committed to promoting AI everywhere, Intel not only creates hardware products including the fourth generation Intel Xeon scalable processors, but also provides all-round assistance to ecological partners through open source software stacks and open ecology, aiming to meet the dynamic requirements of multiple performance, energy efficiency and ease of use of thousands of industries, and further promote the landing of AI applications.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.