Recently, the internationally authoritative AI benchmark evaluation organization MLPerf™ released the latest AI Inference V3.1 ranking. A total of 25 semiconductor, server, and algorithm manufacturers around the world participated in this evaluation. In the fierce competition, H3C stood out in the AI server category and achieved 25 world firsts, demonstrating H3C ‘s strong technological innovation and product development capabilities in the AI field.
MLPerf™ was launched by Turing Award winner David Patterson in conjunction with top academic institutions. It is the world’s most well-known and participated artificial intelligence benchmark test. Including natural language processing, medical image segmentation, intelligent recommendation and other classic model tracks. It provides a fair assessment of a manufacturer’s hardware, software, service training and inference performance. The test results have wide application and reference value. In the current competition for AI infrastructure, MLPerf can provide authoritative and effective data guidance for measuring equipment performance, becoming a “touchstone” for manufacturers’ technical strength in the AI field. With years of focus and strong strength, H3C has won 157 championships in MLPerf.
In this AI Inference benchmark test, the H3C R5300 G6 server performed well, ranking first in 23 configurations in data centers and edge scenarios, and first in 1 absolute configuration, proving its strong support for large-scale, diversified, and advanced applications.ability. Complex computing scenarios.
In the ResNet50 model track, the R5300 G6 server can classify 282,029 images in real time per second, providing efficient and accurate image processing and recognition capabilities.
On the RetinaNet model track, the R5300 G6 server can identify objects in 5,268.21 images per second, providing a computing basis for scenarios such as autonomous driving, smart retail, and smart manufacturing.
On the 3D-UNet model track, the R5300 G6 server can segment 26.91 3D medical images per second, with an accuracy requirement of 99.9%, assisting doctors in rapid diagnosis and improving diagnosis efficiency and quality.
As the flagship of multiple computing capabilities in the intelligent era, the R5300 G6 server has excellent performance, flexible architecture, strong scalability, and high reliability. It supports multiple types of AI accelerator cards, with CPU and GPU installation ratios of 1:4 and 1:8, and provides 5 types of GPU topologies to adapt to the needs of different AI scenarios. Moreover, the R5300 G6 adopts an integrated design of computing power and storage, supporting up to 10 double-wide GPUs and 400TB of massive storage to meet the storage space requirements of AI data.
At the same time, with its advanced AI system design and full-stack optimization capabilities, the R5350 G6 server ranked first with the same configuration in the ResNet50 (image classification) evaluation task in this benchmark test. Compared with the previous generation product, R5350 G6 achieves 90% performance improvement and 50% increase in core count. Equipped with 12-channel memory, the memory capacity can reach 6TB. In addition, R5350 G6 supports up to 24 2.5/3.5-inch hard drives, 12 PCIe5.0 slots and 400GE network cards to meet AI’s demand for massive data storage and high-speed network bandwidth. It can be used in various scenarios such as deep learning model training, deep learning inference, high-performance computing, and data analysis.
Every breakthrough and record-breaking performance demonstrates H3C Group’s focus on customer application scenarios and its accumulation of practical experience and technical capabilities. In the future, H3C will adhere to the concept of “precision agriculture, empowering the era of intelligence”, closely integrate product innovation with artificial intelligence application scenarios, and bring the continuous evolution of intelligent computing power to all walks of life.
Post time: Sep-13-2023
