QuantaGrid-D54Q-2U establishes position in MLPerf inference benchmarks

With an even longer list of vendors from previous years, QCT was named amongst AI inference leaders in the latest MLPerf results released by MLCommons. MLCommons is an open engineering consortium with a mission to benefit society by accelerating innovation in machine learning. The foundation for MLCommons began with the MLPerf benchmark in 2018, which rapidly scaled as a set of industry metrics to measure machine learning performance and promote transparency of machine learning techniques.

In MLPerf Inference v0.7, QCT submitted 2 models showcasing QCT’s capabilities in data center inference. (Shown below: server, CPU, GPU)

• QCT D43KQ; 2 x AMD EPYC 7742 processors; 2 x NVIDIA A100 PCIe
• QCT D52G; 2 x Intel Xeon Gold 6248 processors; 10 x NVIDIA A100 PCIe

For the latest – and now seventh edition – round of MLPerf Inference v3.0, QCT submitted its QuantaGrid-D54Q-2U system to the data center closed division. QCT’s submission included tasks in computer vision, language, and speech areas, achieving 99.9% accuracy in medical image segmentation and 99% accuracy in image classification, language processing, and speech-to-text using its QuantaGrid-D54Q-2U. The system powered by dual 4th Gen Intel® Xeon® Scalable processors provides scalability with flexible PCle expansion slot options, up to 16TB DDR5 memory capacity and 26 drive bays can scale along AI-related workloads. With the innovative hardware design, system tuning and software optimization, the D54Q-2U system equipped with dual Intel Xeon Platinum 8480+ performed 1.2X-1.4X compared to previous version of MLPerf Inference on image classification of server and offline (batch) processing.

QCT has years of experience building supercomputers, cloud solutions, AI infrastructures and was involved in the design of two of the Top500 supercomputers in the world, namely the Taiwania 2 and Taiwania 3. As a global data center solution provider, QCT works closely with its ecosystem of hardware components and software partners to design, manufacture, integrate and service cutting edge offerings for 5G Telco/Edge, AI/HPC, Cloud, and Enterprise infrastructure via its own global network. As a member of MLCommons, QCT actively promotes the MLPerf benchmark suite, optimizing the characteristics of different hardware and software for machine learning and will continue to provide comprehensive hardware systems, solutions, and services to academic and industrial users and share our MLPerf results with the MLCommons community in the MLPerf inference and training benchmarks.To view the QCT submission results please visit: https://mlcommons.org/en/inference-datacenter-30/