Quanta Cloud Technology continues to push the boundaries of innovation in high-performance computing (HPC) and artificial intelligence (AI). With a commitment to delivering top-tier performance, QCT participated in the latest MLPerf Inference v5.0 benchmark, showcasing their prowess across various system configurations in the data center closed division.

Diverse System Configurations for Varied Workloads

In this round of MLPerf Inference v5.0 submissions, QCT highlighted their versatile system designs, catering to both CPU-centric and GPU-centric architectures. This approach ensures that they meet the diverse needs of modern computational workloads.

CPU-Centric Systems:

• QuantaGrid D55X-1U: This high-density 1U server is powered by dual Intel^® Xeon^® 6 processors. Supporting the Intel^® AMX instruction set, it delivers optimized inference performance across a range of AI models including ResNet-50, RetinaNet, 3D-UNet, and DLRMv2. It also supports small language models like GPT-J-6B, providing a cost-efficient alternative to GPU-based solutions.

GPU-Centric Systems:

• QuantaGrid D54U-3U: A flexible 3U x86_64-based platform that supports up to four dual-width or up to eight single-width PCIe GPUs including the NVIDIA H100 PCIe GPU. This allows users to customize configurations based on their different workload requirements.
• QuantaGrid D74H-7U: Designed for large-scale AI workloads, this powerful 7U system supports up to eight NVIDIA H100 SXM5 GPUs. It leverages the NVIDIA NVLink™ interconnect for high-speed GPU-to-GPU communication and supports GPUDirect Storage for ultra-fast, low-latency data transfers.
• QuantaGrid S74G-2U: A next-generation 2U system based on NVIDIA GH200 Grace Hopper™ Superchip. The integration of CPU and GPU via NVLink C2C interconnect establishes a unified memory architecture, enhancing computational efficiency and seamless data access.

Commitment to Transparency and Performance

QCT’s comprehensive AI infrastructure solutions cater to a wide range of users, from academic researchers to enterprise clients, who require robust and scalable AI infrastructure. By actively participating in MLPerf benchmarking and openly sharing their results, QCT underscores its dedication to transparency and reliability. This empowers customers to make informed, data-driven decisions based on validated performance metrics.

In conclusion, QCT’s participation in the MLPerf Inference v5.0 benchmark not only demonstrates their cutting-edge capabilities but also their unwavering commitment to advancing AI and HPC technologies. Their diverse and powerful system configurations ensure that they remain at the forefront of innovation, ready to meet the evolving demands of modern computational workloads.