Top 10 Must‑Watch DPU Talks from NVIDIA GTC China 2020: AI, 5G, Edge & More

The 2020 GTC China online conference attracted over 50,000 registrants and featured more than 200 technical talks across 14 sub‑forums. This article selects ten highlighted presentations from NVIDIA, UCloud and Shanghai Jinstar Software that focus on DPU applications in finance, 5G, edge computing, data science and high‑performance networking.

UCloud Kuajie Cloud Host: Software and Hardware Optimization

Speaker: Peng Jingxin, UCloud Storage Director

Kuajie cloud host is UCloud's latest generation compute product with 10 M pps internal network performance and 1.2 M IOPS storage performance. By leveraging SR‑IOV and RDMA on NVIDIA ConnectX‑5 25G NICs, internal network throughput rose from 300 k pps to 10 M pps, storage IOPS increased from 30 k to 1.2 M, and latency dropped from 1 ms to 100 µs, delivering near‑bare‑metal performance for databases, messaging queues and big‑data workloads.

BlueField‑2 DPU and SNAP in the Financial Industry

Speaker: Zhang Songbai, NVIDIA Senior Systems Engineer

NVIDIA DPU SNAP technology emulates remote storage as local, supporting NVMe and Virtio devices. This enables diskless compute nodes, full utilization of backend storage pools, reduced failure rates, and lower capital and operational costs, meeting the high‑performance demands of AI, big data and HPC workloads in finance.

DPU Accelerates 5G Mobile Network Communication

Speaker: Shen Yuxi, NVIDIA Solution Architect

DPU provides precise clock synchronization and scheduling, combined with GPU and general‑purpose servers, to support O‑RAN CU/DU implementations and accelerate 5G core and edge UPF devices, fulfilling the high‑performance requirements of modern 5G deployments.

NVIDIA Ethernet Products Accelerate Data Science Applications

Speaker: Wang Dong, NVIDIA China Ethernet Product Director

Data science workloads demand high‑performance networking; NVIDIA Ethernet solutions deliver the bandwidth, low latency, and reliability required for modern data‑science pipelines.

DPU Accelerates Edge Computing

Speaker: Feng Gaofeng, NVIDIA HPC‑AI Technology Market Director

The latest NVIDIA DPU, built on the DOCA architecture, provides breakthrough network, storage, and security performance for edge computing, offering programmable, zero‑trust security to protect data and workloads.

Intelligent Network Accelerates 5G Cloud Services

Speaker: Shen Yuxi, NVIDIA Solution Architect

Smart NICs offload core network functions, enabling high‑performance, low‑latency 5G services. Precise clock sync and scheduling, combined with GPU acceleration, support O‑RAN CU/DU and high‑performance UPF implementations.

NVIDIA Network Technology for Fast Financial Systems

Speaker: Zhang Peng, Shanghai Jinstar Software, Trading‑Product Pre‑Sales Lead

By applying NVIDIA RDMA networking, Jinstar reduces latency in financial trading systems, improving market price discovery and enabling high‑frequency, low‑delay transactions.

NVIDIA Ethernet Empowers Next‑Gen Edge Cloud Networks

Speaker: Wang Chenxi, NVIDIA Senior Pre‑sales Engineer

The ESF solution uses high‑performance, low‑latency Ethernet switches and RoCE to quickly build flexible, small‑form‑factor edge cloud networks for remote work, CDN, telecom edge, and autonomous driving.

Building Open Network Infrastructure to Accelerate Modern Data Centers

Speaker: Wang Dong, NVIDIA China Ethernet Product Director

Open network technologies replace closed architectures, meeting the scalability, flexibility, and AI‑driven performance requirements of modern data centers.

Rivermax – Easy Entry to Ultra‑HD IP Video

Speaker: Chen Long, NVIDIA Senior Marketing Manager

ConnectX NICs with kernel bypass, GPU‑direct and virtualization support enable 10‑200 Gbps video streaming, replacing traditional SDI solutions for 4K/8K broadcasting.