Showcase: OCP Global Summit 2019
Expanding beyond hardware, Inspur explores new avenues to enable open infrastructure benefits in broader applications: dynamic AI solutions for the open data center, high-density cloud optimized platforms, and rack management and open RMC software for OCP.
19" Standard Server Rack Solutions
4-Socket Server for Hyperscale Cloud Workloads
Olympus 4-Socket Server: NF8380M5
Based on Microsoft’s Project Olympus open standard to provide a modular and flexible open compute standard for cloud workloads.
High-Density Cloud-Optimized Platform
Flexible High-Density 2U 4-Socket Server: NF8260M5
High-density, flexible server with ultra-dense memory design, scalable 4P performance, up to eight PCI-E expansion cards and two 2000W power supplies. Delivers 18TB of memory in a single platform and up to 360TB in a 42U rack. Fully optimized with Intel’s latest Optane DC persistent memory.
Advanced New Architecture for AI
This configuration offers increased computational efficiency through a 4-socket design, and delivers 0.5TB aggregated memory at 16TB/s.
- higher GPU P2P bandwidth of 2.4TB/s
- higher GPU memory bandwidth of 16TB/s
- greater full bandwidth GPU clustering aggregation of 16 GPU
4-Socket Server for Hyperscale Cloud Workloads: NF8380M5
Based on Microsoft’s Project Olympus, delivers M.2 riser flash array expandability for GPU, FPGA, and NVMe to enable flexible configurations for multi-purpose applications.
8U 16GPU Supercomputing Server: AGX-5
Equipped with 16 NVIDIA Tesla® V100 Tensor Core GPUs, delivers 2 Petaflops of performance and NVSwitch™ fabric interconnectivity.
21" OCP Rack Solutions
Composable Infrastructure and Open RMC
Inspur’s GPU Composable Infrastructure rack, integrated with Liqid Composable, delivers unparalleled rack-scale agility and enables dynamic configuration, allocation and deployment of GPU elements.
High Expansion OCP Compute Node: ON5293M5
Provides high I/O expansion supporting 1 PCIe 3.0 x16 or 2 PCIe 3.0 x8, ideal for data acceleration.
High-Density NVlink GPU Expansion: ON5388M5
This high-density JBOG features an NVlink-enabled architecture that provides flexible topology for different applications.
I/O Expansion OCP Compute Node ON5293M5 + Intel® Programmable Acceleration Card (Intel® PAC)
Compute node’s high I/O expansion paired with Intel PAC® and OpenVINO enables massive data acceleration and rapid AI scaling.
Inspur OCP Contributions
As a longtime collaborator in the OCP community, Inspur has contributed a number of hardware innovations in compute, storage and GPU platforms based on the newest open standards.
OCP Compute Nodes
3-node compute nodes for different applications:
OCP AI Node: ON5488M5
Designed for hyperscale data center AI training applications. Ultra-high GPU density and optimized architecture for AI.
OCP Storage JBOD: ON5266M5
34 hard-drive 2OU JBOD with NVMe SSD/HDD support and flexible architecture.