Global Site
Breadcrumb navigation
The Challenge
As AI inference, data processing, and shared research workloads continue to grow, organizations are investing heavily in GPU infrastructure. However, conventional server-centric architectures often lead to low GPU utilization, rising power and cooling demands, poor investment efficiency, and limited operational flexibility.
| The Challenge | What NEC CDI Solution Delivers |
|---|---|
| GPU resources are statically attached to servers, resulting in idle GPUs and low overall utilization. | Pools and dynamically allocate GPUs to servers based on workload demand, improving utilization and operational efficiency. |
| Increasing GPU power consumption and cooling requirements make data center expansion difficult. | Separates GPUs from compute servers, enabling flexible placement in locations optimized for power and cooling. |
| Expanding GPU capacity often requires purchasing additional GPU servers, increasing capital costs. | Allows independent scaling of servers and GPUs, improving investment efficiency and reducing unnecessary hardware purchases. |
| GPU or server failures can interrupt services and reduce resource availability. | Supports resource pooling and redundant network configurations to improve availability and operational resilience. |
Use Cases
NEC CDI Solution addresses common infrastructure challenges in research IT environments where GPU resources must be shared, scaled, and operated efficiently.
 |
 |
 |
|---|---|---|
| Campus-wide GPU Resource Pooling | Dynamic GPU Allocation for AI Inference and Data Processing | Remote GPU and PCIe Device Sharing Across Buildings or Data Centers |
| Use Case | Setting | What NEC CDI Solution Enables |
|---|---|---|
| Campus-wide GPU Resource Pooling | Universities and research institutions with multiple departments sharing GPU resources | Shared GPU pools enable efficient utilization, centralized management, and fair resource allocation across organizations. |
| Dynamic GPU Allocation for AI Inference and Data Processing | AI inference platforms, shared research IT environments, and data processing workloads with changing resource demands | GPU resources can be reassigned according to workload demand, improving utilization while reducing infrastructure costs. |
| Remote GPU and PCIe Device Sharing Across Buildings or Data Centers | Distributed research facilities, multi-building campuses, and data centers | Compute servers and GPUs/PCIe devices can be physically separated while maintaining near-native PCIe device access over Ethernet. |
Features
NEC CDI Solution with ExpEther technology enables GPUs and other PCIe devices to be dynamically assigned, shared, and managed across research and AI environments, improving utilization, flexibility, and operational efficiency.
Flexible Resource Composition
- Enables dynamically assigning GPUs and PCIe devices to servers according to workload requirements
- No application modification is required, as PCIe devices appear as locally attached resources
Data Center-scale Disaggregation
- Separates servers and PCIe devices across rooms, floors, or buildings
- Supports long-distance connectivity through Ethernet-based ExpEther technology for a maximum of 2 km
High Performance
- Provides low-latency PCIe device access over Ethernet for AI inference and shared accelerator use cases
- Enables efficient utilization of high-performance accelerators
High Availability
- Dual-path redundancy for PCIe connectivity improves service continuity
- Automatic path switching helps maintain operation when failures occur
Simplified Operations
- Centralized resource management through NEC CDI Manager Software
- Improves monitoring, allocation, and lifecycle management of shared resources
Architecture
NEC CDI Solution combines compute servers, Ethernet networking, ExpEther technology, and external PCIe device pools into a flexible composable infrastructure. Resources can be assigned to servers on demand while remaining centrally managed.
| Host Server | Runs AI inference, data processing, and research applications |
| ExpEther Board | Extends PCIe connectivity across Ethernet networks |
| 100G Ethernet Switch | Provides high-speed interconnection between servers and resource pools |
| Expansion IO Box | Hosts GPUs, SSDs, FPGAs, and other PCIe devices |
| NEC CDI Manager | Provides centralized monitoring and resource management |
Contact
To learn more about NEC CDI Solution with ExpEther technology and explore how GPU resource pooling can improve your AI and research infrastructure, please get in touch.