Single-Device Preparation
Confirm storage capacity, network access, device management and installation requirements for your target models and applications.
H3C LinSeer MegaCube
Technical Specifications
Explore the computing architecture, unified system memory, network ports and deployment requirements.
Compute & Storage
- Computing Platform
- NVIDIA GB10 Grace Blackwell Superchip
- Processor
- 20-core Arm CPU
10 × Cortex-X925 + 10 × Cortex-A725 - GPU Compute Cores
- 6144 CUDA cores, fifth-generation Tensor Cores
- AI Computing Performance
- Up to 1 PFLOPS
Theoretical peak performance at FP4 precision with sparsity. For measured model inference performance, see the report for the relevant configuration. - Unified System Memory
- 128 GB LPDDR5x · 273 GB/s
Shared by CPU and GPU for the operating system, model weights, inference caches and application data. - Storage Options
- 1 TB / 2 TB / 4 TB NVMe M.2
Available configurations are subject to your quotation.
Networking & Ports
- Wired Networking
- 1 × RJ-45, 10GbE
- High-Speed Interconnect
- ConnectX-7 · 200G high-speed interconnect
Dual QSFP ports with RoCE support. - USB
- 4 × Type-C
The official English specifications list USB 4 at up to 40Gb/s. Confirm the functions of individual ports on the actual device. - Display Output
- 1 × HDMI 2.1a
- Wireless
- Wi-Fi 7 · Bluetooth 5.4
Dimensions & Environment
- Chassis Dimensions
- 150 × 150 × 50.5 mm
- Weight
- Approx. 1.2 kg
- Operating System
- NVIDIA DGX OS · Arm architecture
Check compatibility with your applications, containers and model dependencies. - Power & Cooling
- Confirm the power requirements of your chosen model and allow space for the power supply, ventilation and cables.
Power consumption figures differ between official pages. Final values are subject to the delivered specifications.
Deployment Environment & Requirements
Hardware specifications are the starting point. A complete deployment also requires the right models, network and software environment.
Multi-Device Preparation
Choose direct high-speed cabling or a switched network based on device count and configuration, then validate services across nodes using your actual workloads.
Software & Acceptance Testing
Confirm preinstalled components, model versions, PINVOU configuration, software licenses and acceptance tests. Production use of NVIDIA AI Enterprise requires an additional license.
Specification references: H3C official product specifications ↗ · H3C dedicated product page ↗ · FP4 metric definition ↗. Images and specifications are provided to help you choose. Final models, preinstalled components and services are subject to the agreed delivery list.
Explore model deployment and performance validation
Browse model deployments and validation results by device count.