Solution
Tailored solutions to address industry challenges
High Performance Computing (HPC) has become one of the most important computing foundations driving the digital economy, scientific research, AI innovation, industrial simulation, and advanced engineering applications. The computing density, operational stability, thermal efficiency, and data transmission performance of HPC systems directly determine research capability, industrial competitiveness, and technological breakthroughs.
High Performance Computing (HPC), as a core foundation of modern computing infrastructure, is widely used in supercomputing centers, scientific simulation, AI training, financial quantitative analysis, weather forecasting, bioinformatics, and other mission-critical fields. It supports large-scale parallel computing, massive high-speed data processing, and high-concurrency computing workloads.
As the core hardware carrier for HPC clusters, server chassis house key components such as high-performance CPUs, GPU accelerators, high-speed storage, and interconnect modules. They play a critical role in ensuring stable system operation, continuous computing power output, and high-speed data transmission.
Unlike the security industry, which focuses more on protection, surveillance, and long-term data retention, the HPC industry focuses on:
High-density computing integration
Advanced thermal efficiency
High-speed interconnect compatibility
Stable continuous operation
Flexible expansion and upgrade capability
Standard server chassis can no longer fully meet the demands of HPC environments, which are characterized by high computing density, heavy heat output, complex hardware integration, and strict low-latency interconnect requirements.
Based on deep customization capabilities, this solution focuses on the core scenarios and challenges of HPC high-performance computing. It provides full-chain customization services from single high-performance chassis to full-rack cluster systems, helping research institutions and enterprises build efficient, scalable, and stable HPC infrastructure while maximizing computing performance.
This solution is built around four core principles:
High-density integration
Advanced thermal management
High-speed interconnect
Stable and reliable operation
It is designed for key HPC scenarios including:
Supercomputing center clusters
AI training clusters
Scientific simulation nodes
Edge high-performance computing
The solution precisely matches HPC workloads that require intensive computing power, high-speed data flow, and long-term high-load operation.
The internal chassis layout is optimized to break through the space limitations of standard chassis designs. It supports high-density integration of multiple CPUs, GPUs, and memory modules, maximizing rack space utilization and improving computing output per cabinet.
This helps customers support large-scale HPC cluster deployment while reducing data center space requirements and deployment costs.
Customized thermal systems are designed for the high heat output generated by HPC workloads. Through optimized airflow and cooling architecture, core hardware components such as CPUs and GPUs are maintained within safe temperature ranges.
This prevents thermal throttling, hardware failure, and computing performance loss, ensuring continuous and stable computing power output.
The chassis reserves sufficient high-speed interconnect interfaces and expansion slots, supporting:
PCIe 5.0 / PCIe 6.0
InfiniBand
Ethernet
100G / 400G high-speed networking
Internal cable routing is optimized to reduce data transmission latency and ensure efficient data exchange between nodes and hardware modules.
The solution adopts highly reliable structural and redundant designs. Core components such as power supplies and cooling fans support N+1 redundancy.
With MTBF of more than 150,000 hours, the system is suitable for 7×24 high-load continuous operation, reducing the risk of computing task interruption, data loss, and maintenance costs.
The modular design supports rapid upgrades and expansion of CPUs, GPUs, storage, and interconnect modules without replacing the entire chassis.
This extends hardware lifecycle, lowers investment costs, and adapts to the fast evolution of HPC computing power requirements.
Supercomputing centers support large-scale scientific computing, weather forecasting, astrophysics simulation, and other demanding workloads. They require extremely high computing density and interconnect speed.
Key challenges include:
Multi-CPU and multi-GPU integration per node
High power consumption and concentrated heat output
Node-to-node latency requirement ≤10μs
High rack space utilization
Long-term high-load operation
Strict redundancy and reliability requirements
Fast expansion for growing computing demand
Mainly based on 2U / 4U rackmount chassis, the internal structure is deeply optimized for compact high-density deployment.
A 4U chassis can support:
2 high-end Intel Xeon / AMD EPYC CPUs
8–16 GPU accelerators
Or 4 CPUs + 8 GPUs
Up to 2TB DDR5 memory
16–24 NVMe high-speed drive bays
Computing density is improved by more than 60% compared with standard chassis.
Reinforced SECC galvanized steel with 1.2–1.5mm thickness provides load-bearing capacity of more than 150kg, preventing chassis deformation under high-density hardware loads.
A hybrid cooling solution combines air cooling and liquid cooling.
Key features include:
Liquid cooling for CPU and GPU core components
Industrial high-static-pressure fan array
Independent airflow zones for CPU, GPU, memory, and storage
Front-to-back airflow design
Intelligent thermal control system
Cooling efficiency is improved by more than 40%, keeping core hardware temperature below 45°C and preventing computing performance throttling.
The chassis reserves 8–12 full-height, full-length PCIe 5.0 / 6.0 expansion slots.
It supports:
InfiniBand HDR / NDR cards
100G / 400G Ethernet cards
Low-loss high-speed cables
Multi-node cluster networking
Internal cable routing is optimized to shorten interconnect paths and control node-to-node latency within 8μs.
Core components support N+1 redundancy:
Hot-swappable power supplies
Redundant cooling fans
Modular CPU, GPU, storage, and power modules
Integrated chassis-level BMC management supports:
Remote monitoring
Fault alarms
Log query
Centralized multi-node management
Fault response time can be controlled within 5 minutes to ensure continuous cluster operation.
AI training clusters rely heavily on GPU computing and require dense multi-GPU deployment with high-speed interconnect. Long training tasks generate extreme heat, and uneven cooling may interrupt training processes.
Key requirements include:
Multi-GPU high-density deployment
High-speed GPU interconnect
Massive NVMe storage
Flexible expansion
Compatibility with AI training frameworks
Multi-node batch management
The solution uses 2U / 4U GPU-optimized chassis.
A 4U chassis can support:
8–12 dual-width GPU accelerator cards
NVIDIA A100 / H100 class GPUs
GPU spacing optimized to more than 30mm
NVLink / NVSwitch high-speed interconnect
GPU-to-GPU bandwidth above 1.6TB/s
1–2 high-performance CPUs
This structure supports distributed training for large-scale AI models.
Each GPU is equipped with an independent cooling channel and dedicated airflow path.
The cooling system includes:
Directional GPU airflow
Liquid cooling support
Enlarged air intake and exhaust structure
Dust filters
Intelligent thermal control
GPU temperature can be controlled below 50°C, helping prevent training interruption caused by overheating.
The chassis supports:
16–32 NVMe high-speed drive bays
U.2 interfaces
Storage bandwidth above 100GB/s
TensorFlow and PyTorch compatibility
CPU-GPU-storage collaborative optimization
It also supports domestic GPU and CPU platforms, meeting localization requirements for AI training infrastructure.
The modular design supports hot-swappable GPUs, drives, and power modules.
The system supports:
Flexible GPU expansion
Storage capacity expansion
Remote monitoring
Batch firmware upgrades
Multi-node fault diagnosis
GPU status, temperature, and workload monitoring
This reduces maintenance costs and simplifies cluster management.
Scientific simulation workloads vary greatly across physics simulation, bioinformatics, materials science, engineering simulation, and other research fields.
Typical challenges include:
Compatibility with different computing cards and simulation modules
Flexible single-node or small-cluster deployment
Limited research budgets
Limited maintenance staff
Frequent hardware upgrades
Noise control requirements in laboratory environments
Available chassis options include:
1U
2U
4U
A single node can support:
1–2 CPUs
2–8 GPUs or computing cards
FPGA acceleration cards
Dedicated simulation cards
Multiple expansion modules
The internal layout reserves sufficient expansion space for different scientific workloads while maintaining cost efficiency.
The system adopts efficient air cooling with industrial-grade low-noise fans.
Key features include:
Noise level ≤50dB
Independent cooling for CPU, GPU, and expansion cards
Smart fan speed adjustment
Stable thermal performance under different workloads
This makes the system suitable for laboratory environments.
The chassis supports:
Intel Xeon
AMD EPYC
Domestic CPUs
NVIDIA GPUs
AMD GPUs
Domestic GPUs
PCIe 4.0 / 5.0 expansion
High-speed storage expansion
Small-scale cluster networking
Customers can upgrade memory, storage, GPUs, and expansion cards without replacing the entire chassis.
The modular hot-swappable design allows quick replacement of:
Drives
Fans
Power supplies
Fault response time can be controlled within 10 minutes.
Integrated remote management supports:
Remote monitoring
Fault alarms
Log export
Basic remote troubleshooting
This helps research institutions reduce on-site maintenance workload.
Edge HPC systems are often deployed in industrial sites, autonomous driving test environments, and remote computing locations.
Key challenges include:
Limited space
Strict weight requirements
Low-latency local computing
Dust, humidity, and temperature fluctuation
Limited power supply
Need for local high-speed storage
Remote maintenance and self-healing capability
The solution uses:
Short-depth 1U chassis, 450–600mm
Compact 2U chassis
Aerospace-grade aluminum alloy
Compared with traditional chassis:
Volume is reduced by 35%
Weight is reduced by 40%
The system supports:
1–2 CPUs
2–4 GPUs or edge computing modules
Wall-mounted or rack-mounted installation
The system adopts low-power high-performance hardware configuration.
Key specifications include:
Standby power ≤40W
Operating power ≤120W
Wide-temperature operation from -10°C to 60°C
IP54 dust and water resistance
Sealed structural design
This ensures stable operation in industrial edge environments.
The chassis supports:
8–16 NVMe high-speed drive bays
Data read/write latency ≤1ms
High-speed interconnect interfaces
5G / 4G module compatibility
This reduces dependence on core data centers and enables low-latency edge computing.
Integrated smart remote management supports:
IPMI / Redfish protocols
Remote power on/off
Remote diagnostics
Firmware upgrades
Fault alarms
Self-healing redundancy
Fans and power supplies support redundant backup and automatic failover to ensure uninterrupted edge computing operation.
Main materials include:
SECC galvanized steel
Reinforced 1.2–1.5mm steel for supercomputing and AI training nodes
Aerospace-grade aluminum alloy for edge HPC
Wear-resistant and anti-corrosion powder coating
These materials provide:
High strength
EMC protection
Rust resistance
Electromagnetic shielding
Long-term durability
The solution uses:
Precision sheet metal fabrication
CNC machining
±0.5mm tolerance accuracy
Fully welded reinforced structures
Modular architecture
Standardized internal cable management
This improves installation accuracy, cooling efficiency, interconnect reliability, and maintenance convenience.
The thermal architecture uses:
Front-to-back airflow
Independent airflow zones
Dedicated cooling paths for CPU, GPU, memory, drives, and expansion cards
Cooling efficiency is improved by more than 40%.
For supercomputing and AI training clusters, the design works with precision data center cooling systems to deliver cold air directly to core hardware.
For edge scenarios, airflow is optimized together with sealing design to prevent dust and moisture intrusion.
Supported cooling methods include:
Air cooling
Hybrid cooling
Liquid cooling
For supercomputing and AI training nodes, liquid cooling can reduce CPU and GPU temperatures by more than 20–25°C.
For scientific simulation and edge computing, high-efficiency air cooling provides a balance of thermal performance, energy efficiency, and low noise.
The system uses industrial-grade high-reliability fans with:
MTBF ≥150,000 hours
N+1 redundancy
Hot-swappable design
High-static-pressure options for HPC clusters
Low-noise options for laboratories and edge environments
Noise can be controlled below 50dB in applicable scenarios.
The chassis supports:
Intel Xeon
AMD EPYC
Domestic CPUs
ATX / EEB / ITX / custom motherboards
1U / 2U / high-power redundant power supplies
PCIe 4.0 / 5.0 / 6.0
NVIDIA GPUs
AMD GPUs
Domestic GPUs
FPGA acceleration cards
Dedicated computing cards
SAS / SATA / NVMe drives
InfiniBand and Ethernet interconnect cards
It also supports domestic hardware platforms for HPC localization requirements.
The chassis can reserve:
Up to 12 PCIe expansion slots
Up to 32 NVMe drive bays
Hot-swappable drives and expansion cards
5G / 4G module interfaces for edge scenarios
Backup power interface expansion
Multi-node cluster expansion
It is compatible with mainstream cluster management systems for large-scale deployment.
The solution supports:
Lightning protection
Anti-static protection
Over-current protection
Over-voltage protection
Surge protection
Physical lock and anti-tamper alarm
EMC electromagnetic interference protection
IP54 or higher protection for edge scenarios
Illegal chassis opening can automatically trigger alerts and push notifications to the maintenance platform.
The chassis supports:
CE certification
FCC certification
CCC certification
ISO9001 quality management system
HPC IT equipment safety standards
Each chassis undergoes:
High-temperature testing
Low-temperature testing
Vibration testing
EMC testing
Thermal efficiency testing
Long-term high-load stability testing for core scenarios
MTBF exceeds 150,000 hours.
The delivery process is optimized for HPC projects with clear computing requirements, strict delivery schedules, high maintenance standards, and complex compatibility needs.
A dedicated HPC industry team communicates with the customer to confirm:
Application scenario
Computing requirements
Hardware configuration
Cooling requirements
Interconnect standards
Expansion planning
A requirement confirmation document is provided to ensure the solution accurately matches the HPC workload.
Based on the requirements, the engineering team performs:
3D modeling
Thermal simulation
Interconnect compatibility verification
Airflow optimization
Internal layout optimization
Deliverables include:
Detailed design proposal
BOM list
Cost quotation
Thermal design description
High-speed interconnect compatibility notes
Rapid prototyping includes:
Hardware compatibility testing
Thermal performance testing
High-speed interconnect testing
Reliability testing
GPU interconnect testing for AI training scenarios
Protection testing for edge scenarios
Simple structural modifications can be completed within 3–5 days, while complex high-density or hybrid cooling designs may require 10–15 days.
With an in-house sheet metal fabrication workshop and automated production lines, the company supports scalable production.
Quality inspection includes:
48-hour high-temperature high-load testing
Thermal testing
Vibration testing
EMC testing
OEM/ODM branding is supported.
Monthly capacity can reach tens of thousands of units, supporting orders from dozens to thousands of units.
Support includes:
On-site installation guidance
Hardware debugging
Cluster networking assistance
7×24 technical support
HPC cluster management system integration
AI training framework integration
Scientific simulation software debugging
1–3 year warranty
Lifetime technical support
Spare parts inventory
Fault response within 24 hours
On-site maintenance for critical scenarios
HPC operation training
A customized 4U high-density liquid-cooled chassis was developed for a provincial supercomputing center.
Configuration:
2 AMD EPYC CPUs
16 NVIDIA H100 GPUs
48 NVMe high-speed drives
Hybrid cooling system
InfiniBand NDR high-speed interconnect
Results:
Core hardware temperature controlled below 42°C
Node-to-node latency ≤7μs
100-node cluster deployment
Total computing power reached 100 PFlops
Supports weather forecasting and astrophysics simulation
Annual downtime ≤2 hours
Maintenance efficiency improved by 80%
A customized 4U GPU chassis was developed for a technology company.
Configuration:
8 NVIDIA A100 GPUs
NVLink high-speed interconnect
1.6TB/s GPU-to-GPU bandwidth
32 NVMe high-speed drives
Directional GPU cooling design
Results:
GPU temperature controlled below 48°C
50-node training cluster deployment
Supported 100-billion-parameter model training
Training efficiency improved by 50%
Training interruption rate reduced below 0.5%
A customized 2U research chassis was developed for a university.
Configuration:
1 Intel Xeon CPU
4 NVIDIA A6000 GPUs
Low-noise air cooling
Remote management module
Results:
Noise level ≤48dB
Suitable for laboratory environments
Supports materials science and bioinformatics simulation
20-node remote management
Reduced maintenance workload
A customized short-depth 1U edge chassis was developed for an automotive company.
Configuration:
1 low-power high-performance CPU
2 NVIDIA Orin GPUs
IP54 protection
-10°C to 60°C wide-temperature design
5G module integration
Results:
Standby power ≤38W
Stable operation in industrial test environments
Supports autonomous driving inference and simulation
Latency ≤1ms
Remote self-healing fault management
7×24 HPC industry technical consultation
Preliminary solution within 24 hours
Dedicated HPC engineering team
Focus on high-density integration, high-speed interconnect, and thermal design
ISO9001 quality management system
Full inspection before shipment
High-temperature and low-temperature testing
Vibration testing
EMC testing
Thermal efficiency testing
High-speed interconnect testing
Long-term high-load stability testing
MTBF ≥150,000 hours
Complete quality inspection reports provided
Supports:
Prototype from one unit
Large-volume fast delivery
Structural customization
Thermal customization
Interconnect customization
Interface customization
Appearance customization
Domestic hardware platform adaptation
1–3 year warranty
Lifetime technical support
Spare parts inventory
Fault response within 24 hours
On-site maintenance for supercomputing and AI training clusters
Cluster networking assistance
Software integration support
HPC operation training
The company invests 8% of annual revenue into R&D and collaborates with supercomputing centers, research institutions, and GPU manufacturers to continuously improve:
High-density integration
Thermal management
High-speed interconnect technology
PCIe 6.0 compatibility
Liquid cooling upgrades
Domestic HPC platform adaptation
This ensures that the solution continuously meets evolving HPC industry requirements and helps enterprises and research institutions improve computing efficiency.
