The rise of AI necessitates robust power backup systems. As companies increasingly adopt AI models, the risk of power outages becomes a critical concern. AI data centers require near-constant uptime, often aiming for 99.99999% reliability. Even a brief power interruption can disrupt training processes or data integrity, leading to significant financial losses and operational setbacks.
Core Components of AI Data Center Design
Core Computing Section
-
NVIDIA A100 and H100 GPUs: These GPUs are pivotal for AI model training and inference. The A100, based on Ampere architecture, delivers 19.5 TFLOPS of FP64 performance and 9.7 TFLOPS of TF32, making it highly efficient for complex AI tasks. The H100, built on Hopper architecture, offers 2.0 PFLOPS of FP64 performance and 4.0 PFLOPS of FP8, significantly accelerating AI workflows.
-
Huawei Ascend 910 AI Processors: With a 7nm process, it provides 256 TOPS of INT8 performance and 64 TOPS of FP16, ideal for large-scale AI model training.
Data Storage Section
High-capacity storage systems are vital for AI data centers. Solutions like Huawei OceanStor Dorado V6, with its FlashLink® architecture and 20 μs latency, ensure rapid data access. Also, OceanStor 2910 V5, supporting 1,200 SATA disks and offering 5 TB/s backup rates, meets large-scale storage needs.
Network Switch Section
High-speed network switches are crucial. The Huawei CloudEngine 16800, with 768 100GE ports and a 140 Tbps switching capacity, supports large-scale data center networks. It also complies with IEEE 802.1Q standards for VLAN segmentation, ensuring efficient data transmission and management within the data center.
Power Backup Solution with Cummins KTA50 Generator Drive Engine
The Cummins KTA50 is a powerful and reliable backup power solution for AI data centers. As a V-type 16-cylinder, 50-liter four-stroke engine with a bore and stroke of 159 mm (6.25 in), it delivers exceptional power output and efficiency. The engine is equipped with a PT Fuel System and comes with a turbocharger for enhanced performance.
Here are the key specifications of the Cummins KTA50 generator drive engine:
| Specification | Details |
|---|---|
| Model | KTA50/KTTA50/QSK50 |
| Engine Type | 4 – Cycle; 60° Vee; 16 – Cylinder Diesel |
| Displacement | 50 L |
| Rated Speed | 1500 – 1800 RPM |
| Emission Standard | Euro II |
| Fuel System | PT |
| Standby Power | 1210 – 1760 kW @ 1500 – 1800 RPM |
| Prime Power | 1097 – 1650 kW @ 1500 – 1800 RPM |
| Bore * Stroke | 159 mm * 159 mm |
| Wet Weight | 5934 kg |
| Packing Size (L * W * H) | 2794 mm * 1507 mm * 1846 mm |
The KTA50 series offers multiple configurations to meet diverse power requirements. It features turbocharged (T), turbocharged with water-to-air intercooler (TA), and turbocharged with air-to-air intercooler (TAA) options. JOZ provides full lifecycle services for customers, including design, installation, commissioning, after-sales training, spare parts supply, and technical support.
When designing AI data centers, selecting the right hardware components is crucial for meeting the high computational demands of AI models. NVIDIA and Huawei's advanced AI cards provide the necessary computing power, while Cummins generator drive engine ensures reliable backup power. This comprehensive solution addresses both performance and reliability requirements, enabling AI data centers to operate efficiently and sustainably in an increasingly AI-driven world.