GPU/CPU Cluster Platform Construction Engineer

Job Responsibilities

1. Responsible for the overall architecture design, planning and construction of the GPU/CPU high-performance computing platform (HPC/AI training cluster), including the selection, deployment and integration of computing, networking and storage resources.

2. Lead the deployment, configuration, and performance tuning of cluster management and scheduling systems (such as Slurm, Kubernetes, OpenPBS, YARN, etc.) to ensure the platform has high availability, high performance, and good scalability.

3. Responsible for GPU server operating system installation, driver deployment (including CUDA, NVIDIA Fabric, ROCm), and container execution environment (Docker, CRI-O, Containerd) configuration and management.

4. Design and implement a cluster monitoring system that covers node status, GPU utilization, network throughput, work queue and system performance.

5. Participate in the integration and performance optimization of large-scale model training platforms and distributed training frameworks (such as Megatron-LM, DeepSpeed).

6. According to actual business and research needs, support platform construction scale evaluation, computing power planning, capacity management and system expansion deployment.

7. Write and maintain platform construction documents, system architecture documents, standard operating procedures (SOP), and related automation scripts (such as Python, Bash, Ansible, Terraform, etc.)