Operations, Security, and Delivery
Development Tools
Q: Can GPUs be used through these tools?
Answer: Yes. Users can create a development IDE with the required CPU/NPU resource flavor.
Distributed Capabilities
Q: Distributed model training / inference
Answer: Yes.
Q: RAY framework support
Answer: Yes.
Monitoring
Q: How do we collect monitoring metrics?
Answer: NPU/GPU card metrics are collected through NPU exporters or the DCGM exporter. CPU, RAM, and ephemeral disk metrics are collected through the Kubernetes Metrics Server.
Q: NVIDIA has DCGM metrics and debugging tools such as Nsight. What is the equivalent on HCS?
Answer: The DCGM exporter is deployed when GPU nodes are managed.
Q: How can we observe card-level metrics?
Answer: Card-level metrics are collected through NPU exporters or the DCGM exporter.
AI Security
Q: Are there AI security solutions specific to the HCS platform?
Answer: None by default. The platform can be integrated with third-party AI security solutions.
Security capabilities that can be integrated include:
- Model scanning
- Guardrail
- LLM Firewall
- Data masking
- Compliance