Product Usage

Products & Practice

What products and services does Alaya NeW offer?

Core compute product

• Compute Power Pack — Elastic high-performance compute metered in DCU (Data Computing Unit, "度"). 1 DCU = 312 TFLOPS × 1 hour, supports on-demand consumption and dynamic scaling. Available in two forms: Subscription Compute Power Pack and On-demand Compute Power Pack — see the comparison.

Compute delivery services

• Virtual Kubernetes Service (VKS) — Multi-node, multi-GPU Kubernetes clusters with high-speed interconnect, suited for large-scale training.
• Cloud Container Instance (CCI) — Lightweight, fast-launching containerized environments for inference and development.

LLM & AI development tools

• LM Lab — End-to-end LLM training and fine-tuning, covering data management, training jobs, and orchestration.
• Inference — Inference service for LLMs, computer vision, and NLP, with deployment, compression, and invocation.
• Alaya Studio — AI agent development and application building, with LLM API integration, open-source model fine-tuning, and Agent-as-API publishing.

For the full matrix, see Product family.

What can LLMs do? Are there reference cases?

The platform ships several best-practice recipes covering training, fine-tuning, inference, and agent development:

• Training — Ray distributed training.
• Inference — Coming soon
• Integration — Dify business-data integration.

More recipes: Tutorials overview.

Virtual Kubernetes Service (VKS)

What's the billing rule for VKS?

VKS is a shared resource pool service on Alaya NeW, billed in DCU on demand:

• Only actual compute work is billed. Environment setup and data transfer are not.
• After a cluster starts, storage consumes a small amount when no GPU is in use; the bulk of consumption comes from jobs that actually request GPUs.
• On-demand consumption with real-time deduction — no charge when idle.

For DCU rates per GPU model and storage pricing, see Billing overview and DCU itemization.

How do I release a VKS cluster?

1. Open the Product Overview and click Cluster Management to enter the cluster list.
2. Locate the target cluster and click its name to open Cluster detail.
3. In the action area, click Stop or Release.

If release fails, file a ticket via the Alaya NeW website's online inquiry with your account info (username, phone, email) and the VKS ID.

Which region should I use for H100 / H200 / L40S?

L40S, H200, and other high-performance GPUs are deployed across multiple regions. Availability depends on your account permissions and real-time inventory — there is no fixed answer.

For the best match, submit the following via the Alaya NeW website's online inquiry so the back office can check:

• GPU model and count
• Estimated duration
• Region preference (data compliance, network proximity, etc.)

Cloud Container Instance (CCI)

What are the main use cases for CCI?

CCI (Cloud Container Instance) is a Kubernetes-based serverless container service. Typical scenarios:

See CCI introduction for details.

What's the billing rule for CCI?

CCI is serverless — billed by actual resource consumption, per second, with charges stopping the moment the instance shuts down.

Full rate table: Billing overview.

CCI won't shut down — what should I check?

Possible causes:

1. Foreground process still running — Jupyter, training scripts, or other foreground processes inside the container.
2. Platform delay — Shutdown commands can briefly queue during peak load.
3. Permission or UI issue — The current account may lack the operation permission, or the console page is stale.

What to try:

• SSH into the CCI instance and kill the main process before shutting down. See SSH into CCI.

• Try Force stop / Delete instance from the console.

  Force delete loses non-persisted data (any in-container files not stored on a mounted PVC). Back up to PVC or object storage first if needed.

• Still stuck? File a ticket via the Alaya NeW website's online inquiry with your account info, CCI instance ID, and the specific symptom (button unresponsive, status frozen, etc.).