PAI AI computing resources provide stable, high-performance compute for latency-sensitive workloads. Two broad categories are available: cloud-native resources for model development, training, and inference, and big data engine resources for data processing pipelines.
Resource types
-
Cloud-native resources
Used for PAI-DSW, PAI-EAS, and PAI-DLC:
-
General computing resources
General computing resources — ECS, ECI, and EGS — deliver a flexible, stable, high-performance environment for deep learning training. Activating PAI automatically creates a public resource quota for general computing resources, which you can then associate with a workspace.
-
Lingjun resources
Lingjun resources are PAI's purpose-built compute for large-scale deep learning and intelligent computing. Built on hardware-software co-optimization, they provide a high-performance heterogeneous computing foundation optimized for development, training, and inference.
-
-
Big data engine resources
Used for PAI-Designer:
-
MaxCompute
MaxCompute is an enterprise-level Software as a Service (SaaS) cloud data warehouse for data analytics. It provides a fast, fully managed online data warehouse service that eliminates the scalability and elasticity limitations of traditional data platforms — reducing O&M costs while enabling cost-effective large-scale data processing. For more information, see What is MaxCompute?.
-
Fully managed Flink resources
Realtime Compute for Apache Flink is a one-stop, real-time big data analytics platform built on Apache Flink with end-to-end sub-second latency. For more information, see What is Realtime Compute for Apache Flink?.
-
Usage
On the AI Computing Resources > Resource Pool page, create resource groups and purchase compute:
General computing resources: For details, see Create a resource group and purchase general computing resources.
Lingjun resources: For details, see Create a resource group and purchase Lingjun resources.
Then, on the AI Computing Resources > Resource Quota page, create a resource quota and associate it with a workspace for AI development and training.
-
Cloud-native resource quota
Allocate purchased resources from one or more resource groups to one or more resource quotas. Create child quotas to form a quota tree for enhanced queuing and scheduling.
Lingjun resources: For details, see Create a resource quota.
General computing resources: For details, see General computing resource quotas.
-
Big data resource quota
For instructions on activating, purchasing, and using big data resource quotas, see:
MaxCompute: For details, see MaxCompute resource quotas.
Fully managed Flink resources: For details, see Manage fully managed Flink resources.