ECS

EBM (bare-metal)

GPU Instance

Preemptible instance

Virtual Node

HPA

Heterogeneous computing is supported.

All major machine learning frameworks are supported.

Optimized K8S scheduling is provided for jobs running on the GPU. For example, it can ensure

pods are scheduled on the GPU node, provide shared GPU resources, isolate GPU resources, and do

co-scheduling.

GPU and shared GPU utilization can be monitored via Prometheus on Alibaba Cloud.

The "Arena" tool is developed by the community but integrated with Alibaba Cloud ACK.

True

False

True

False

Al ML job lifecycle management

Optimized scheduler for AI/ML jobs

Unified heterogeneous resources management

Accelerate computation with distributed data cache

Auto Scaling

GPU monitoring and problem detection

Cost management

In one zone

In two zones

In three zones

In four zones

FIFO

Capacity

Fair share

Short job first