Cut Up to 90% Cloud Costs with Alibaba Cloud PAI‑EAS Spot Instances
Alibaba Cloud's PAI‑EAS now supports Spot Instances, letting users run latency‑tolerant inference tasks on pre‑emptible compute resources and achieve up to 90% cost savings compared with on‑demand instances, while offering flexible bidding, protection periods, multi‑spec selection, and automatic scaling.
In the drive for efficiency in AI development and services, Alibaba Cloud’s Machine Learning platform PAI announces support for Spot Instances . During model inference, users can select Spot Instances via the PAI‑EAS online service platform for tasks that are less sensitive to latency, achieving up to 90% cost optimization compared with pay‑as‑you‑go instances.
What is a Spot Instance
A Spot Instance is a purchasing model for compute resources whose price fluctuates in real time based on market demand and supply. PAI‑EAS Spot Instances draw from idle resources in the public resource pool, offering prices as low as 10% of the pay‑as‑you‑go rate.
Unprotected Spot < Spot with protection < Prepaid < Pay‑as‑you‑go
Before using Spot Instances, users set a maximum bid price and optionally a 1‑hour protection period. Once the service is deployed, PAI‑EAS automatically bids for the appropriate resources.
Spot Instance Usage Steps
Purchase Spot Instance: success requires sufficient inventory and a bid not lower than the current market price.
Use Spot Instance: with a 1‑hour protection period, the instance is guaranteed for at least one hour even if the market price exceeds the bid; after the hour, the instance may be released if inventory is insufficient or the bid is too low. Without protection, the instance is released immediately under those conditions.
Multi‑spec Instance Selection: when deploying with Spot Instances, PAI‑EAS can iterate through a list of specifications to launch resources, greatly reducing deployment risk caused by Spot Instance release.
Applicable Scenarios
Spot Instances are suitable for price‑sensitive but latency‑tolerant workloads, such as:
AIGC asynchronous generation
Image recognition, OCR batch processing
Video segmentation, classification batch processing
Speech segmentation and transcription batch processing
Stable Diffusion and other AI painting batch jobs
When real‑time inference results are not required and a delay of up to an hour is acceptable, Spot Instances can significantly lower service costs.
Configuration Guide
Open the PAI‑EAS console and click “Deploy Service”.
In the “Resource Deployment” section, select “Public Resource Group” and switch to “Advanced Resource Configuration” to set Spot Instance resources.
Choose a protection period:
1‑hour protection: guarantees at least one hour of usage after successful acquisition.
No fixed protection: no guarantee, but cheaper pricing.
Select the machine type; the interface shows Spot and original prices for comparison. Add instance specifications with the “+” button; PAI‑EAS will traverse the spec list to launch resources, mitigating release risk.
Complete other basic settings and click “Deploy”.
References
EAS Spot Instance Overview
Advanced Configuration: Multi‑Spec Instance Selection
Alibaba Cloud Big Data AI Platform
The Alibaba Cloud Big Data AI Platform builds on Alibaba’s leading cloud infrastructure, big‑data and AI engineering capabilities, scenario algorithms, and extensive industry experience to offer enterprises and developers a one‑stop, cloud‑native big‑data and AI capability suite. It boosts AI development efficiency, enables large‑scale AI deployment across industries, and drives business value.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
