Data Science Workshop (DSW) of Alibaba Cloud Machine Learning Platform for AI recently has scaled the resources available on the platform. Besides different types of GPU resources, CPU resources are now available for you to save costs and increase performance. Currently, DSW supports GPU P100, GPU M40, and multiple CPU models. DSW allows you to change the resource type or model for instances that use the pay-as-you-go billing method.
Pricing: A pay-as-you-go instance is billed based on the actual runtime. Fees charged per hour vary depending on the resource model. For more information, see Pricing(https://www.alibabacloud.com/product/machine-learning/pricing)
|Resource type||Resource model||Region|
|GPU: pai.medium.1xp100. An instance deployed in China (Beijing) supports up to eight GPUs.||P100||China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and Singapore|
|GPU: pai.medium.1xm40||M40||China (Shanghai)|
|CPU: dsw.large.2core4g||2-core 4 GB||China (Beijing) and China (Shanghai)|
|CPU: dsw.xlarge.4core8g||4-core 8 GB||China (Beijing) and China (Shanghai)|
|CPU: dsw.2xlarge.8core16g||8-core 16 GB||China (Beijing) and China (Shanghai)|
|CPU: dsw.4xlarge.16core32g||16-core 32 GB||China (Beijing) and China (Shanghai)|
|CPU: dsw.6xlarge.24core48g||24-core 48 GB||China (Beijing) and China (Shanghai)|
When you use an instance to train models, if you find that the current computing resource cannot meet your requirements, you can change the resource. A pay-as-you-go instance allows you to switch between CPUs and GPUs. It also supports switching between different CPU models or GPU models. The following figure shows the procedure of switching the computing resource.Select a resource type and model, and then click Switch to switch to the specified computing resource. The title of the right-side panel displays Switching and a loading icon is shown. This means that the instance is switching to the specified resource type. It may take one to two minutes for the instance to complete the switchover. After the switchover is complete, the title of the right-side panel displays the specified computing resource type and the title of the code tab displays the kernel that the instance is connected to. After the instance is switched to a new computing resource, you must re-execute the existing code for it to take effect.
- After the instance is switched to a new computing resource, it is billed based on the new resource. Please pay close attention to your bill statements.
- After the instance is switched to a new computing resource, if the page is not updated or shows that the switchover failed, manually refresh the page.
- Do not change the computing resource frequently unless it is necessary. Do not click any button during the resource switchover process.