> For the complete documentation index, see [llms.txt](https://lwang010.gitbook.io/longw/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://lwang010.gitbook.io/longw/mlops/chap-3.-hardware-maintain/t-or-g-or-c-pu/gpu-source-in-cloud.md).

# \[info] GPU source in cloud

When you'd like to train some models, instead of buying a server with GPUs, you could also use the cloud service. Here are some GPU sources that are commonly used and not expensive. 

* GCP ($300 credit)
* Azure ($200 credit)
* AWS (gain credit by participating in their summits)

\
**AWS: updated for 2021-06-20**

* Inf1 instance: for deployment (note that it is not an NV GPU card).

**AWS: updated for 2020-11-08**

* G4: Best instance for cost-efficient deep learning training, and high-performance inference deployments
* p3dn.24xlarge: Fastest GPU instance in the cloud
* P2: Cost-effective for training and prototyping, but consider G4 instances first