A VSCode server hosted in the Cloud, backed with optional GPUs for accelerated ML training and experimentation.
You are a data scientist that wants to train/fine tune a complex Neural Network. Your hardware is not powerful enough and whilst you've heard great things about GPUs, you cannot afford or justify buying one.
vscode-gpus solves this issue by running an instance of the popular VSCode IDE on
VMs hosted in the cloud. Now you have the power to attach and unattach GPUs at
will for a fraction of the cost compared to buying one.
VSCode introduced a feature that allows you to access VSCode instances running on remote servers. This project uses this feature to create a VM that is hosted on Google Cloud Platform that you can connect to.
graph LR;
your_machine[Your Machine]
internet[Internet]
your_machine --> |tunnel| internet
internet --> |tunnel| VSCode
subgraph VM
VSCode --> GPU..1
VSCode --> GPU..N
end
You simply configure the VM's vCPU and GPU count to match your workload. The VM has batteries-included software that enables you to stay focussed on completing data science without having to worry about underlying software packages, versions and so on.
- Build the VM.
- Run the VM.
- Access the instance in your browser via https://vscode.dev/tunnel/vscode-gpus.
- When you are finished with the instance and your source code is pushed,
run
$ make cleanto destroy VSCode and stop getting billed.