Launch GPU/CPU-powered instances
Launching an instance is easy and flexible.π Just choose the framework, GPU type, GPU quantity, and storage; then click the launch button. Your instance will be up and running in less than 30 seconds!β°
Detailed Walkthroughβ
You can launch an instance in 4 easy steps:
Choose the framework:
- Pytorch
- FastAI
- TensorFlow
- BYOC (Choose any public container of choice - Advanced Feature)
Choose the GPU type in ascending order of performance:
- RTX 5000
- RTX 6000
- A5000
- A6000
- A100
- CPU
Choose the number of GPUs:
- Choose a value between 1 to 8
Choose a disk size:
- Choose any value between 20GB to 500 GB.
Reserved:
- Yes
- No
Duration:
- Hourly
- Weekly
- Monthly
Reserved vs Not Reservedβ
We offer 2 kinds of instances:
- Reserved
- Not Reserved
Reservedβ
When you choose reserved instances, the resources are guaranteed. The resources that are allocated will not be released until you hit pause/destroy or run out of balance.
Not Reserved or Spot Instancesβ
Instances that are not reserved are called spot or preemptible instances. You can use spare GPU capacity at discounted prices and save up to 60% on cost. But these instances will be auto-paused when demand increases. These instances are best suited for:
- Learning
- When you have implemented checkpoints and want to save on costs.
Weekly/Monthly Durationβ
Get up to 10% discounts for weekly usage and up to 30% discounts on monthly usage. Instances that are launched for a week/month cannot be paused or destroyed during that duration. Post the duration, instances will continue at discounted prices till you choose to pause/destroy.
Optional Featuresβ
Click the down arrow β¬οΈ for additional options
Startup Script πβ
You can choose a startup script that gets executed when the instance starts. You can use it for things like:
- Installing your favorite libraries
- Cloning GitHub repos
Name the Instanceβ
It is common to run multiple experiments while training models. Name the instance to remember what project you are using it for.
Billing starts once an instance is created, so ensure that you either pause or destroy the instance when not in use to avoid unnecessary costs.
β οΈ The instances will be paused automatically if the account balance becomes negative. You can enable subscription or maintain sufficient credits in the wallet.