CoCalc feature
CPU, RAM, and GPU Compute for Research
Compute for demanding analysis and simulation.
Run a project on a dedicated machine or connect a notebook to a remote Jupyter kernel.
Run the whole project on a dedicated machine or send notebook computation to a machine you already have.
Two ways to get more compute
On CoCalc.ai, creating a dedicated machine on your own account needs a paid membership, and its usage is billed to your account.
Move the whole project to a dedicated machine.
Its files, applications, and processes run on the machine you choose, including GPU machines.
Connect a notebook to a machine you already have.
Keep the notebook in CoCalc while its kernel runs on your machine over SSH. Files are not synchronized.
Size the job from a representative run.
Record a smaller run's memory, CPU, GPU, and storage use, then choose a machine that fits. A listed machine may still be unavailable; check the live machine catalog before you plan around it.
Technical details
- Parallel workers can share cores with other projects, so more visible cores do not always mean faster runs.
- GPU-enabled projects on a host can use all of the host's GPUs, and projects on the same host can use the same devices.
- Moving a project goes through backup and restore: saved files move to the new machine, but running processes and files in /tmp do not.
- A remote Jupyter kernel runs over SSH from a CoCalc project; files are not synchronized. Check that your project has the network access the connection needs.
- Save checkpoints and logs to files so a stopped or interrupted run can resume.
- Signed in to your account, the CoCalc CLI shows the hosts visible to you (cocalc host list) and the project-host catalog (cocalc host catalog).