Accelerate: GPU Infrastructure for your AI Stack
Accelerate are instances based on Metal powered by NVIDIA's H100 GPU. You can start training or inferencing on models with pre-configured deep learning tools like TensorFlow, Pytorch, and Jupyter.
With H100, you can train your models up to 9x faster than with the previous generation A100.
Improvements and fixes
- Latitude.sh Terraform Provider: Shipped version 0.2.1 with support for deploying servers with SSH and user data and fixed an issue that prevented project settings from being updated.
- Updated API URL: The default API URL is now api.latitude.sh. The old URL will continue working indefinitely.
- **Go client library: **Released version 0.1.3.
- **Dashboard: **Improved several aspects of the dashboard to provide better interactions with dropdowns, selects, and more.
- **API: **Added the ipmi_status to GET /servers responses to retrieve the health status of a server's IPMI connection.
- User permissions: Collaborators can now update the deploy config of a server.
- Bandwidth alerts: Improved notifications for bandwidth alerts to avoid sending too many emails.
Deploy servers in 15 seconds with instant deployments
You can now deploy bare metal servers in as little as 15 seconds. Instant deployments are available in all locations for select operating systems.
Views & filters
Create views and filters for your servers and IP addresses right inside the dashboard.
By creating a filtered view a new tab is added to your server or IP resource list. All of your team members get immediate access so you can easily share just the most relevant data with them.
Troubleshoot your servers with Rescue Mode
Misconfiguration and bad installs can cause you to get locked out of your server.
Today we are releasing Rescue Mode as an additional way to troubleshoot your servers and regain access to your system.
When you put a server in rescue mode, we load a lightweight Ubuntu image into your server's memory. This allows you to SSH to the machine to fix issues, recover and transfer files, or simply change access credentials.
Rescue is available for all servers in all locations.