Skip to content

Instances overview

An Instance is a single, dedicated compute machine you provision in Verda — either CPU-only or GPU-accelerated. You get full control of the operating system and software stack, so Instances are a good fit when you want a familiar "one server" model (SSH, install packages, run services, manage files, etc.).

Instances are typically used for:

  • interactive experimentation (e.g., notebooks, dev environments)
  • training or batch jobs that you want to run on a single machine
  • inference and services you want to deploy directly on a VM

CPU vs. GPU: which should you use?

Choose a CPU Instance when your workload is CPU-bound, such as:

  • data preprocessing / ETL
  • orchestration and pipelines
  • web services, APIs, and background workers
  • lightweight development and testing

Choose a GPU Instance when you need GPU acceleration, such as:

  • deep learning training and fine-tuning
  • high-throughput inference
  • computer vision / generative AI workloads
  • anything that benefits from CUDA-capable compute

As a rule of thumb: start with CPU if you're unsure, and move to GPU once profiling shows the workload is GPU-bound or you already know the model requires it.

Instances vs. Clusters

An Instance is a single machine, while a Cluster is multiple nodes working together for distributed workloads.

Pick an Instance when you want the simplest "one server" setup: a single notebook/dev box, a single training job that fits on one machine, or a service/inference deployment that can run on one VM and scale vertically (bigger CPU/GPU, more RAM).

Pick a Cluster when you need to scale horizontally across machines (more total GPUs/CPU than one node can provide), run distributed training frameworks, or you want higher availability and easier scaling/rolling updates for production services.

Quick comparison

Instances and Clusters are both ways to get compute from us. They differ in how much of the underlying machine you take on, and in what shape the workload needs to be.

Aspect Instances Clusters
You provide Nothing; you build on the machine Your jobs, submitted to the orchestrator
You manage The OS, drivers, server process and uptime Your workloads on the cluster
You reach it through SSH, with root on the machine Kubernetes or Slinky, pre-installed
Scaling Manual: you start and stop it Fixed cluster size, from 16 to 128 GPUs
You pay For as long as the instance runs For the cluster, Pay As You Go
Suits Development, fine-tuning, interactive or stateful work Distributed training and multi-node inference

Quickstart

Getting a working instance takes three steps: deploy it, add an SSH key so you can reach it, then connect.

  1. Create an instance: choose a GPU or CPU model, storage, and an operating system, then deploy.
  2. Manage SSH keys: generate a key pair (or reuse one) and add it to your project so it's available when you deploy.
  3. Connect to your instance: SSH in using the IP address shown on your instance card.

Do steps 1 and 2 in either order. You can add an SSH key while deploying, or beforehand from Credentials. Once your instance is running and your key is attached, connecting takes one command.