DEVUP Docs
Back to Dashboard

Compute

Compute Overview

Deploy GPU and CPU instances directly from your DEVUP AI dashboard — full root access via SSH and Jupyter, per-second billing, and persistent storage options.

DEVUP AI Compute lets you launch cloud instances on demand. Choose between GPU instances for machine learning, training, and GPU-accelerated workloads, or CPU instances for general-purpose compute, scripting, and lightweight tasks.

GPU vs CPU — when to use which

GPU InstancesCPU Instances
Best forML training, fine-tuning, inference, rendering, GPU-accelerated computeData processing, scripting, web scraping, CI/CD, general-purpose workloads
HardwareNVIDIA GPUs (RTX, A-series, H-series) with dedicated VRAMMulti-core vCPUs with configurable RAM
StorageContainer Disk + Volume Disk (ephemeral) or Network Volumes (persistent)Container Disk (ephemeral) or Network Volumes (persistent)
BillingPer-second while running, reduced idle rate while pausedPer-second while running, reduced idle rate while paused
AccessSSH + JupyterLabSSH + JupyterLab

Resource sizing guide

When selecting an instance size for Large Language Models (LLMs), a general rule of thumb is ~2GB of VRAM per billion parameters for unquantized models. For example:

  • 7B models: ~14GB+ VRAM
  • 13B models: ~26GB+ VRAM
  • 70B models: ~140GB+ VRAM (or ~35GB+ when using 4-bit quantization)

Use tools like the Hugging Face Model Memory Calculator to estimate exact requirements.

WorkloadRecommended Tier
LLM InferenceMid-range to High-end GPU
Training / Fine-TuningHigh-end GPU (multi-GPU for large models)
Image Generation & 3D RenderingEntry to Mid-range GPU
Computer VisionMid-range GPU
Data Processing & ScriptingCPU Instances

Availability depends on your selected data center. See the GPU Deploy and CPU Deploy pages for current options.

How it works

  1. Choose a tier — select a GPU model or CPU flavor from the live catalog in your dashboard. Availability and pricing are shown in real time.
  2. Configure your instance — pick a container template, set your storage allocation, and optionally attach a persistent Network Volume.
  3. Deploy — your instance provisions in seconds. Once running, connect via SSH or JupyterLab.
  4. Manage — pause to save costs (a reduced idle rate applies), resume when ready, or terminate to stop billing entirely.

Prerequisites

Next steps