Skip to main content

Understanding Instance Tiers

RunGen.AI offers three GPU tiers—Basic, Standard, and Performance—to accommodate a wide range of model sizes and workloads. Each tier provides different levels of VRAM, CPU, and performance at varying costs, so you can select the option best suited for your model’s requirements and budget.

Overview of Tiers​

1. Basic​

  • Approx. 6 vCPU, 24GB VRAM
  • Suitable for smaller models and apps that don’t need large-scale memory or multiple workers.
  • Most cost-effective tier for testing or lightweight usage.

2. Standard​

  • Approx. 8 vCPU, 48GB VRAM
  • Enhanced performance and visibility for medium to larger models.
  • A balance of cost vs. capability.

3. Performance​

  • Approx. 14 vCPU, 80GB VRAM
  • Superior performance for very large models.
  • Best for tasks requiring significant GPU memory and compute.

Why Are Some Tiers Unavailable?​

Not every model can run on every tier. Some models (especially very large ones) require more VRAM and compute resources than the Basic or even Standard tiers can provide. If your selected model has a large memory footprint, you may only see Performance (or Performance + Standard) as available options. Conversely, smaller models can often run on any tier.

Here are some common scenarios:

  • Small to Medium Models: Typically compatible with Basic or Standard tiers.
  • Large Models: Often need the Performance tier exclusively due to high VRAM requirements.

You can easily switch tiers if you find your app needs more (or less) capacity at any point.

How to Choose a Tier​

  1. Estimate Model Size: Check your model’s memory requirements.
  2. Select Your Tier: Choose from Basic, Standard, or Performance based on your usage, performance goals, and budget.
  3. Review Cost: Each tier has a different per-minute rate; verify you’ve chosen a tier that aligns with your budget.
  4. Update Anytime: You can update your tier within RunGen.AI if your model’s needs change.

Under the Hood​

While we don’t publicly display the exact GPU hardware backing each tier, know that each tier’s specifications (CPU, VRAM, cost) are carefully managed to ensure reliability and performance for the intended use cases.

Summary​

  • Three Tiers: Basic, Standard, and Performance.
  • Different Caps on VRAM and CPU to handle various model sizes.
  • Some Models may only show Performance if they exceed memory limits of lower tiers.
  • Costs and Requirements scale with tier level.

By selecting the appropriate tier, you can run your AI workloads smoothly while balancing performance and cost. If you outgrow one tier, simply upgrade to the next. And if you find your project is smaller than expected, you can downgrade to save on expenses.

Continue to the next tutorial or explore advanced usage in our documentation to get the most out of your chosen GPU tier!