Skip to content

Recommend Inferentia2/Trainium for inference workloads #13

@maksimov

Description

@maksimov

When an NVIDIA G-class instance is detected running inference with low
utilization, recommend AWS Inferentia2 (inf2) or Trainium (trn1) as
alternatives. These offer up to 3x price-performance for supported models.

  • Detect inference patterns (steady invocation rate, low batch variability)
  • Map compatible model architectures to inf2/trn1 support
  • Show concrete price-performance comparison vs current NVIDIA instance
  • Caveat: not all models are compatible — flag this clearly in recommendations

Major trend in 2026: NVIDIA → AWS silicon migration for inference workloads.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestv0.2Version 0.2 milestone

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions