@meta
  v: 1
  route: /gpu/gaudi-3
  generated: 2026-06-10T09:13:07.926Z

@intent
  purpose:    Show what AI models fit on the Gaudi 3.
  audience:   ai-engineer, self-hoster, homelab-buyer
  capability: inspect_specs, find_fitting_models, compare_to_other_gpus

@state
  slug: gaudi-3
  name: Gaudi 3
  vram_gb: 128
  bandwidth_gbs: 3700
  fp16_tflops: 1835
  budget_8k_gb: 102
  category: intel
  summary: 128GB HBM2e. Habana's transformer accelerator. 70B at FP16 in production.
  top_fitting_models[5]: mixtral-8x22b, gpt-oss-120b, command-r-plus, qwen-2-5-72b, llama-3-3-70b

@actions
  - id: compare
    method: GET
    href: /gpu/dgx-spark
  - id: compare
    method: GET
    href: /gpu/m4-max-128
  - id: compare
    method: GET
    href: /gpu/m5-max-128
  - id: compare
    method: GET
    href: /gpu/h200
  - id: view_calculator
    method: GET
    href: /#calculator
  - id: view_math
    method: GET
    href: /the-math

@context
  > 128GB HBM2e. Habana's transformer accelerator. 70B at FP16 in production.

@nav
  self:      /gpu/gaudi-3
  parents:   [/]
  peers:     [/gpu/dgx-spark, /gpu/m4-max-128, /gpu/m5-max-128, /gpu/h200]
  drilldown: /the-math