NVIDIA Hopper

NVIDIA H200

H200 refreshes Hopper with faster HBM3e, boosting throughput for memory-bound foundation models.

Launch year
2024
Memory
141 GB HBM3e
Memory bandwidth
4.8 TB/s
Peak FP16 / FP32
450 TFLOPS · 80 TFLOPS

Market snapshot

$1.08 /hr

Range $1.08 – $28.72

Catalog coverage

172 live offerings

Across 3 providers · 17 regions

  • HBM3e memory
  • Drop-in H100 replacement
  • FP8 Transformer Engine
NVIDIA H200 render

Last refreshed Oct 20, 2025, 1:51 AM

Performance snapshot

Normalized versus NVIDIA A100 (=1.0). Values use public reference benchmarks for training and inference workloads.

  • AI Training×2.00
  • AI Inference×1.90
  • Memory Bandwidth×1.80

Provider availability

Price bands per provider pulled from the live catalog.

  • Datacrunch logoDatacrunch32 offers
    $1.08 – $20.72 /hr
  • Nebius logoNebius12 offers
    $1.45 – $28.00 /hr
  • RunPod logoRunPod128 offers
    $2.29 – $28.72 /hr

Popular regions

  • AP-JP-116 offers
  • CA-MTL-316 offers
  • CA-MTL-416 offers
  • US-GA-216 offers
  • US-NC-116 offers
  • EU-FR-114 offers
  • EUR-IS-212 offers
  • FIN-018 offers

Weighted average price $11.42 /hr · median $10.77