Skip to content
local-ai

NVIDIA H100

NVIDIA · Server

A data-centre accelerator with 80GB of HBM memory and enormous bandwidth, built for training and high-throughput serving. It is the workhorse of professional AI infrastructure, and firmly organisational rather than home hardware.

Specifications

VRAM 80GB
Memory bandwidth 3350 GB/s
Power draw 700W
Type accelerator

A data-centre part, typically bought in quantity or rented by the hour from a cloud provider. The PCIe version has lower bandwidth and power than the SXM one.

Strengths

  • 80GB of very high-bandwidth HBM memory for large models
  • Exceptional throughput for training and multi-user serving
  • The reference platform that most serving software targets first

Weaknesses

  • Data-centre pricing and availability, well beyond home use
  • High power and cooling requirements, needing a proper server
  • Overkill for single-user local inference
See what models fit in 80GB →

Entry last verified 30 July 2026. Specifications and especially prices change; verify before buying.