GPU guide / H100 PCIe

NVIDIA H100 PCIe 80GB

A hardware guide for checking model memory and cloud configurations. Hardware facts are separate from provider prices and availability.

GPU memory80 GBHBM2e · per GPU

ArchitectureHopperPCIe

Memory bandwidth2,000 GB/sTheoretical hardware specification

Memory is a budget for your model.

Model weights, context and runtime all need space. A “80 GB GPU” does not mean a 80 GB model will fit.

Model weightsThe model itself

Context + batchWhat it processes at once

Runtime + headroomSoftware and working space

Conceptual allocation only. No model-size guarantee or measured performance claim.

Hardware notes

NVIDIA describes bandwidth as over 2 TB/s; 2,000 GB/s is a rounded reference, not a benchmark.

Hardware source reviewed 11 October 2026 · verified. Read source ↗

Cost of your workload

We currently earn no commission from these links. How we fund this site

Dated public-source record

$2.50 / hour

USD · 1 GPU · full configuration

Compute only · extras excluded

Try it for 1 hour$2.50

An 8-hour day$20.00

Compute estimate: a billing rule is not published. Storage, transfer and tax extra.

Storage, egress and tax are not included in this compute rate. Live availability is unconfirmed.

Use the same workload and usage assumptions. Compare complete configuration prices and check the supporting resources.

Compare hardware specifications without assuming equal GPU-hours deliver equal performance.