Compute Comparison
NVIDIAHopper2022

H100 80GB

See rental prices ↓

Current median: $2.80/hr across 81 providers (observed), 0.00 (0.0%) vs yesterday. Typical range: $2.29–$6.16/hr (P10–P90). 30-day: 0.0%

TDP 700W · 989 TFLOPS · 80GB VRAM · 3350 GB/s bandwidth

The H100 80GB has a 700W TDP, 989 TFLOPS of FP16 compute, 80GB of HBM3 VRAM, 3350 GB/s memory bandwidth.

Tensor TFLOPS are shown without sparsity. NVIDIA and AMD often publish figures with sparsity, which are about 2x higher.

Flagship data center GPU. Transformer Engine with FP8 support. SXM5 form factor for max bandwidth.

VRAM
80GB
HBM3
FP16
989.0
TFLOPS
Bandwidth
3.4k
GB/s
TDP
700W
power
Best for:LLM trainingLarge-scale inferenceScientific HPC

H100 80GB currently rents for a median of $2.80/hr across 81 providers (observed on-demand prices). H100 80GB has 80GB of memory. Compare its specs and listed cloud rental prices below. The workload must fit in VRAM with room for runtime memory; a listed price does not confirm stock.

H100 80GB Overview

The NVIDIA H100 80GB is the most capable widely-available GPU for large language model training and high-throughput inference as of 2025. Built on the Hopper architecture (GH100 die, TSMC 4N process), it delivers 989 TFLOPS of FP16 and BF16 throughput (dense) — a 3× improvement over the A100 in raw transformer compute. The headline feature is native FP8 support via the Transformer Engine: at 1,979 TFLOPS FP8 (dense), the H100 can nearly double throughput for inference workloads that tolerate reduced precision, with automatic scaling to maintain accuracy.

Memory configuration is 80GB of HBM3 with 3,350 GB/s bandwidth — 65% higher than the A100 80GB's 2,039 GB/s. This bandwidth advantage is the primary reason H100 outperforms A100 by more than the raw TFLOPS ratio suggests for memory-bound inference: each autoregressive token generation reads the full model weight matrix, so bandwidth directly determines tokens-per-second. The SXM5 form factor connects via NVLink 4.0 at 900 GB/s bidirectional, enabling near-linear tensor-parallel scaling across 8-GPU DGX H100 nodes for 70B+ model inference.

The H100 80GB is the right choice when training models above 30B parameters, running 70B+ inference at production throughput, or working with FP8 quantization pipelines. It is overkill for inference-only workloads on models below 30B parameters — the A100 80GB or L40S typically delivers better cost-per-token at those scales. At $2.50–$5.00/hr on-demand across major providers, the H100 commands a significant premium; validate that your workload is actually compute-bound (not memory-bandwidth-bound) before choosing it over cheaper alternatives.

Memory

VRAM80 GB
Memory TypeHBM3
Bandwidth3350 GB/s
NVLink BW900 GB/s

Compute Performance

FP6433.5 TFLOPS
FP3267 TFLOPS
FP16989 TFLOPS
BF16989 TFLOPS
FP81979 TFLOPS
INT81979 TOPS

Hardware Specifications

Chip

ArchitectureGH100
GenerationHopper
Process NodeTSMC 4N
Transistors80B
Die Size814 mm²
Release DateSeptember 20, 2022

Processors

CUDA / Shader Cores16,896
Tensor Cores528

Clocks

Boost Clock1,980 MHz

Memory

VRAM80 GB
Memory TypeHBM3
Memory Bus5120-bit
Bandwidth3350 GB/s
L2 Cache51 MB
NVLink BW900 GB/s

Power

TDP700 W
InterconnectNVLink 4.0 / PCIe 5.0

Relative Performance

FP16 Compute13%
VRAM Capacity28%
Mem Bandwidth21%

Relative to highest-spec GPU in database

Limitations

Highest cost per hour of widely available GPUs
SXM5 form factor requires specialized server infrastructure
Overkill for inference-only workloads where A100 suffices

Live Cloud PricingOn-demand hourly rates

Live105 providers · -89% price spread
ProviderOn-DemandSpotRegionAvailability
$25.90/hr—us-east-1unknown
$4.50/hr—US/EUunknown
$3.50/hr—AP-Southeastunknown
$2.99/hr—US/EU/Asiaunknown
$2.40/hr—USunknown
$2.49/hr—USunknown
$2.19/hr—US/EUunknown
$2.69/hr—USunknown
$2.09/hr—Asia-Pacificunknown
$2.89/hr—USunknown
$2.15/hr—South Koreaunknown
$2.49/hr—EUunknown
$3.20/hr—APAC/ME/LATAMunknown
$3.99/hr—INunknown
$2.74/hr—US/EUunknown
$4.68/hr—USunknown
$2.27/hr—USunknown
$5.04/hr—us-west-1unknown
$3.81/hr—marketplaceunknown
$21.00/hr—us-east-1unknown
$21.40/hr—EU/USunknown
$2.99/hr—IN-Northunknown
$3.10/hr—EU-Centralunknown
$3.20/hr—EU-Westunknown
$2.45/hr—EUunknown
$2.95/hr—EU-Westunknown
$2.99/hr—USunknown
$2.75/hr—EU-Westunknown
$2.85/hr—US-Westunknown
$2.55/hr—USunknown
$2.49/hr—US/APACunknown
$1.80/hr—Variousunknown
$2.89/hr—EUunknown
$2.49/hr—USunknown
$2.69/hr—EUunknown
$2.55/hr—USunknown
$2.79/hr—EUunknown
$2.99/hr—KRunknown
$2.59/hr—USunknown
$2.75/hr—USunknown
$2.69/hr—EUunknown
$2.95/hr—ISunknown
$2.45/hr—USunknown
$3.15/hr—SEunknown
$2.59/hr—EUunknown
$2.39/hr—Globalunknown
$3.05/hr—EUunknown
$2.65/hr—USunknown
$1.99/hr—Globalunknown
$2.19/hr—USunknown
$2.99/hr—SEunknown
$11.06/hr—us-central1unknown
$2.99/hr—USunknown
$3.20/hr—USunknown
$3.50/hr—USunknown
$2.20/hr—USunknown
$2.70/hr—USunknown
$1.75/hr—US/EUunknown
$2.39/hr—Variousunknown
$2.35/hr—USunknown
$2.85/hr—US/EUunknown
$2.29/hr—Globalunknown
$22.60/hr—us-eastunknown
$2.61/hr—US/EU/APACunknown
$2.25/hr—USunknown
$3.89/hr—US/EUunknown
$3.95/hr—us-eastunknown
$2.99/hr—EU-Westunknown
$2.50/hr—US/EUunknown
$3.99/hr—us-east-1unknown
$23.60/hr—US/EU/BRunknown
$3.44/hr—USunknown
$6.16/hr—us-east-1unknown
$2.65/hr—Variousunknown
$2.89/hr—USunknown
$2.30/hr—US/EU/APACunknown
$5.00/hr—Quebecunknown
$2.75/hr—USunknown
$2.95/hr—USunknown
$2.80/hr—APACunknown
$18.90/hr—US/EU/APACunknown
$2.49/hr—USunknown
$2.99/hr—EUunknown
$3.05/hr—EUunknown
$3.49/hr—EU/US/Asiaunknown
$2.89/hr—NLunknown
$2.48/hr—EU/USunknown
$2.35/hr—US/EUunknown
$4.50/hr—eu-north1unknown
$2.49/hr—EU/US/Asiaunknown
$6.88/hr—us-east-1unknown
$2.80/hr—IN-Westunknown
$10.00/hr—us-ashburn-1unknown
$3.20/hr—EU-Southunknown
$2.65/hr—USunknown
$2.29/hr—INunknown
$2.59/hr—US/Midwestunknown
$2.97/hr—EU-Westunknown
$4.51/hr—EU-Northunknown
$2.49/hr—USunknown
$3.90/hr—us-east1unknown
$2.39/hr—EUunknown
$2.55/hr—US/EUunknown
$2.49/hr—EU (Finland)unknown
$2.95/hr—EU-Northunknown

Salad: Lowest priority tier — preemptible

Cheapest / month
$18907
$25.90/hr × 730h
Median / month
$2183
at median price
Spot savings
N/A
vs on-demand (best)

Job Cost Estimator

At median $2.80/hr across 81 providers (observed on-demand). Spot pricing typically 40–70% lower.

7B LLM fine-tune
8× GPU, 24 h · Full fine-tune, single node
$538
spot ~$242
70B LLM fine-tune
64× GPU, 72 h · Multi-node, 3-day run
$13k
spot ~$5.8k
Inference server
1× GPU, 730 h · 1 month continuous
$2.0k
spot ~$920
Image gen batch
4× GPU, 8 h · Batch generation job
$90
spot ~$40

Estimates only. Actual cost depends on provider, region, availability, and utilization. Use the GPU cost calculator for custom inputs.

Compare H100 80GB vs…

Use Case Guidance

LLM training
Large-scale inference
Scientific HPC

LLM Model Size Guidance

Max model (FP16)~40Bparameters at FP16 precision
Max model (INT8)~80Bparameters at INT8 precision
Max model (INT4)~160Bparameters at INT4/GGUF

Estimates only. Actual capacity depends on context length, KV cache, and framework overhead.

Related Guides

LLM APIs Running on This GPU Class

Providers that serve frontier LLM inference on Hopper-class hardware.

Browse all 42 LLM models
Best LLMs for this GPU
Self-hosting guide & model recommendations

H100 80GB vs Alternatives — Spec Comparison

SpecH100 80GB thisH200 141GBH100 SXM5 80GBH200 SXM 141GB
VRAM80GB HBM3141GB HBM3e▲80GB HBM3141GB HBM3e▲
Memory Bandwidth3350 GB/s4800 GB/s▲3350 GB/s4800 GB/s▲
FP16 TFLOPS989▲989▲989▲989▲
BF16 TFLOPS989▲989▲989▲989▲
FP8 TFLOPS1979▲1979▲1979▲1979▲
INT8 TOPS1979▲1979▲1979▲1979▲
TDP700W▲700W▲700W▲700W▲
Process NodeTSMC 4N▲TSMC 4N▲TSMC 4N▲TSMC 4N▲
ArchitectureGH100▲GH100▲GH100▲GH100▲
Release Year20222024▲20222024▲
Max model (FP16)~40B params~70B params▲~40B params~70B params▲
Max model (INT4)~160B params~282B params▲~160B params~282B params▲
▲ indicates best value in row · FP16/BF16 TFLOPS at full precision · Max model estimates at 2 bytes/param (FP16) and 0.5 bytes/param (INT4)Full side-by-side comparison

Compare H100 80GB

Related GPUs

H100 80GB Frequently Asked Questions

What is the H100 80GB TDP?

The H100 80GB has a TDP (Thermal Design Power) of 700W.

How many TFLOPS does the H100 80GB have?

The H100 80GB delivers 989 TFLOPS of FP16 compute and 989 TFLOPS BF16, and 1979 TFLOPS FP8.

How much VRAM does the H100 80GB have?

The H100 80GB has 80GB of HBM3 VRAM.

What is the H100 80GB's memory bandwidth?

The H100 80GB has a memory bandwidth of 3350 GB/s.

H100 80GB User Reviews

H100 80GB User Reviews

Loading reviews...

Ready to rent?

Compare H100 80GB prices across 131+ providers

Live on-demand & spot rates · monthly cost estimates · availability status

Compare rental prices

LLM Self-Hosting Guide

Best LLM models to run on the H100 80GB

The gold standard for frontier LLM inference and fine-tuning