H100 SXM5 80GB vs A100 SXM4 80GB

Compare NVIDIA H100 SXM5 80GB and NVIDIA A100 SXM4 80GB specs, performance, and cloud pricing

H100 SXM5 80GB

80GB

From $2.20/hr

A100 SXM4 80GB

80GB

From $1.10/hr

Architecture

Hopper

vs Ampere

FP16 Gap

3.2x

H100 SXM5 80GB leads

SpecificationH100 SXM5 80GBA100 SXM4 80GB
VRAM80 GB80 GB
VRAM TypeHBM3HBM2e
FP16 TFLOPS2.0 PFLOPS624 TFLOPS
FP8 TFLOPS4.0 PFLOPSN/A
Memory Bandwidth3.4 TB/s2.0 TB/s
TDP700W400W
InterconnectNVLink 4NVLink 3
ArchitectureHopperAmpere

Price Comparison

MetricH100 SXM5 80GBA100 SXM4 80GB
Cheapest On-Demand$2.20/hr$1.10/hr
Cheapest Spot$1.35/hr$0.760/hr
Providers Available76

Verdict

Best for Training

NVIDIA H100 SXM5 80GB

2.0 PFLOPS FP16 with 80GB VRAM

Best Value

NVIDIA H100 SXM5 80GB

900 TFLOPS per $/hr

Best for Inference

NVIDIA H100 SXM5 80GB

4.0 PFLOPS FP8/FP16

Use-Case Recommendations

Large-Scale Training

Training LLMs and large multi-modal models

Winner

H100 SXM5 80GB

2.0 PFLOPS FP16 with 80GB HBM3 provides the best training throughput.

Inference at Scale

Deploying models in production for real-time inference

Winner

H100 SXM5 80GB

4.0 PFLOPS FP8/FP16 gives superior inference throughput.

Budget-Conscious Workloads

Getting the best performance per dollar

Winner

H100 SXM5 80GB

Starting at $2.20/hr delivers the best TFLOPS per dollar.

Learn More