Hopper (GH100) architecture • 80 GB HBM3

NVIDIA H100 (80GB HBM3) hourly pricing

The NVIDIA H100 Hopper GPU is the enterprise benchmark accelerator for modern generative AI. With fourth-generation Tensor Cores, a dedicated Transformer Engine, and 3.35 TB/s memory bandwidth, it delivers up to 4x faster training and 30x faster inference compared to the previous generation A100.

Starting from $1.85/hr on DePIN

Save up to 52% compared to AWS EC2

H100 price per hour across networks

Ranked from lowest hourly rate

Network	Network type	Hourly price	Billing type	Status	Audit date
Akash Network	DePIN (Cosmos)	$1.85 / hr52% less than AWS	dynamic bid	Last checked: 2026-10-01	2026-10-01Lease on Akash
io.net	DePIN (Solana)	$1.98 / hr49% less than AWS	hourly	Last checked: 2026-09-30	2026-09-30Lease on io.net
Aethir	DePIN (Arbitrum)	$2.10 / hr46% less than AWS	hourly	Last checked: 2026-09-29	2026-09-29Lease on Aethir
Amazon Web Services (AWS)Benchmark	Centralized cloud	$3.88 / hr	hourly	Last checked: 2026-10-02	2026-10-02Lease on AWS
Render Network	DePIN (Solana)	Not verified yetRender Network primarily routes workloads to consumer and workstation GPUs (RTX 4090/3090/A6000). Direct H100 hourly booking is not currently verified on public tiers.	job based	Not verified yet	2026-10-01Lease on Render

NVIDIA H100 (80GB HBM3) technical specifications

ArchitectureHopper (GH100)

Total VRAM80 GB HBM3

Memory bandwidth3.35 TB/s

FP16 compute989 TFLOPS (dense) / 1,979 TFLOPS (sparse)

FP8 tensor compute1,979 TFLOPS (dense) / 3,958 TFLOPS (sparse)

InterconnectNVLink 4.0 (900 GB/s bidirectional)

TDP700W (SXM5) / 350W (PCIe)

Optimal workloads

•Frontier Large Language Model Pre-Training
•Full-Parameter Fine-Tuning (>70B Parameters)
•High-Throughput FP8 / FP16 Production Inference
•Massive Scale Deep Learning Research

Buyer advice

If your workload requires low-latency serving of 70B+ models (like Llama 3 70B, DeepSeek, or Mixtral 8x22B) or full-parameter training, H100 is unmatched. On decentralized networks like Akash and io.net, you can rent H100 instances at $1.85–$2.10/hr compared to $3.88+/hr on AWS EC2 p5.48xlarge.

Disclosure: Some links on this page may be affiliate referral links. When you lease compute through these links, DecentralGPU may earn a small referral commission at no additional cost to you. See our affiliate disclosure and pricing methodology.