Nvidia H20

Export-compliant Hopper inference GPU with lower compute than H100.

Arch: Hopper
Memory: 96 GB HBM3
Bandwidth: 4.00 TB/s
Released: Q1 2024

Compare with another GPU

At a glance

We don't currently track any H20 offers. See alternative GPUs.

Technical Specifications

Nvidia H20 · Per GPU

Compute
Precision support	FP8FP16BF16TF32FP32FP64INT8

Memory
Capacity	96 GB HBM3
Bandwidth	4,000 GB/s
ECC	Yes

Silicon
Architecture	Hopper

Fabric and host
Host interface	PCIe 5.0 x16

Power
Board power	400 W

Platform
Partitioning	MIG, up to 7 instances

Lium

Built for agents: CLI, SDK and MCP server. Your agent rents and runs GPUs itself.

A permissionless marketplace. 1000+ nodes from independent providers worldwide.

Hard to find nodes available; B200 and B300 Blackwell nodes listed now.

Consumer + DC grade cards for cheap. H100s from $1/hr, RTX 5090s from $0.35/hr.

View Pricing

Frequently Asked Questions

Why choose the H20?

96GB HBM3 at lower cost than H100. Large memory capacity suits inference for 70B models. Available in China where H100/H200 are restricted.

When is the H20 not a good fit?

Significantly reduced compute compared to H100. PCIe-only with no NVLink support. Better suited for memory-bound inference than training or latency-sensitive workloads.

Alternatives to Nvidia H20

Nvidia H100 80 GB HBM3 · 16% less bandwidth than the H20
From $1.41 /GPU/hr Compare
Nvidia A100 80 GB HBM2e · 52% less bandwidth than the H20
From $0.43 /GPU/hr Compare
Nvidia L40S 48 GB GDDR6 · 78% less bandwidth than the H20
From $0.38 /GPU/hr Compare

Browse all GPUs

Last updated Sept. 14, 2026