Two Nvidia data center GPUs on the Hopper architecture: 96 GB HBM3 against 80 GB HBM2e, 2x the memory bandwidth on the H20, and NVLink 400 GB/s on the H800 where the H20 is PCIe only.
At a glance
Memory
- H20
- 96 GB
- H800
- 80 GB
1.2x more on H20
Memory bandwidth
- H20
- 4.00 TB/s
- H800
- 2.00 TB/s
2x higher on H20
Cheapest on-demand
- H20
- --
- H800
- --
Price comparison
Median price per GPU per hour by billing type, with the cheapest listing under it.
Specs comparison
| Spec | H20 | H800 | Difference |
|---|---|---|---|
| Memory | |||
| Capacity | 96 GB HBM3 | 80 GB HBM2e | 1.2x more on H20 |
| Bandwidth | 4,000 GB/s | 2,000 GB/s | 2x higher on H20 |
| Bus width | 5,120-bit | ||
| Platform | |||
| Interconnect | PCIe only | NVLink 400 GB/s | NVLink on H800 only |
| Board power | 400 W | 350 W | 1.1x higher on H20 |
| Released | Q1 2024 | Q1 2023 | 10 months newer on H20 |
| Precision | |||
| Supported formats | FP8FP16BF16TF32FP32FP64INT8 | FP8FP16BF16TF32FP32FP64INT8 | |
Similar GPUs
Other comparisons in the same class.
-
Nvidia H100
From $1.41 /GPU/hr Compare -
Nvidia A100
From $0.47 /GPU/hr Compare -
Nvidia L40S
From $0.38 /GPU/hr Compare -
Nvidia H200
From $2.09 /GPU/hr Compare
Our data for Nvidia H20 was last updated on Sept. 7, 2026, and for Nvidia H800 on Sept. 7, 2026.