Two Nvidia data center GPUs on the Hopper architecture: 96 GB HBM3 against 141 GB HBM3e, 20% more memory bandwidth on the H200, and NVLink 900 GB/s on the H200 where the H20 is PCIe only.
At a glance
Updated 4 minutes ago- Memory
- 96 GB
- HBM3
- Memory bandwidth
- 4.00 TB/s
- From
- --
- Memory
- 141 GB
- HBM3e
- Memory bandwidth
- 4.80 TB/s
- From
- $2.09 /GPU/hr
- 45 providers
Price comparison
Cheapest listed price per GPU per hour by billing type, and the median across providers.
| Billing | H20 | H200 | Difference |
|---|---|---|---|
| On-demand | -- | $2.09 | |
| Spot | No spot listings | $1.32 | |
| Reserved | No reserved listings | $2.29 | |
| Median | -- | $4.29 |
Median: Median across providers, one vote each, so a large catalog does not outweigh a small one.
Price history
H20 vs H200 price history
Weekly median on-demand price per GPU per hour
Specs comparison
| Spec | H20 | H200 | Difference |
|---|---|---|---|
| Memory | |||
| Capacity | 96 GB HBM3 | 141 GB HBM3e | 1.5x more on H200 |
| Bandwidth | 4,000 GB/s | 4,800 GB/s | 1.2x higher on H200 |
| Platform | |||
| Interconnect | PCIe only | NVLink 900 GB/s | NVLink on H200 only |
| Board power | 400 W | 700 W | 1.8x higher on H200 |
| Released | Q1 2024 | Q4 2023 | |
| Precision | |||
| Supported formats | FP8FP16BF16TF32FP32FP64INT8 | FP8FP16BF16TF32FP32FP64INT8 | |
Source: H200 datasheet.
Where to rent them
Providers listing these GPUs, those with both first, so you can switch without moving clouds.
| Provider | H20 /GPU/hr | H200 /GPU/hr |
|---|---|---|
Koyeb
|
-- | On-Demand $3.00 |
Lium
|
-- | On-Demand $3.00 |
GPU.ai
|
-- | On-Demand $3.29 |
Vast.ai
|
-- | On-Demand $3.59 |
Massed Compute
|
-- | On-Demand $3.62 |
Verda
|
-- | On-Demand $4.00 |
Lyceum
|
-- | On-Demand $4.29 |
Runpod
|
-- | On-Demand $4.59 |
0 providers list the H20 and 45 the H200. Every listing is on the H20 page and the H200 page.
Similar GPUs
Other comparisons in the same class.
Our data for Nvidia H20 was last updated on Sept. 7, 2026, and for Nvidia H200 on Sept. 7, 2026.