Two Nvidia data center GPUs, one Hopper and one Pascal: 80 GB HBM2e against 24 GB GDDR5, 5.8x the memory bandwidth on the H800, and NVLink 400 GB/s on the H800 where the P40 is PCIe only.
At a glance
Memory
- H800
- 80 GB
- P40
- 24 GB
3.3x more on H800
Memory bandwidth
- H800
- 2,000 GB/s
- P40
- 346 GB/s
5.8x higher on H800
Cheapest on-demand
- H800
- --
- P40
- --
Price comparison
Median price per GPU per hour by billing type, with the cheapest listing under it.
Price history
H800 vs P40 price history
Weekly median on-demand price per GPU per hour
Specs comparison
| Spec | H800 | P40 | Difference |
|---|---|---|---|
| Memory | |||
| Capacity | 80 GB HBM2e | 24 GB GDDR5 | 3.3x more on H800 |
| Bandwidth | 2,000 GB/s | 346 GB/s | 5.8x higher on H800 |
| Bus width | 5,120-bit | 384-bit | 13x wider on H800 |
| Compute · dense | |||
| FP8 | 1,513 TFLOPS | Not supported | H800 only |
| FP16 / BF16 | 756.5 TFLOPS | ||
| INT8 | 1,513 TOPS | 47 TOPS | 32x higher on H800 |
| FP32 | 51 TFLOPS | 12 TFLOPS | 4.2x higher on H800 |
| Platform | |||
| Interconnect | NVLink 400 GB/s | PCIe only | NVLink on H800 only |
| Board power | 350 W | 250 W | 1.4x higher on H800 |
| Released | Q1 2023 | Q3 2016 | 6 years newer on H800 |
| Precision | |||
| Supported formats | FP8FP16BF16TF32FP32FP64INT8 | FP16FP32INT8 | |
Compute: Vendor peak figures, dense and per GPU. Not measured throughput.
Source: P40 datasheet.
Similar GPUs
Other comparisons in the same class.
Our data for Nvidia H800 was last updated on Sept. 7, 2026, and for Nvidia P40 on Sept. 7, 2026.