Two Nvidia data center GPUs, one Hopper and one Ada Lovelace: 80 GB HBM3 against 48 GB GDDR6, 3.9x the memory bandwidth on the H100, and NVLink 900 GB/s on the H100 where the L40 is PCIe only.
At a glance
Updated 6 minutes ago- Memory
- 80 GB
- HBM3 · 94 GB on NVL
- Memory bandwidth
- 3.35 TB/s
- 2,000 GB/s on PCIe · 3,938 GB/s on NVL
- From
- $1.50 /GPU/hr
- 53 providers
- Memory
- 48 GB
- GDDR6
- Memory bandwidth
- 864 GB/s
- From
- $0.33 /GPU/hr
- 14 providers
Price comparison
Cheapest listed price per GPU per hour by billing type, and the median across providers.
| Billing | H100 | L40 | Difference |
|---|---|---|---|
| On-demand | $1.50 | $0.33 | 4.5x more expensive on H100 |
| Spot | $0.02 | $0.40 | 20x more expensive on L40 |
| Reserved | $1.53 | $0.70 | 2.2x more expensive on H100 |
| Median | $3.39 | $0.98 | 3.4x more expensive on H100 |
Median: Median across providers, one vote each, so a large catalog does not outweigh a small one.
Price history
H100 vs L40 price history
Weekly median on-demand price per GPU per hour
Specs comparison
| Spec | H100 | L40 | Difference |
|---|---|---|---|
| Memory | |||
| Capacity | 80 GB HBM3 94 GB on NVL | 48 GB GDDR6 | 1.7x more on H100 |
| Bandwidth | 3,350 GB/s 2,000 GB/s on PCIe 3,938 GB/s on NVL | 864 GB/s | 3.9x higher on H100 |
| Bus width | 5,120-bit | 384-bit | 13x wider on H100 |
| Compute · dense | |||
| FP8 | 1,979 TFLOPS | 362 TFLOPS | 5.5x higher on H100 |
| FP16 / BF16 | 989.5 TFLOPS | 181.1 TFLOPS | 5.5x higher on H100 |
| INT8 | 1,979 TOPS | 362 TOPS | 5.5x higher on H100 |
| FP32 | 67 TFLOPS | 90.5 TFLOPS | 1.4x higher on L40 |
| FP64 | 34 TFLOPS | Not supported | H100 only |
| Platform | |||
| Interconnect | NVLink 900 GB/s NVLink 600 GB/s on PCIe, NVL | PCIe only | NVLink on H100 only |
| Board power | 700 W | 300 W | 2.3x higher on H100 |
| Released | Q3 2022 | Q4 2022 | |
| Precision | |||
| Supported formats | FP8FP16BF16TF32FP32FP64INT8 | FP8FP16BF16TF32FP32INT4INT8 | |
Figures are for each model's headline configuration. Where a form factor carries a different number, it is named under the figure.
Compute: Vendor peak figures, dense and per GPU. Not measured throughput.
Source: H100 datasheet and L40 datasheet.
Where to rent them
Providers listing these GPUs, those with both first, so you can switch without moving clouds.
| Provider | H100 /GPU/hr | L40 /GPU/hr |
|---|---|---|
Lium
|
On-Demand $1.50 | On-Demand $0.33 |
Vast.ai
|
On-Demand $1.73 | On-Demand $0.46 |
GPU.ai
|
On-Demand $1.98 | On-Demand $0.78 |
Massed Compute
|
On-Demand $2.73 | On-Demand $0.86 |
HyperAI
|
On-Demand $1.80 | -- |
Runcrate
|
On-Demand $2.19 | -- |
Koyeb
|
On-Demand $2.50 | -- |
Spheron
|
On-Demand $2.64 | -- |
Runpod
|
-- | On-Demand $0.69 |
EmpirioLabs AI
|
-- | On-Demand $0.70 |
Thunder Compute
|
-- | On-Demand $0.79 |
Sesterce
|
-- | On-Demand $0.97 |
53 providers list the H100 and 14 the L40. Every listing is on the H100 page and the L40 page.
Similar GPUs
Other comparisons in the same class.
Our data for Nvidia H100 was last updated on Sept. 7, 2026, and for Nvidia L40 on Sept. 7, 2026.