H100 vs L40S
The H100 delivers 2.7× the FP16 compute of the L40S while costing 3.0× more on average, making the L40S 9% more cost-efficient per TFLOP.
Spec Comparison
Spec
H100
L40S
Architecture
Hopper
Ada Lovelace
VRAM
80 GB
48 GB
FP16 TFLOPS
989
362
Memory Bandwidth
3,350 GB/s
864 GB/s
MSRP
$30,000
$14,000
Used/Market Price
N/A
$9,000
Live Pricing
Cost Efficiency
Metric
H100
L40S
$/TFLOP/hr
$0.0091
$0.0083
$/GB-VRAM/hr
$0.1121
$0.0626
Lower $/TFLOP and $/GB-VRAM is better. Based on median hourly pricing.
Provider Availability
Only on H100 (25)
Only on L40S (8)
H100 Cloud Pricing
327 instances| Provider | GPUs | VRAM | Region | Type | Price/hr | |
|---|---|---|---|---|---|---|
| CHEAPESTVerda | 1x | 80 GB | EU-Central | Spot | $0.801 | Rent |
| Verda | 1x | 80 GB | EU-Central | Spot | $0.801 | Rent |
| Nebius | 1x | 80 GB | eu-north1 | Spot | $1.25 | Rent |
| GCP | 1x | 80 GB | europe-north1-c | Spot | $1.29 | Rent |
| GCP | 1x | 80 GB | EU-West | Spot | $1.39 | Rent |
| Vast.ai | 1x | 80 GB | fr-france | Spot | $1.47 | Rent |
| Vast.ai | 1x | 80 GB | fr-france | On-Demand | $1.55 | Rent |
| Verda | 2x | 160 GB | EU-Central | Spot | $1.60 | Rent |
| Shadeform | 1x | 80 GB | US-East | On-Demand | $1.66 | Rent |
| E2E Networks | 1x | 80 GB | Asia-Pacific | On-Demand | $1.69 | Rent |
L40S Cloud Pricing
180 instances| Provider | GPUs | VRAM | Region | Type | Price/hr | |
|---|---|---|---|---|---|---|
| CHEAPESTVerda | 1x | 48 GB | EU-Central | Spot | $0.320 | Rent |
| AWS | 1x | 44.7 GB | US-East | Spot | $0.446 | Rent |
| Crusoe | 1x | 48 GB | US-West | Spot | $0.500 | Rent |
| AWS | 1x | 44.7 GB | US-East | Spot | $0.532 | Rent |
| AWS | 1x | 44.7 GB | Asia-Pacific | Spot | $0.560 | Rent |
| AWS | 1x | 44.7 GB | US-East | Spot | $0.587 | Rent |
| Verda | 2x | 96 GB | EU-Central | Spot | $0.640 | Rent |
| AWS | 1x | 44.7 GB | Asia-Pacific | Spot | $0.669 | Rent |
| AWS | 1x | 44.7 GB | eu-north-1 | Spot | $0.723 | Rent |
| AWS | 1x | 44.7 GB | eu-north-1 | Spot | $0.723 | Rent |