Two Nvidia data center GPUs, one Grace Blackwell and one Ada Lovelace: 186 GB HBM3e against 48 GB GDDR6, 9.3x the memory bandwidth on the GB200, and NVLink 1,800 GB/s on the GB200 where the L40S is PCIe only.
At a glance
Memory
- GB200
- 186 GB
- L40S
- 48 GB
3.9x more on GB200
Memory bandwidth
- GB200
- 8,000 GB/s
- L40S
- 864 GB/s
9.3x higher on GB200
Cheapest on-demand
- GB200
- $10.50 /GPU/hr
- L40S
- $0.55 /GPU/hr
19x more expensive on GB200
Price comparison
Median price per GPU per hour by billing type, with the cheapest listing under it.
| Billing | GB200 | L40S | Difference |
|---|---|---|---|
| On-demand | $16.00 /GPU/hr from $10.50 | $1.64 /GPU/hr from $0.55 | 9.8x more expensive on GB200 |
| Reserved | -- /GPU/hr from $10.58 | $1.11 /GPU/hr from $0.60 (12 mo) | |
| Spot | -- /GPU/hr from $27.04 | $1.11 /GPU/hr from $0.74 |
Heads up: A provider's own page may quote a different figure, for example on tax, region or a promotion. A monthly-only plan shows a derived hourly rate. Medians are across providers, one vote each, so a large catalog does not outweigh a small one. No median for reserved and spot on the GB200: we only show one when at least 3 providers list the GPU on that billing type. Verify before provisioning. More on how we price.
Price history
GB200 vs L40S price history
Weekly median on-demand price per GPU per hour
Specs comparison
| Spec | GB200 | L40S | Difference |
|---|---|---|---|
| Memory | |||
| Capacity | 186 GB HBM3e | 48 GB GDDR6 | 3.9x more on GB200 |
| Bandwidth | 8,000 GB/s | 864 GB/s | 9.3x higher on GB200 |
| Bus width | 384-bit | ||
| Compute · dense | |||
| FP4 | 10,000 TFLOPS | Not supported | GB200 only |
| FP8 | 5,000 TFLOPS | 733 TFLOPS | 6.8x higher on GB200 |
| FP16 / BF16 | 2,500 TFLOPS | 362.1 TFLOPS | 6.9x higher on GB200 |
| INT8 | 5,000 TOPS | 733 TOPS | 6.8x higher on GB200 |
| FP32 | 80 TFLOPS | 91.6 TFLOPS | 1.1x higher on L40S |
| FP64 | 40 TFLOPS | Not supported | GB200 only |
| Platform | |||
| Interconnect | NVLink 1,800 GB/s | PCIe only | NVLink on GB200 only |
| Board power | 1,200 W | 350 W | 3.4x higher on GB200 |
| Released | Q1 2024 | Q3 2023 | 7 months newer on GB200 |
| Precision | |||
| Supported formats | FP4FP6FP8FP16BF16TF32FP32FP64INT8 | FP8FP16BF16TF32FP32INT4INT8 | |
Compute: Vendor peak figures, dense and per GPU. Not measured throughput.
Source: GB200 datasheet and L40S datasheet.
Where to rent them
Providers listing these GPUs, those with both first, so you can switch without moving clouds.
| Provider | GB200 /GPU/hr | L40S /GPU/hr |
|---|---|---|
CoreWeave
|
On-Demand $10.50 | -- |
Oracle Cloud
|
On-Demand $16.00 | -- |
Azure
|
On-Demand $27.04 | -- |
AWS
|
Reservation $10.58 | -- |
CUDO
|
Custom on request | -- |
Canopy Wave
|
Custom on request | -- |
Nebius
|
Custom on request | -- |
Together AI
|
Custom on request | -- |
Novita
|
-- | On-Demand $0.55 |
GPU.ai
|
-- | On-Demand $0.80 |
Packet·ai
|
-- | On-Demand $0.92 |
Sesterce
|
-- | On-Demand $0.97 |
Massed Compute
|
-- | On-Demand $0.97 |
Runcrate
|
-- | On-Demand $0.97 |
UpCloud
|
-- | On-Demand $1.19 |
Lyceum
|
-- | On-Demand $1.19 |
8 providers list the GB200 and 34 the L40S. Every listing is on the GB200 page and the L40S page.
Similar GPUs
Other comparisons in the same class.
Our data for Nvidia GB200 was last updated on Sept. 7, 2026, and for Nvidia L40S on Sept. 7, 2026.