Two Nvidia data center GPUs, one Ampere and one Blackwell: 48 GB GDDR6 against 180 GB HBM3e, 11x the memory bandwidth on the B200, and NVLink 1,800 GB/s on the B200 where the A40 is PCIe only.
At a glance
Updated 6 minutes ago- Memory
- 48 GB
- GDDR6
- Memory bandwidth
- 696 GB/s
- From
- $1.05 /GPU/hr
- 6 providers
- Memory
- 180 GB
- HBM3e
- Memory bandwidth
- 7.70 TB/s
- From
- $3.75 /GPU/hr
- 35 providers
Price comparison
Cheapest listed price per GPU per hour by billing type, and the median across providers.
| Billing | A40 | B200 | Difference |
|---|---|---|---|
| On-demand | $1.05 | $3.75 | 3.6x more expensive on B200 |
| Spot | No spot listings | $3.12 | |
| Reserved | $0.61 | $3.75 | 6.2x more expensive on B200 |
| Median | $1.05 | $6.83 | 6.5x more expensive on B200 |
Median: Median across providers, one vote each, so a large catalog does not outweigh a small one.
Price history
A40 vs B200 price history
Weekly median on-demand price per GPU per hour
Specs comparison
| Spec | A40 | B200 | Difference |
|---|---|---|---|
| Memory | |||
| Capacity | 48 GB GDDR6 | 180 GB HBM3e | 3.8x more on B200 |
| Bandwidth | 696 GB/s | 7,700 GB/s | 11x higher on B200 |
| Bus width | 384-bit | ||
| Compute · dense | |||
| FP4 | Not supported | 9,000 TFLOPS | B200 only |
| FP8 | Not supported | 4,500 TFLOPS | B200 only |
| FP16 / BF16 | 149.7 TFLOPS | 2,250 TFLOPS | 15x higher on B200 |
| INT8 | 299.3 TOPS | 4,500 TOPS | 15x higher on B200 |
| FP32 | 37.4 TFLOPS | 75 TFLOPS | 2x higher on B200 |
| FP64 | Not supported | 37 TFLOPS | B200 only |
| Platform | |||
| Interconnect | PCIe only | NVLink 1,800 GB/s | NVLink on B200 only |
| Board power | 300 W | 1,000 W | 3.3x higher on B200 |
| Released | Q4 2020 | Q1 2024 | 3 years newer on B200 |
| Precision | |||
| Supported formats | FP16BF16TF32FP32INT4INT8 | FP4FP6FP8FP16BF16TF32FP32FP64INT8 | |
Compute: Vendor peak figures, dense and per GPU. Not measured throughput.
Source: A40 datasheet and B200 datasheet.
Where to rent them
Providers listing these GPUs, those with both first, so you can switch without moving clouds.
| Provider | A40 /GPU/hr | B200 /GPU/hr |
|---|---|---|
Runpod
|
Reservation on request | On-Demand $6.79 |
EmpirioLabs AI
|
On-Demand $1.49 | On-Demand $9.00 |
Exoscale
|
On-Demand $1.05 | -- |
Database Mart
|
Reserved (24mo) $0.61 | -- |
Utho
|
Reserved (36mo) $0.65 | -- |
Sesterce
|
On-Demand $1.21 | -- |
Packet·ai
|
-- | On-Demand $3.75 |
Vast.ai
|
-- | On-Demand $5.31 |
GPU.ai
|
-- | On-Demand $5.40 |
Lium
|
-- | On-Demand $5.50 |
Spheron
|
-- | On-Demand $7.20 |
Oblivus
|
-- | On-Demand $8.40 |
6 providers list the A40 and 35 the B200. Every listing is on the A40 page and the B200 page.
Similar GPUs
Other comparisons in the same class.
Our data for Nvidia A40 was last updated on Sept. 7, 2026, and for Nvidia B200 on Sept. 7, 2026.