A100 vs B200
A100 starts at $0.89/GPU-hour on-demand; B200 starts at $3.49/GPU-hour on-demand. Compare the exact variants, memory and published specifications below. Peak compute figures do not establish application speed.
A100
B200
Spec sheet
| A100 | B200 | |
|---|---|---|
| Spec variant | A100 SXM 40GB | B200 |
| Architecture | Ampere | Blackwell |
| VRAM (spec variant) | 40 GB | 180 GB |
| FP16 TFLOPS (dense) | 312 | 2250 |
| Memory bandwidth | 1555 GB/s | 8000 GB/s |
| TDP | 400 W | 1000 W |
| Interconnect | NVLink 600GB/s | NVLink 1800GB/s |
| Launched | 2020 | 2024 |
Rental price per manufacturer-rated FP16 compute: A100 $2.85 per PFLOP-hour vs B200 $1.55. Formula: cheapest $/hr ÷ dense FP16 PFLOPS (1 PFLOP = 1000 TFLOPS). This uses A100 SXM 40GB and B200, whose datasheets may differ from the variants above. It is not measured job throughput. Specs: A100 manufacturer, B200 manufacturer.
What they cost to rent (2026-10-02)
| A100 | B200 | |
|---|---|---|
| Cheapest on-demand $/hr | $0.89 | $3.49 |
| Cheapest spot/community $/hr | $0.679 | $3.20 |
| Cheapest source | Jarvislabs | Prime Intellect |
| Providers offering it | 21 | 18 |
Lowest recorded on-demand rate per configuration; highlighted = cheaper. Spot/marketplace tiers shown on the individual pages.
FAQ
A100 is cheaper: the A100 rents from $0.89 per GPU-hour and the B200 from $3.49 on-demand, updated 2026-10-02.
The largest recorded A100 variant has 80 GB and the largest B200 variant has 180 GB. Other variants may have less memory. The B200 has more.
Compare more rental options: the screener · price movers · methodology.