A40 vs L4
A40 starts at $0.49/GPU-hour on-demand; L4 starts at $0.44/GPU-hour on-demand. Compare the exact variants, memory and published specifications below. Peak compute figures do not establish application speed.
Spec sheet
| A40 | L4 | |
|---|---|---|
| Spec variant | A40 | L4 |
| Architecture | Ampere | Ada Lovelace |
| VRAM (spec variant) | 48 GB | 24 GB |
| FP16 TFLOPS (dense) | 150 | 121 |
| Memory bandwidth | 696 GB/s | 300 GB/s |
| TDP | 300 W | 72 W |
| Interconnect | NVLink 112.5GB/s | PCIe Gen4 |
| Launched | 2020 | 2023 |
Rental price per manufacturer-rated FP16 compute: A40 $3.27 per PFLOP-hour vs L4 $3.64. Formula: cheapest $/hr ÷ dense FP16 PFLOPS (1 PFLOP = 1000 TFLOPS). This uses A40 and L4, whose datasheets may differ from the variants above. It is not measured job throughput. Specs: A40 manufacturer, L4 manufacturer.
What they cost to rent (2026-10-02)
| A40 | L4 | |
|---|---|---|
| Cheapest on-demand $/hr | $0.49 | $0.44 |
| Cheapest spot/community $/hr | – | $0.105 |
| Cheapest source | RunPod | Jarvislabs |
| Providers offering it | 2 | 10 |
Lowest recorded on-demand rate per configuration; highlighted = cheaper. Spot/marketplace tiers shown on the individual pages.
FAQ
L4 is cheaper: the A40 rents from $0.49 per GPU-hour and the L4 from $0.44 on-demand, updated 2026-10-02.
The largest recorded A40 variant has 48 GB and the largest L4 variant has 24 GB. Other variants may have less memory. The A40 has more.
Compare more rental options: the screener · price movers · methodology.