NVIDIA A100 80GB PCIe vs NVIDIA L40 48GB
A100 80GB PCIe has 67% more memory than L40 48GB (80 GB vs 48 GB). Its memory bandwidth is 124% more (1.935 vs 0.864 TB/s). In the cloud, the median on-demand A100 80GB PCIe price is $4.78 per GPU-hour against $0.86 for the L40 48GB. Per gigabyte of GPU memory, that is $0.060 vs $0.018 per hour.
Specifications
Manufacturer figures as recorded by Kovara. Higher memory and bandwidth are highlighted; neither is a benchmark of your workload.
| A100 80GB PCIe | L40 48GB | |
|---|---|---|
| Manufacturer | NVIDIA | NVIDIA |
| Architecture | Ampere | Ada Lovelace |
| Memory | 80 GB HBM2E | 48 GB GDDR6 |
| Memory bandwidth | 1.935 TB/s | 0.864 TB/s |
| Power (TDP) | 300 W | 300 W |
| Form factor | PCIe dual-slot | PCIe dual-slot FHFL |
Cloud prices today
Public offers Kovara tracks, normalised to US dollars per GPU-hour. Prices use on-demand terms where available; spot and reserved offers are on each GPU’s price pages. Lower is highlighted.
| A100 80GB PCIe | L40 48GB | |
|---|---|---|
| Lowest price | $1.59 | $0.79 |
| Median price | $4.78 | $0.86 |
| Highest price | $7.35 | $1.25 |
| Median per GB of memory | $0.060 | $0.018 |
| Public offers (all terms) | 83 | 13 |
| Providers | 5 | 5 |
Where to rent each
Lowest tracked price per provider. Open a provider to see every offer, region and term.
Questions
Which is cheaper to rent, the A100 80GB PCIe or the L40 48GB?
Right now the cheapest on-demand A100 80GB PCIe Kovara tracks is $1.59 per GPU-hour and the cheapest L40 48GB is $0.79. Medians are $4.78 and $0.86. The cheaper GPU per hour is not always cheaper per job: a faster or larger GPU can finish sooner or need fewer units.
Which has more memory?
The A100 80GB PCIe has more memory: 80 GB against 48 GB. Memory decides which models and batch sizes fit on one GPU.
Are these benchmark results?
No. This page compares manufacturer specifications and public prices. Real performance depends on the model, precision, software and interconnect. Kova can help you estimate cost for a specific workload.