NVIDIA L40S rental price per hour
Series 1 · GPU prices · NVIDIA · snapshot 16 Oct 2026
The L40S is a 48 GB PCIe card that rents for $1.53 an hour at the median. Its median has been one of the most volatile in the index, swinging between $0.90 and $1.55 within two months.
Figure 1L40S: median rental price and published range, USD per GPU-hour, 2025-06-01 to 2026-10-14
§1Reading the L40S market
The floor is $0.38 on Lium.io, confirmed through ComputePrices' live tracker; Lium's own page also shows a $0.35 reference rate, but marked unavailable with no pods in stock, so $0.38 is the operative number. A cluster of low-cost clouds follows: Runpod ~$0.79, gpu.ai and Vast.ai ~$0.80, CoreWeave ~$0.99.
At the top, Oracle's bare-metal $3.50 and AWS's $3.77 (an eight-GPU configuration in us-east-1) are roughly four times the cheap cluster. GetDeploying's tracked median reads ~$1.50; the index carries $1.53.
Civo's rate card is a staircase in ten-cent steps: $1.29, $1.19, $1.09, $0.99, $0.89 for on-demand, 6, 12, 24 and 36 months. The tracker originally logged the 36-month figure as the 1-year rate and later corrected it to the 12-month $1.09.
Table 1L40S prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| Lium.io | Marketplace | $0.38 | via ComputePrices live tracker | 2026-09-21 |
| Runpod | On-demand | ~$0.79 | 2026-09-21 | |
| gpu.ai / Vast.ai | Marketplace | ~$0.80 | recorded as one data point for both | 2026-09-21 |
| Civo | Reserved, 36 mo | $0.89 | 2026-09-21 | |
| CoreWeave | On-demand | ~$0.99 | 2026-09-21 | |
| Civo | Reserved, 24 mo | $0.99 | 2026-09-21 | |
| Civo | Reserved, 12 mo | $1.09 | 2026-09-21 | |
| Civo | Reserved, 6 mo | $1.19 | 2026-09-21 | |
| Civo | On-demand | $1.29 | 2026-09-21 | |
| GetDeploying median | Third-party index | ~$1.50 | 2026-09-21 | |
| Oracle Cloud | On-demand | $3.50 | bare metal | 2026-09-21 |
| AWS | On-demand | $3.77 | us-east-1, 8-GPU configuration | 2026-09-21 |
Rows are the prices recorded in the tracker note for this model; 6 of the 12 come from the provider's own price page, the rest through comparison trackers. Third-party index rows are shown for context and not counted as providers. Every page behind this table: Sources, L40S.
§2Specifications
Table 2L40S (single PCIe card)
| Memory per GPU (index) | 48 GB |
| Architecture | Ada Lovelace |
| Memory | 48GB GDDR6 with ECC |
| Memory bandwidth | 864GB/s |
| Max power | 350W |
| FP8 Tensor Core | 733 TFLOPS dense; 1,466 with sparsity |
| NVLink | No |
| Form factor | Dual slot, 4.4 in x 10.5 in |
Quoted from NVIDIA — L40S GPU, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the L40S price
- Simple hosting: NVIDIA lists the L40S at 350W in a dual-slot PCIe form factor with no NVLink, so it runs in standard air-cooled servers — many small hosts can offer it.
- Memory type: 48GB of GDDR6 at 864GB/s, against 3.35TB/s of HBM on an H100 SXM; the card has FP8 (733 TFLOPS dense per NVIDIA) but far less memory bandwidth.
- Marketplace pricing: the floor and part of the low-cost cluster are marketplace rates, which move with available stock from one read to the next.
- Long-term discounts: Civo's 36-month price is −31% against its on-demand rate, one of the steeper published ladders.
§4Rent or buy
Continuous rental of one L40S costs $1,117 a month at the median, $650 on Civo's 36-month term and $277 at the marketplace floor.
Because it is a standard PCIe card that fits ordinary servers, the L40S is among the easiest data center parts in the index to own and colocate; our sources simply contain no dated card price, so we do not quote one. For buyers comparing options, the deciding number is usually the per-rack power available in the hosting site rather than the card itself.
Comparing colocation for PCIe GPU servers: published colocation prices.
Other models with their own page: GB200, B200, MI355X, H200, H100, MI300X, A100, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.