Datacenter · Ada Lovelace · launched 2023 · prices checked 2026-09-03
Rent NVIDIA L40S — 48 GB, $0.466/hr on-demand
- VRAM 48 GBGDDR6
- FP16 tensor 362TFLOPS
- PowerScore 255RTX 3090 = 100
- Configs 1–8×PCIe 4.0
- Online now 59 8 regions
Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — the whole fee schedule. Next weekly market re-check: 2026-09-10.
Datacenter · Ada Lovelace architecture
The L40S is NVIDIA's universal datacenter card for Ada: 48 GB GDDR6, 18,176 CUDA cores, FP8 support and passive cooling at 350 W. It is the go-to for 24/7 inference endpoints, SDXL/Flux serving and 32B-class LLMs — server-grade reliability at a consumer-adjacent price.
A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.
NVIDIA L40S specs: VRAM, TFLOPS, bandwidth
| GPU model | NVIDIA L40S | Architecture | Ada Lovelace (2023) |
|---|---|---|---|
| VRAM | 48 GB GDDR6 | Memory bandwidth | 864 GB/s |
| FP16 tensor perf. | 362 TFLOPS | FP32 perf. | 91.6 TFLOPS |
| CUDA cores | 18,176 | TDP | 350 W |
| PowerScore (RTX 3090 = 100) | 255 | PCIe generation | Gen 4.0 |
| Multi-GPU | 1× – 8× | Max instance storage | 8,000 GB NVMe |
| Network up to | 5,000 Mbps | CUDA | 12.4 – 13.0 |
Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.
L40S price per hour: on-demand, interruptible, reserved
One public rule sets every price on this page: the marketplace median for the L40S ($0.67/hr, snapshot 2026-09-03) × 0.70, rounded down — so on-demand is $0.466, 30% below market. Interruptible halves it; a 3-month reservation takes another 35% off.
| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
|---|---|---|---|---|
| On-demand | $0.466 | $11.18 | $340 | Guaranteed capacity, price locked at deploy, stop anytime |
| Interruptible | $0.233 | $5.59 | $170 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| Reserved (3 months) | $0.302 | $7.25 | $220 | −35% on on-demand, rate locked for the term, capacity held |
Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the GPU cost calculator.
What you can run on a L40S (48 GB VRAM)
With 48 GB of GDDR6, a single card holds a ~14B-parameter LLM in FP16 or up to ~72B parameters quantized to 4-bit, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.466.
- Recommended for llm inference — Best $/token in class
- Recommended for image generation — Flux dev ~2 s/image on 4090
- Recommended for video generation — 32–141 GB VRAM on tap
- One-click template: vLLM on a L40S
- One-click template: PyTorch on a L40S
- One-click template: ComfyUI on a L40S
- Sizing help: LLM VRAM requirements guide
L40S availability by region
59 × L40S across 10 machines, live from inventory:
London
Frankfurt
Ashburn, VA- Chicago, IL
Sydney- Los Angeles, CA
Singapore- Seattle, WA
L40S vs alternatives: price per TFLOP
| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
|---|---|---|---|---|
| L40S this card | 48 GB | 362 | $0.466 | $1.29‰ |
| A100 PCIE | 80 GB | 312 | $0.662 | $2.12‰ |
| L40 | 48 GB | 181 | $0.234 | $1.29‰ |
| L4 | 24 GB | 121 | $0.225 | $1.86‰ |
| A10 | 24 GB | 125 | $0.168 | $1.34‰ |
‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.
- RTX 4090 vs L40S — specs, price per hour, which to rent
- L40S vs A100 PCIE — specs, price per hour, which to rent
- L40S vs H100 PCIE — specs, price per hour, which to rent
- RTX 5090 vs L40S — specs, price per hour, which to rent
- All GPU comparisons
Renting a L40S: frequently asked questions
How much does it cost to rent an NVIDIA L40S per hour?
$0.466 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.67. Interruptible capacity costs $0.233/hr and a 3-month reservation $0.302/hr. Around $340/month if you keep one running non-stop, billed per second.
What can a L40S with 48 GB VRAM run?
In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.
Is the L40S available to rent right now?
Yes — 59 GPUs across 10 machines in 8 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.
How do I deploy a L40S?
Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by L40S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu l40s --template pytorch.
Why is the L40S cheaper here than on GPU marketplaces?
We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-10) and published — no auctions, no per-host roulette, no bidding.