L40S 48GB Price Comparison

48GB VRAM · Data Center GPU · Sweet-spot 48GB VRAM GPU — optimized for quantized 70B inference and batch inference · Updated: 2026-10-26

⚡ Direct answer：L40S 48GB is currently cheapest at RunPod (USA), ¥5.69/hour (spot), ~¥137/day. Prices update hourly; final order page prevails.

Currently monitoring lower prices

¥5.69/hour

RunPod (USA) · spot

PromotionSuanjiayunA100/5090/4090 from ¥1.24 per GPU-hour; 100+ model images availableLearn more →

Platform	Price ⇅	24 hours	Inventory	Billing Information
RunPod (USA)Top overseas favoriteNorth America Stability 61·C	¥5.69/hourCurrently Monitored Low PriceReal-timeSpot	¥137/day	sufficient	RunPod Community Cloud lowest price: $0.79/hr (NVIDIA L40S; Secure Cloud: $1.09/hr)	Rent Now
Verda (Finland)European value-for-moneyEurope Stability 62·C	¥6.58/hour▲1.1%Real-timeSpot	¥158/day	sufficient	Verda (Finland, formerly DataCrunch) spot: $0.9135/hour (1× L40S 48GB); on-demand: $1.827/hour	Rent Now
Gcore (Luxembourg)European edge cloudEurope Stability 59·D	¥9.91/hourReal-timePay-as-you-go	¥238/day	sufficient	Gcore (Luxembourg, global edge cloud): pay-as-you-go rate starting at €1.27/hour (L40S)	Rent Now
Crusoe (USA)Green energyNorth America Stability 59·D	¥10.80/hourReal-timePay-as-you-go	¥259/day	sufficient	Crusoe (USA, powered by clean energy): pay-as-you-go rate starting at $1.50/GPU·hr (NVIDIA L40S 48GB)	Rent Now
Vultr (USA)Maximum number of data centersNorth America Data accumulation in progress: 7/29 days	¥12.03/hourReal-timePay-as-you-go	¥289/day	sufficient	Vultr official on-demand price: $1.671/GPU·hour (NVIDIA L40S, 48GB VRAM full server, prorated per GPU)	Rent Now

Some “Go to Rent” links include affiliate commissions, which do not affect ranking or indexing.

📈 L40S 48GB price trends (lowest daily price over the past 14 days)

14-day low

¥9.91

10-13 · Gcore (Luxembourg)

14-day high

¥9.91

10-13 · Gcore (Luxembourg)

Average within range

¥9.91

Fluctuation: stable

Market Median Price

¥10.80

8% above lowest price · Quoted by 3 platforms

Current vs 30-Day Lowest

No change

9.91 / 9.91

Price Range Change

No change

10-13 → 10-26

Lowest price across all platforms (¥/hour) Market Median Price (median of all platform quotes) Lowest point in range (with date and platform indicated)

Hover over the dot to see the specific platform for that day. Prices are stable with almost no fluctuations—you can place your order directly at current rates.

Is 48GB VRAM sufficient?
Not sure which GPU your task needs? Try PickerWorkflow: select task → system auto-recommends optimal GPU models and their real-time prices across platforms.

Weekly/monthly plans are typically 20–40% cheaper than pay-as-you-go. For long-running tasks (>1 week), check monthly pricing first. Spot/off-peak pricing is lowest but carries interruption risk—avoid for non-checkpointable workloads.

FAQ

How much does it cost to rent an L40S 48GB for one hour?

Pricing varies across platforms; the lowest observed rate is ¥5.69/hour (RunPod US, spot). Weekly or monthly plans typically offer an additional 20–40% discount; spot/idle instances are cheaper but carry interruption risk.

What tasks is the L40S 48GB suitable for?

L40S 48GB (48GB VRAM): a sweet-spot GPU for 48GB VRAM workloads—ideal for 70B quantized inference and batch inference. Unsure which GPU to rent? Use our free hardware recommendation tool, which auto-suggests optimal GPUs by task type.

Which platform offers the cheapest L40S 48GB rental?

Prices change hourly—this site automatically scrapes and permanently sorts by price across 26 global platforms. The top row in the table above shows the currently lowest-priced platform; click “Rent Now” to go directly to its official website for ordering. Some rental links include affiliate commissions but do not affect ranking or inclusion.

Cite This Site(AI and articles may cite directly under CC BY 4.0):

L40S 48GB 租用price｜Data Source：GPU computing power radar rentgpu.link（2026-10-26 更新，per hour自动抓取各Platform公开定价）｜https://rentgpu.link/gpu/l40s｜CC BY 4.0

Related Guides

Model Fine-tuningVRAM

How Much VRAM Do You Need to Fine-tune a 7B Model? LoRA / QLoRA / Full-parameter Training — All in One Table.

VRAM requirements for fine-tuning 7B models explained: LoRA needs 16 GB, QLoRA needs 12 GB, full-parameter fine-tuning requires 80 GB — clear guidance on which GPU to rent.

2026-09-28

Avoid pitfallsCost

GPU rental pitfalls guide: 5 hidden costs beyond listed prices

Why does the final bill double for two GPUs both priced at ¥2/hour? Full breakdown: storage fees, bandwidth fees, billing during shutdown, minimum spend requirements, and image-related pitfalls.

2026-09-28

Save moneyBilling

Pay-as-you-go, daily, or monthly — which plan suits you? A simple arithmetic question

How to Choose a GPU Rental Billing Model: Break-even Formula + Cost-Optimized Plans for Every Usage Tier — Stop Overthinking It.

2026-09-28