Rent an NVIDIA L4 from $0.05 per GPU-hour
NVIDIA L4 price per hour: $0.05 per GPU-hour on spot, $0.08 on-demand, $0.06 reserved, billed by the minute. A 72-watt inference card: 24 GB, hardware video engines, and the lowest power draw in the data-center range. Built for many small concurrent jobs rather than one large one.
Billed per started minute. 96 GB of GPU memory on a 4× node. Persistent storage $0.08 per GB-month. No egress fees.
- GPU memory
- 24GBGDDR6
- Bandwidth
- 300GB/sSets token throughput on inference
- Node sizes
- 1× · 2× · 4×Same price per GPU at every size
- Reclaim rate
- <5%Spot instances reclaimed over 30 days
What you get with each L4 node.
| Node | GPU memory | vCPU | RAM | Local NVMe | Spot | On-demand | Action |
|---|---|---|---|---|---|---|---|
| 1× L4 | 24 GB | 8 | 64 GB | 500 GB | $0.05/h | $0.08/h | Configure |
| 2× L4 | 48 GB | 16 | 128 GB | 1 TB | $0.10/h | $0.16/h | Configure |
| 4× L4 | 96 GB | 32 | 256 GB | 2 TB | $0.20/h | $0.32/h | Configure |
GPU-to-GPU link: PCIe Gen4 x16 64 GB/s. Local NVMe is scratch space wiped when the instance ends; keep anything you need on a persistent disk.
96 L4 GPUs free right now.
Straight from the capacity pool the console books against. Prices are identical in every region — pick the one closest to your data.
A region with zero free GPUs still accepts reserved capacity requests — we hold hardware for a term rather than sell what is already taken.
What fits in 24 GB — and what it costs per hour.
Weights need about 2 GB per billion parameters at FP16 / BF16, 1 GB per billion parameters at FP8, 0.5 GB per billion parameters at INT4. We keep 20% of the memory free for activations, the KV cache and the CUDA context, then take the smallest node that still fits.
| Open model | Parameters | FP16 / BF16node · spot price | FP8node · spot price | INT4node · spot price |
|---|---|---|---|---|
| Llama 3.1 8BThe default small model for fine-tuning and agents | 8B | 1× $0.05/h | 1× $0.05/h | 1× $0.05/h |
| Qwen3 14BMid-size dense Qwen3 | 14.8B | 2× $0.10/h | 1× $0.05/h | 1× $0.05/h |
| gpt-oss-20bSmall open-weight reasoning model for local and edge serving | 21B | 4× $0.20/h | 2× $0.10/h | 1× $0.05/h |
| Gemma 4 31BLargest dense Gemma 4, multimodal, 256K context | 31.3B | 4× $0.20/h | 2× $0.10/h | 1× $0.05/h |
| Qwen3.6 35B A3BSmall hybrid-attention mixture of experts | 35B | 4× $0.20/h | 2× $0.10/h | 1× $0.05/h |
| Llama 3.3 70BThe common 70B baseline (same size and layout as Llama 3.1 70B) | 70.6B | over 4× | 4× $0.20/h | 2× $0.10/h |
| gpt-oss-120bOpen-weight reasoning model, mixture of experts, tool use | 117B | over 4× | over 4× | 4× $0.20/h |
| Qwen3 235B A22BFlagship Qwen3 mixture of experts, the 2507 checkpoint | 235B | over 4× | over 4× | over 4× |
| DeepSeek R1Reasoning model, the R1-0528 checkpoint | 671B | over 4× | over 4× | over 4× |
Each model links to its full sizing: KV cache by context length and every GPU that runs it — GPU requirements for 64 open models.
Real usage depends on context length, batch size and the serving engine: a long context can add tens of gigabytes of KV cache. Treat the table as the floor, not the ceiling — and if a job is close to the limit, take the next node size or quantise one step further.
What a dollar buys on an L4.
Ranked 6 of 21 data-center models in this catalogue on cost per gigabyte of GPU memory.
GPU memory you get for $1.00 of spot time, per hour.
Bandwidth decides token throughput far more often than raw FLOPS.
Spot price divided by published FP16 tensor throughput.
One GPU on spot, billed per minute, storage excluded.
Against the rest of the market
37 providers · September 2026- Median on-demand elsewhere
- $0.80Our on-demand is 90% below it
- Cheapest on-demand seen
- $0.11Our on-demand $0.08 is 27% below it
- Cheapest spot seen
- $0.20Our spot $0.05 is 75% below it
- 24 hours on one GPU
- $1.20 on spot$1.92 on-demand · $2.64 at the cheapest rate found elsewhere
Index built from published prices for the same GPU across 102 providers (provider pricing pages and public price indexes), September 2026. How the guarantee is enforced.
L4 price per hour at 28 other providers.
The lowest published on-demand rate we recorded at each provider, cheapest first: $0.11 to $1.65 per GPU-hour, with a median of $0.75. Every row links to its source; a provider without a published price is left out rather than estimated.
| # | Provider | On-demandper GPU-hour | Spot | 730 h | Source |
|---|---|---|---|---|---|
| — | OnlineGPUthis site · how the price is set | $0.08 | $0.05 | $58.40 | pricing page |
| 1 | Liumrecorded 15 Sep 2026 | $0.11 | not published | $80.30 | lium.io |
| 2 | GPU Outletrecorded 15 Sep 2026 | $0.13 | not published | $94.90 | computeprices.com |
| 3 | GPU.airecorded 15 Sep 2026 | $0.32 | not published | $233.60 | getdeploying.com |
| 4 | Vast.airecorded 15 Sep 2026 | $0.32 | $0.20 | $233.60 | getdeploying.com |
| 5 | Hexabyterecorded 15 Sep 2026 | $0.42 | not published | $306.60 | getdeploying.com |
| 6 | Seewebrecorded 15 Sep 2026 | $0.44 | not published | $320.47 | computeprices.com |
| 7 | Jarvis Labsrecorded 15 Sep 2026 | $0.44 | $0.29 | $321.20 | computeprices.com |
| 8 | Google Cloudrecorded 15 Sep 2026 | $0.45 | $0.38 | $328.50 | cloud.google.com |
| 9 | Leasewebrecorded 15 Sep 2026 | $0.45 | not published | $328.50 | getdeploying.com |
| 10 | RunPodrecorded 15 Sep 2026 | $0.49 | not published | $357.70 | runpod.io |
| 11 | AceCloudrecorded 15 Sep 2026 | $0.66 | not published | $483.26 | getdeploying.com |
| 12 | UpCloudrecorded 15 Sep 2026 | $0.68 | $0.68 | $496.40 | getdeploying.com |
| 13 | EmpirioLabs AIrecorded 15 Sep 2026 | $0.69 | not published | $503.70 | getdeploying.com |
| 14 | Koyebrecorded 15 Sep 2026 | $0.70 | not published | $511.00 | getdeploying.com |
| 15 | Modalrecorded 15 Sep 2026 | $0.80 | not published | $583.27 | computeprices.com |
Show the other 13 providers
| # | Provider | On-demandper GPU-hour | Spot | 730 h | Source |
|---|---|---|---|---|---|
| 16 | AWSrecorded 15 Sep 2026 | $0.80 | $0.58 | $584.00 | computeprices.com |
| 17 | Cerebriumrecorded 15 Sep 2026 | $0.80 | not published | $584.00 | getdeploying.com |
| 18 | Hugging Facerecorded 15 Sep 2026 | $0.80 | not published | $584.00 | getdeploying.com |
| 19 | AtmosComputerecorded 15 Sep 2026 | $0.80 | not published | $585.46 | computeprices.com |
| 20 | OVHcloudrecorded 15 Sep 2026 | $0.87 | not published | $635.10 | getdeploying.com |
| 21 | Scalewayrecorded 15 Sep 2026 | $0.91 | not published | $664.30 | shadeform.com |
| 22 | Shadeformrecorded 15 Sep 2026 | $0.95 | not published | $693.50 | computeprices.com |
| 23 | Omega Gradientrecorded 15 Sep 2026 | $1.00 | not published | $730.00 | computeprices.com |
| 24 | Elastxrecorded 15 Sep 2026 | $1.02 | not published | $744.60 | getdeploying.com |
| 25 | IO.NETrecorded 15 Sep 2026 | $1.04 | not published | $759.20 | computeprices.com |
| 26 | Sestercerecorded 15 Sep 2026 | $1.04 | not published | $759.20 | getdeploying.com |
| 27 | Runcraterecorded 15 Sep 2026 | $1.05 | not published | $766.50 | getdeploying.com |
| 28 | Hinoderecorded 15 Sep 2026 | $1.65 | not published | $1,204.50 | computeprices.com |
Recorded 15 September 2026 from published price pages (provider pricing pages and public price indexes), compute only: storage, egress and taxes excluded. Prices move — if a row is out of date, send us the link and it is corrected.
What people run on an L4.
Fine-tuning small models
LoRA on 7B–13B models, and full fine-tuning below 3B.
Image and video generation
Diffusion pipelines run entirely in memory at this capacity.
Anything that can checkpoint
On the spot tier this model is 38% below its own on-demand price, with a 2-minute notice before a reclaim.
NVIDIA L4, as published by NVIDIA.
Taken from the vendor datasheet. Where a figure is not published, the row is absent rather than estimated.
- Vendor
- NVIDIA
- Architecture
- Ada Lovelace (2023)
- Segment
- Datacenter
- Memory
- 24 GB GDDR6
- Memory bandwidth
- 300 GB/s
- FP16 / BF16 tensor
- 121 TFLOPS
- FP8 tensor
- 243 TFLOPS
- FP32
- 30 TFLOPS
- CUDA cores
- 7,424
- Board power
- 72 W
- Form factor
- PCIe single-slot
- Interconnect
- PCIe Gen4 x16 64 GB/s
- Multi-instance (MIG)
- Not available
- 8-bit float (FP8)
- Supported by the architecture
Source: www.nvidia.com · images.nvidia.com · verified against the vendor document
Software that runs on it
8 templatesPre-built environments, pulled on the node before you land on it. Or bring any OCI image from a public or private registry.
On the spot tier
<5% reclaimed · 30 d- 2 minutes of notice on the metadata endpoint, a webhook and the console
- The instance is stopped, not deleted — disk and IP stay attached
- Auto-relaunch on the next free L4, or switch the same disk to on-demand
- You save $0.72 a day per GPU against on-demand
If the L4 is not the right fit.
Next step up in memory: 32 GB instead of 24 GB
L40S 48 GB GDDR6 · Ada Lovelace $0.15/h spotSame Ada Lovelace generation
L40 48 GB GDDR6 · Ada Lovelace $0.09/h spotSame Ada Lovelace generation
V100 16GB 16 GB HBM2 · Volta $0.04/h spotFaster memory: 900 GB/s against 300 GB/s
Or put two of them side by side: L4 vs V100 · L4 vs L40S · L4 vs L40
L4 — the questions that come up
How much does an L4 cost per hour?
Spot is $0.05 per GPU-hour and on-demand is $0.08, both billed per minute — every started minute costs the hourly price divided by 60, so $0.0008 on spot. Reserved capacity is $0.06 per GPU-hour on a monthly commitment. Storage and egress are not included in that rate: persistent disks are $0.08 per GB-month and there are no egress fees.
How many GPUs can I put in one instance?
Node sizes are 1×, 2×, 4× — up to 4 L4 GPUs in a single virtual machine, with CPU, RAM and local NVMe scaled with the GPU count. The price per GPU is identical at every node size.
Can it run a 70B model?
Yes. At FP8, a 70B model needs about 70 GB for the weights, so it fits on 4× L4 ($0.20 per hour on spot) with roughly 20% of the memory left for activations and the KV cache.
Which regions have it?
Available now in US East (Virginia), US West (Oregon), EU Central (Frankfurt). Prices are identical in every region; pick the one closest to your data.
How often is a spot L4 reclaimed?
Over the trailing 30 days, <5% of spot instances on this model were reclaimed. You get a 2-minute notice on the metadata endpoint, through a webhook and in the console; the instance is then stopped, never deleted, and the persistent disk stays attached.
Anything else about this model? Ask an engineer — the same people run the nodes. See also all prices, how we compare with 41 other providers and how to pay in crypto.
An L4 for $0.05 an hour, running in under 60 seconds.
Prepaid in crypto, pay as you go: no card, no contract, no minimum commitment. Add credit, launch, stop whenever you want.
Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription. 96 L4 GPUs available right now.